text
null | inputs
dict | prediction
null | prediction_agent
null | annotation
list | annotation_agent
null | multi_label
bool 1
class | explanation
null | id
stringlengths 1
5
| metadata
null | status
stringclasses 2
values | event_timestamp
null | metrics
null |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
null | {
"abstract": " We construct an analog of the intrinsic normal cone of Behrend-Fantechi in\nthe equivariant motivic stable homotopy category over a base-scheme B and\nconstruct a fundament class in E-cohomology for any cohomology theory E in\nSH(B). For affine B, a perfect obstruction theory gives rise to a virtual\nfundamental class in a twisted Borel-Moore E-homology for arbitrary E. This\nincludes motivic cohomology (homotopy invariant) K-theory algebraic cobordism\nand the oriented Chow groups of Barge-Morel and Fasel. In the case of motivic\ncohomology, we recover the constructions of Behrend-Fantechi, with values in\nthe Chow group.\n",
"title": "The intrinsic stable normal cone"
} | null | null | null | null | true | null | 19901 | null | Default | null | null |
null | {
"abstract": " Most old globular clusters (GCs) in the Galaxy are observed to have internal\nchemical abundance spreads in light elements. We discuss a new GC formation\nscenario based on hierarchical star formation within fractal molecular clouds.\nIn the new scenario, a cluster of bound and unbound star clusters (`star\ncluster complex', SCC) that have a power-law cluster mass function with a slope\n(beta) of 2 is first formed from a massive gas clump developed in a dwarf\ngalaxy. Such cluster complexes and beta=2 are observed and expected from\nhierarchical star formation. The most massive star cluster (`main cluster'),\nwhich is the progenitor of a GC, can accrete gas ejected from asymptotic giant\nbranch (AGB) stars initially in the cluster and other low-mass clusters before\nthe clusters are tidally stripped or destroyed to become field stars in the\ndwarf. The SCC is initially embedded in a giant gas hole created by numerous\nsupernovae of the SCC so that cold gas outside the hole can be accreted onto\nthe main cluster later. New stars formed from the accreted gas have chemical\nabundances that are different from those of the original SCC. Using\nhydrodynamical simulations of GC formation based on this scenario, we show that\nthe main cluster with the initial mass as large as [2-5]x10^5 Msun can accrete\nmore than 10^5 Msun gas from AGB stars of the SCC. We suggest that merging of\nhierarchical star cluster complexes can play key roles in stellar halo\nformation around GCs and self-enrichment processes of GCs.\n",
"title": "Globular cluster formation with multiple stellar populations from hierarchical star cluster complexes"
} | null | null | null | null | true | null | 19902 | null | Default | null | null |
null | {
"abstract": " While domain adaptation has been actively researched in recent years, most\ntheoretical results and algorithms focus on the single-source-single-target\nadaptation setting. Naive application of such algorithms on multiple source\ndomain adaptation problem may lead to suboptimal solutions. As a step toward\nbridging the gap, we propose a new generalization bound for domain adaptation\nwhen there are multiple source domains with labeled instances and one target\ndomain with unlabeled instances. Compared with existing bounds, the new bound\ndoes not require expert knowledge about the target distribution, nor the\noptimal combination rule for multisource domains. Interestingly, our theory\nalso leads to an efficient learning strategy using adversarial neural networks:\nwe show how to interpret it as learning feature representations that are\ninvariant to the multiple domain shifts while still being discriminative for\nthe learning task. To this end, we propose two models, both of which we call\nmultisource domain adversarial networks (MDANs): the first model optimizes\ndirectly our bound, while the second model is a smoothed approximation of the\nfirst one, leading to a more data-efficient and task-adaptive model. The\noptimization tasks of both models are minimax saddle point problems that can be\noptimized by adversarial training. To demonstrate the effectiveness of MDANs,\nwe conduct extensive experiments showing superior adaptation performance on\nthree real-world datasets: sentiment analysis, digit classification, and\nvehicle counting.\n",
"title": "Multiple Source Domain Adaptation with Adversarial Training of Neural Networks"
} | null | null | [
"Computer Science",
"Statistics"
]
| null | true | null | 19903 | null | Validated | null | null |
null | {
"abstract": " We report the discovery of the minor planet 2013 SY$_{99}$, on an\nexceptionally distant, highly eccentric orbit. With a perihelion of 50.0 au,\n2013 SY$_{99}$'s orbit has a semi-major axis of $730 \\pm 40$ au, the largest\nknown for a high-perihelion trans-Neptunian object (TNO), well beyond those of\n(90377) Sedna and 2012 VP$_{113}$. Yet, with an aphelion of $1420 \\pm 90$ au,\n2013 SY$_{99}$'s orbit is interior to the region influenced by Galactic tides.\nSuch TNOs are not thought to be produced in the current known planetary\narchitecture of the Solar System, and they have informed the recent debate on\nthe existence of a distant giant planet. Photometry from the\nCanada-France-Hawaii Telescope, Gemini North and Subaru indicate 2013 SY$_{99}$\nis $\\sim 250$ km in diameter and moderately red in colour, similar to other\ndynamically excited TNOs. Our dynamical simulations show that Neptune's weak\ninfluence during 2013 SY$_{99}$'s perihelia encounters drives diffusion in its\nsemi-major axis of hundreds of astronomical units over 4 Gyr. The overall\nsymmetry of random walks in semi-major axis allow diffusion to populate 2013\nSY$_{99}$'s orbital parameter space from the 1000-2000 au inner fringe of the\nOort cloud. Diffusion affects other known TNOs on orbits with perihelia of 45\nto 49 au and semi-major axes beyond 250 au, providing a formation mechanism\nthat implies an extended population, gently cycling into and returning from the\ninner fringe of the Oort cloud.\n",
"title": "OSSOS: V. Diffusion in the orbit of a high-perihelion distant Solar System object"
} | null | null | null | null | true | null | 19904 | null | Default | null | null |
null | {
"abstract": " Using Lindblad dynamics we study quantum spin systems with dissipative\nboundary dynamics that generate a stationary nonequilibrium state with a\nnon-vanishing spin current that is locally conserved except at the boundaries.\nWe demonstrate that with suitably chosen boundary target states one can solve\nthe many-body Lindblad equation exactly in any dimension. As solution we obtain\npure states at any finite value of the dissipation strength and any system\nsize. They are characterized by a helical stationary magnetization profile and\na superdiffusive ballistic current of order one, independent of system size\neven when the quantum spin system is not integrable. These results are derived\nin explicit form for the one-dimensional spin-1/2 Heisenberg chain and its\nhigher-spin generalizations (which include for spin-1 the integrable\nZamolodchikov-Fateev model and the bi-quadratic Heisenberg chain). The\nextension of the results to higher dimensions is straightforward.\n",
"title": "Solution of the Lindblad equation for spin helix states"
} | null | null | null | null | true | null | 19905 | null | Default | null | null |
null | {
"abstract": " We study the spatially homogeneous time dependent solutions and their\nbifurcations of the Gray-Scott model. We find the global map of bifurcations by\na combination of rigorous verification of the existence of Takens Bogdanov and\na Bautin bifurcations, in the space of two parameters k and F. With the aid of\nnumerical continuation of local bifurcation curves we give a global description\nof all the possible bifurcations\n",
"title": "Global bifurcation map of the homogeneus states in the Gray-Scott model"
} | null | null | null | null | true | null | 19906 | null | Default | null | null |
null | {
"abstract": " We consider the minimization of composite objective functions composed of the\nexpectation of quadratic functions and an arbitrary convex function. We study\nthe stochastic dual averaging algorithm with a constant step-size, showing that\nit leads to a convergence rate of O(1/n) without strong convexity assumptions.\nThis thus extends earlier results on least-squares regression with the\nEuclidean geometry to (a) all convex regularizers and constraints, and (b) all\ngeome-tries represented by a Bregman divergence. This is achieved by a new\nproof technique that relates stochastic and deterministic recursions.\n",
"title": "Stochastic Composite Least-Squares Regression with convergence rate O(1/n)"
} | null | null | null | null | true | null | 19907 | null | Default | null | null |
null | {
"abstract": " Generative Adversarial Networks (GANs) are powerful models for learning\ncomplex distributions. Stable training of GANs has been addressed in many\nrecent works which explore different metrics between distributions. In this\npaper we introduce Fisher GAN which fits within the Integral Probability\nMetrics (IPM) framework for training GANs. Fisher GAN defines a critic with a\ndata dependent constraint on its second order moments. We show in this paper\nthat Fisher GAN allows for stable and time efficient training that does not\ncompromise the capacity of the critic, and does not need data independent\nconstraints such as weight clipping. We analyze our Fisher IPM theoretically\nand provide an algorithm based on Augmented Lagrangian for Fisher GAN. We\nvalidate our claims on both image sample generation and semi-supervised\nclassification using Fisher GAN.\n",
"title": "Fisher GAN"
} | null | null | null | null | true | null | 19908 | null | Default | null | null |
null | {
"abstract": " We propose a statistical model for natural language that begins by\nconsidering language as a monoid, then representing it in complex matrices with\na compatible translation invariant probability measure. We interpret the\nprobability measure as arising via the Born rule from a translation invariant\nmatrix product state.\n",
"title": "Language as a matrix product state"
} | null | null | null | null | true | null | 19909 | null | Default | null | null |
null | {
"abstract": " The low energy optical conductivity of conventional superconductors is\nusually well described by Mattis-Bardeen (MB) theory which predicts the onset\nof absorption above an energy corresponding to twice the superconducing (SC)\ngap parameter Delta. Recent experiments on strongly disordered superconductors\nhave challenged the application of the MB formulas due to the occurrence of\nadditional spectral weight at low energies below 2Delta. Here we identify three\ncrucial items which have to be included in the analysis of optical-conductivity\ndata for these systems: (a) the correct identification of the optical threshold\nin the Mattis-Bardeen theory, and its relation with the gap value extracted\nfrom the measured density of states, (b) the gauge-invariant evaluation of the\ncurrent-current response function, needed to account for the optical absorption\nby SC collective modes, and (c) the inclusion into the MB formula of the energy\ndependence of the density of states present already above Tc. By computing the\noptical conductvity in the disordered attractive Hubbard model we analyze the\nrelevance of all these items, and we provide a compelling scheme for the\nanalysis and interpretation of the optical data in real materials.\n",
"title": "On the application of Mattis-Bardeen theory in strongly disordered superconductors"
} | null | null | null | null | true | null | 19910 | null | Default | null | null |
null | {
"abstract": " This work demonstrates nanoscale magnetic imaging using bright circularly\npolarized high-harmonic radiation. We utilize the magneto-optical contrast of\nworm-like magnetic domains in a Co/Pd multilayer structure, obtaining\nquantitative amplitude and phase maps by lensless imaging. A\ndiffraction-limited spatial resolution of 49 nm is achieved with iterative\nphase reconstruction enhanced by a holographic mask. Harnessing the unique\ncoherence of high harmonics, this approach will facilitate quantitative,\nelement-specific and spatially-resolved studies of ultrafast magnetization\ndynamics, advancing both fundamental and applied aspects of nanoscale\nmagnetism.\n",
"title": "Nanoscale Magnetic Imaging using Circularly Polarized High-Harmonic Radiation"
} | null | null | null | null | true | null | 19911 | null | Default | null | null |
null | {
"abstract": " Superconducting detectors are now well-established tools for low-light\noptics, and in particular quantum optics, boasting high-efficiency, fast\nresponse and low noise. Similarly, lithium niobate is an important platform for\nintegrated optics given its high second-order nonlinearity, used for high-speed\nelectro-optic modulation and polarization conversion, as well as frequency\nconversion and sources of quantum light. Combining these technologies addresses\nthe requirements for a single platform capable of generating, manipulating and\nmeasuring quantum light in many degrees of freedom, in a compact and\npotentially scalable manner. We will report on progress integrating tungsten\ntransition-edge sensors (TESs) and amorphous tungsten silicide superconducting\nnanowire single-photon detectors (SNSPDs) on titanium in-diffused lithium\nniobate waveguides. The travelling-wave design couples the evanescent field\nfrom the waveguides into the superconducting absorber. We will report on\nsimulations and measurements of the absorption, which we can characterize at\nroom temperature prior to cooling down the devices. Independently, we show how\nthe detectors respond to flood illumination, normally incident on the devices,\ndemonstrating their functionality.\n",
"title": "Towards integrated superconducting detectors on lithium niobate waveguides"
} | null | null | null | null | true | null | 19912 | null | Default | null | null |
null | {
"abstract": " We prove that the Hilbert scheme of 11 points on a smooth threefold is\nirreducible. In the course of the proof, we present several known and new\ntechniques for producing curves on the Hilbert scheme.\n",
"title": "The Hilbert scheme of 11 points in A^3 is irreducible"
} | null | null | null | null | true | null | 19913 | null | Default | null | null |
null | {
"abstract": " We propose a novel type of tunable Yagi-Uda nanoantenna composed of\nmetal-dielectric (Ag-Ge) core-shell nanoparticles. We show that, due to the\ncombination of two types of resonances in each nanoparticle, such hybrid\nYagi-Uda nanoantenna can operate in two different regimes. Besides the\nconventional nonresonant operation regime at low frequencies, characterized by\nhighly directive emission in the forward direction, there is another one at\nhigher frequencies caused by hybrid magneto-electric response of the core-shell\nnanoparticles. This regime is based on the excitation of the van Hove\nsingularity, and emission in this regime is accompanied by high values of\ndirectivity and Purcell factor within the same narrow frequency range. Our\nanalysis reveals the possibility of flexible dynamical tuning of the hybrid\nnanoantenna emission pattern via electron-hole plasma excitation by 100\nfemtosecond pump pulse with relatively low peak intensities $\\sim$200\nMW/cm$^2$.\n",
"title": "Dynamically reconfigurable metal-semiconductor Yagi-Uda nanoantenna"
} | null | null | [
"Physics"
]
| null | true | null | 19914 | null | Validated | null | null |
null | {
"abstract": " Modified structures of SAPO-34 were prepared using polyethylene glycol as the\nmesopores generating agent. The synthesized catalysts were applied in\nmethanol-to-olefins (MTO) process. All modified synthesized catalysts were\ncharacterized via XRD, XRF, FESEM, FTIR, N2 adsorption-desorption techniques,\nand temperature-programmed NH3 desorption and they were compared with\nconventional microporous SAPO-34. Introduction of non-ionic PEG capping agent\naffected the degree of homogeneity and integrity of the synthesis media and\nthus reduced the number of nuclei and order of coordination structures\nresulting in larger and less crystalline particles compared with the\nconventional sample. During the calcination process, decomposition of absorbed\nPEG moieties among the piled up SAPO patches formed a great portion of tuned\nmesopores into the microporous matrix. These tailored mesopores were served as\nauxiliary diffusion pathways in MTO reaction. The effects of molecular weight\nof PEG and PEG/Al molar ratio on the properties of the synthesized materials\nwere investigated in order to optimize their MTO reaction performance. It was\nrevealed that both of these two parameters can significantly change the\nstructural composition and physicochemical properties of resultant products.\nUsing PEG with MW of 6000 has led to the formation of RHO and CHA structural\nframeworks i.e. DNL-6 and SAPO-34, simultaneously, while addition of PEG with\nMW of 4000 resulted the formation of pure SAPO-34 phase. Altering the PEG/Al\nmolar ratio in the precursor significantly influenced the porosity and acidity\nof the synthesized silicoaluminophosphate products. SAPO-34 impregnated with\nPEG molecular weight of 4000 and PEG/Al molar ratio of 0.0125 showed superior\ncatalytic stability in MTO reaction because of the tuned bi-modal porosity and\ntailored acidity pattern.\n",
"title": "Synthesis and In Situ Modification of Hierarchical SAPO-34 by PEG with Different Molecular Weights; Application in MTO Process"
} | null | null | null | null | true | null | 19915 | null | Default | null | null |
null | {
"abstract": " Support Vector Machines (SVMs) with various kernels have played dominant role\nin machine learning for many years, finding numerous applications. Although\nthey have many attractive features interpretation of their solutions is quite\ndifficult, the use of a single kernel type may not be appropriate in all areas\nof the input space, convergence problems for some kernels are not uncommon, the\nstandard quadratic programming solution has $O(m^3)$ time and $O(m^2)$ space\ncomplexity for $m$ training patterns. Kernel methods work because they\nimplicitly provide new, useful features. Such features, derived from various\nkernels and other vector transformations, may be used directly in any machine\nlearning algorithm, facilitating multiresolution, heterogeneous models of data.\nTherefore Support Feature Machines (SFM) based on linear models in the extended\nfeature spaces, enabling control over selection of support features, give at\nleast as good results as any kernel-based SVMs, removing all problems related\nto interpretation, scaling and convergence. This is demonstrated for a number\nof benchmark datasets analyzed with linear discrimination, SVM, decision trees\nand nearest neighbor methods.\n",
"title": "Support Feature Machines"
} | null | null | null | null | true | null | 19916 | null | Default | null | null |
null | {
"abstract": " Voltage deviations occur frequently in power systems. If the violation at\nsome buses falls outside the prescribed range, it will be necessary to correct\nthe problem by controlling reactive power resources. In this paper, an optimal\nalgorithm is proposed to solve this problem by identifying the voltage buses,\nthat will have a maximum effect on the affected buses, and setting their new\nset-points. This algorithm is based on the Eigen-Value Decomposition of the\nfast decoupled load flow Jacobian matrix. Different Case studies including IEEE\n9, 14, 30 and 57 bus systems have been used to verify the method.\n",
"title": "Voltage Control Using Eigen Value Decomposition of Fast Decoupled Load Flow Jacobian"
} | null | null | [
"Computer Science"
]
| null | true | null | 19917 | null | Validated | null | null |
null | {
"abstract": " Convolutional Neural Networks (CNNs) are a popular deep learning architecture\nwidely applied in different domains, in particular in classifying over images,\nfor which the concept of convolution with a filter comes naturally.\nUnfortunately, the requirement of a distance (or, at least, of a neighbourhood\nfunction) in the input feature space has so far prevented its direct use on\ndata types such as omics data. However, a number of omics data are metrizable,\ni.e., they can be endowed with a metric structure, enabling to adopt a\nconvolutional based deep learning framework, e.g., for prediction. We propose a\ngeneralized solution for CNNs on omics data, implemented through a dedicated\nKeras layer. In particular, for metagenomics data, a metric can be derived from\nthe patristic distance on the phylogenetic tree. For transcriptomics data, we\ncombine Gene Ontology semantic similarity and gene co-expression to define a\ndistance; the function is defined through a multilayer network where 3 layers\nare defined by the GO mutual semantic similarity while the fourth one by gene\nco-expression. As a general tool, feature distance on omics data is enabled by\nOmicsConv, a novel Keras layer, obtaining OmicsCNN, a dedicated deep learning\nframework. Here we demonstrate OmicsCNN on gut microbiota sequencing data, for\nInflammatory Bowel Disease (IBD) 16S data, first on synthetic data and then a\nmetagenomics collection of gut microbiota of 222 IBD patients.\n",
"title": "Convolutional neural networks for structured omics: OmicsCNN and the OmicsConv layer"
} | null | null | null | null | true | null | 19918 | null | Default | null | null |
null | {
"abstract": " Traditional neural network approaches for traffic flow forecasting are\nusually single task learning (STL) models, which do not take advantage of the\ninformation provided by related tasks. In contrast to STL, multitask learning\n(MTL) has the potential to improve generalization by transferring information\nin training signals of extra tasks. In this paper, MTL based neural networks\nare used for traffic flow forecasting. For neural network MTL, a\nbackpropagation (BP) network is constructed by incorporating traffic flows at\nseveral contiguous time instants into an output layer. Nodes in the output\nlayer can be seen as outputs of different but closely related STL tasks.\nComprehensive experiments on urban vehicular traffic flow data and comparisons\nwith STL show that MTL in BP neural networks is a promising and effective\napproach for traffic flow forecasting.\n",
"title": "Neural Network Multitask Learning for Traffic Flow Forecasting"
} | null | null | null | null | true | null | 19919 | null | Default | null | null |
null | {
"abstract": " The Tolman paradox is well known as a base for demonstrating the causality\nviolation by faster-than-light signals within special relativity. It is\nconstructed using a two-way exchange of faster-than-light signals between two\ninertial observers who are in a relative motion receding one from another.\nRecently a one-way superluminal signalling arrangement was suggested as a\npossible construction of a causal paradox. In this note we show that this\nsuggestion is not correct, and no causality principle violation can occur in\nany one-way signalling by the use of faster-than-light particles and signals.\n",
"title": "A note on a new paradox in superluminal signalling"
} | null | null | [
"Physics"
]
| null | true | null | 19920 | null | Validated | null | null |
null | {
"abstract": " For analysing text algorithms, for computing superstrings, or for testing\nrandom number generators, one needs to compute all overlaps between any pairs\nof words in a given set. The positions of overlaps of a word onto itself, or of\ntwo words, are needed to compute the absence probability of a word in a random\ntext, or the numbers of common words shared by two random texts. In all these\ncontexts, one needs to compute or to query overlaps between pairs of words in a\ngiven set. For this sake, we designed COvI, a compressed overlap index that\nsupports multiple queries on overlaps: like computing the correlation of two\nwords, or listing pairs of words whose longest overlap is maximal among all\npossible pairs. COvI stores overlaps in a hierarchical and non-redundant\nmanner. We propose an implementation that can handle datasets of millions of\nwords and still answer queries efficiently. Comparison with a baseline solution\n- called FullAC - relying on the Aho-Corasick automaton shows that COvI\nprovides significant advantages. For similar construction times, COvI requires\nhalf the memory FullAC, and still solves complex queries much faster.\n",
"title": "The Compressed Overlap Index"
} | null | null | null | null | true | null | 19921 | null | Default | null | null |
null | {
"abstract": " Spin-relaxation is conventionally discussed using two different approaches\nfor materials with and without inversion symmetry. The former is known as the\nElliott-Yafet (EY) theory and for the latter the D'yakonov-Perel' (DP) theory\napplies, respectively. We discuss herein a simple and intuitive approach to\ndemonstrate that the two seemingly disparate mechanisms are closely related. A\ncompelling analogy between the respective Hamiltonian is presented and that the\nusual derivation of spin-relaxation times, in the respective frameworks of the\ntwo theories, can be performed. The result also allows to obtain the less\ncanonical spin-relaxation regimes; the generalization of the EY when the\nmaterial has a large quasiparticle broadening and the DP mechanism in ultrapure\nsemiconductors. The method also allows a practical and intuitive numerical\nimplementation of the spin-relaxation calculation, which is demonstrated for\nMgB$_2$ that has anomalous spin-relaxation properties.\n",
"title": "An intuitive approach to the unified theory of spin-relaxation"
} | null | null | null | null | true | null | 19922 | null | Default | null | null |
null | {
"abstract": " In the information overloaded web, personalized recommender systems are\nessential tools to help users find most relevant information. The most\nheavily-used recommendation frameworks assume user interactions that are\ncharacterized by a single relation. However, for many tasks, such as\nrecommendation in social networks, user-item interactions must be modeled as a\ncomplex network of multiple relations, not only a single relation. Recently\nresearch on multi-relational factorization and hybrid recommender models has\nshown that using extended meta-paths to capture additional information about\nboth users and items in the network can enhance the accuracy of recommendations\nin such networks. Most of this work is focused on unweighted heterogeneous\nnetworks, and to apply these techniques, weighted relations must be simplified\ninto binary ones. However, information associated with weighted edges, such as\nuser ratings, which may be crucial for recommendation, are lost in such\nbinarization. In this paper, we explore a random walk sampling method in which\nthe frequency of edge sampling is a function of edge weight, and apply this\ngenerate extended meta-paths in weighted heterogeneous networks. With this\nsampling technique, we demonstrate improved performance on multiple data sets\nboth in terms of recommendation accuracy and model generation efficiency.\n",
"title": "Weighted Random Walk Sampling for Multi-Relational Recommendation"
} | null | null | null | null | true | null | 19923 | null | Default | null | null |
null | {
"abstract": " Scholarly communication has the scope to transcend the limitations of the\nphysical world through social media extended coverage and shortened information\npaths. Accordingly, publishers have created profiles for their journals in\nTwitter to promote their publications and to initiate discussions with public.\nThis paper investigates the Twitter presence of humanities and social sciences\n(HSS) journal titles obtained from mainstream citation indices, by analysing\nthe interaction and communication patterns. This study utilizes webometric data\ncollection, descriptive analysis, and social network analysis. Findings\nindicate that the presence of HSS journals in Twitter across disciplines is not\nyet substantial. Sharing of general websites appears to be the key activity\nperformed by HSS journals in Twitter. Among them, web content from news portals\nand magazines are highly disseminated. Sharing of research articles and\nretweeting was not majorly observed. Inter-journal communication is apparent\nwithin the same citation index, but it is very minimal with journals from the\nother index. However, there seems to be an effort to broaden communication\nbeyond the research community, reaching out to connect with the public.\n",
"title": "Understanding the Twitter Usage of Humanities and Social Sciences Academic Journals"
} | null | null | null | null | true | null | 19924 | null | Default | null | null |
null | {
"abstract": " Machine learning algorithms, when applied to sensitive data, pose a distinct\nthreat to privacy. A growing body of prior work demonstrates that models\nproduced by these algorithms may leak specific private information in the\ntraining data to an attacker, either through the models' structure or their\nobservable behavior. However, the underlying cause of this privacy risk is not\nwell understood beyond a handful of anecdotal accounts that suggest overfitting\nand influence might play a role.\nThis paper examines the effect that overfitting and influence have on the\nability of an attacker to learn information about the training data from\nmachine learning models, either through training set membership inference or\nattribute inference attacks. Using both formal and empirical analyses, we\nillustrate a clear relationship between these factors and the privacy risk that\narises in several popular machine learning algorithms. We find that overfitting\nis sufficient to allow an attacker to perform membership inference and, when\nthe target attribute meets certain conditions about its influence, attribute\ninference attacks. Interestingly, our formal analysis also shows that\noverfitting is not necessary for these attacks and begins to shed light on what\nother factors may be in play. Finally, we explore the connection between\nmembership inference and attribute inference, showing that there are deep\nconnections between the two that lead to effective new attacks.\n",
"title": "Privacy Risk in Machine Learning: Analyzing the Connection to Overfitting"
} | null | null | null | null | true | null | 19925 | null | Default | null | null |
null | {
"abstract": " Recently, the fabrication of CdSe nanoplatelets became an important research\ntopic. Nanoplatelets are often described as having a similar electronic\nstructure as 2D dimensional quantum wells and are promoted as colloidal quantum\nwells with monolayer precision width. In this paper, we show, that\nnanoplatelets are not ideal quantum wells, but cover depending on the size: the\nstrong confinement regime, an intermediate regime and a Coulomb dominated\nregime. Thus, nanoplatelets are an ideal platform to study the physics in these\nregimes. Therefore, the exciton states of the nanoplatelets are numerically\ncalculated by solving the full four dimensional Schrödinger equation. We\ncompare the results with approximate solutions from semiconductor quantum well\nand quantum dot theory. The paper can also act as review of these concepts for\nthe colloidal nanoparticle community.\n",
"title": "Nanoplatelets as material system between strong confinement and weak confinement"
} | null | null | null | null | true | null | 19926 | null | Default | null | null |
null | {
"abstract": " Advances in GNC, particularly from miniaturized control electronics,\nreaction-wheels and attitude determination sensors make it possible to design\nsurface probes and small robots to perform surface exploration and science on\nlow-gravity environments. These robots would use their reaction wheels to roll,\nhop and tumble over rugged surfaces. These robots could provide 'Google\nStreetview' quality images of off-world surfaces and perform some unique\nscience using penetrometers. These systems can be powered by high-efficiency\nfuel cells that operate at 60-65 % and utilize hydrogen and oxygen electrolyzed\nfrom water. However, one of the major challenges that prevent these probes and\nrobots from performing long duration surface exploration and science is thermal\ndesign and control. In the inner solar system, during the day time, there is\noften enough solar-insolation to keep these robots warm and power these\ndevices, but during eclipse the temperatures falls well below storage\ntemperature. We have developed a thermal control system that utilizes chemicals\nto store and dispense heat when needed. The system takes waste products, such\nas water from these robots and transfers them to a thermochemical storage\nsystem. These thermochemical storage systems when mixed with water (a waste\nproduct from a PEM fuel cell) releases heat. Under eclipse, the heat from the\nthermochemical storage system is released to keep the probe warm enough to\nsurvive. In sunlight, solar photovoltaics are used to electrolyze the water and\nreheat the thermochemical storage system to release the water. Our research has\nshowed thermochemical storage systems are a feasible solution for use on\nsurface probes and robots for applications on the Moon, Mars and asteroids.\n",
"title": "Combined Thermal Control and GNC: An Enabling Technology for CubeSat Surface Probes and Small Robots"
} | null | null | null | null | true | null | 19927 | null | Default | null | null |
null | {
"abstract": " Unsupervised learning is of growing interest because it unlocks the potential\nheld in vast amounts of unlabelled data to learn useful representations for\ninference. Autoencoders, a form of generative model, may be trained by learning\nto reconstruct unlabelled input data from a latent representation space. More\nrobust representations may be produced by an autoencoder if it learns to\nrecover clean input samples from corrupted ones. Representations may be further\nimproved by introducing regularisation during training to shape the\ndistribution of the encoded data in latent space. We suggest denoising\nadversarial autoencoders, which combine denoising and regularisation, shaping\nthe distribution of latent space using adversarial training. We introduce a\nnovel analysis that shows how denoising may be incorporated into the training\nand sampling of adversarial autoencoders. Experiments are performed to assess\nthe contributions that denoising makes to the learning of representations for\nclassification and sample synthesis. Our results suggest that autoencoders\ntrained using a denoising criterion achieve higher classification performance,\nand can synthesise samples that are more consistent with the input data than\nthose trained without a corruption process.\n",
"title": "Denoising Adversarial Autoencoders"
} | null | null | null | null | true | null | 19928 | null | Default | null | null |
null | {
"abstract": " We define monodromy maps for tropical Dolbeault cohomology of algebraic\nvarieties over non-Archimedean fields. We propose a conjecture of Hodge\nisomorphisms via monodromy maps, and provide some evidence.\n",
"title": "Monodromy map for tropical Dolbeault cohomology"
} | null | null | null | null | true | null | 19929 | null | Default | null | null |
null | {
"abstract": " In this paper, we study the problems in the discrete Fourier transform (DFT)\ntest included in NIST SP 800-22 released by the National Institute of Standards\nand Technology (NIST), which is a collection of tests for evaluating both\nphysical and pseudo-random number generators for cryptographic applications.\nThe most crucial problem in the DFT test is that its reference distribution of\nthe test statistic is not derived mathematically but rather numerically\nestimated, the DFT test for randomness is based on a pseudo-random number\ngenerator (PRNG). Therefore, the present DFT test should not be used unless the\nreference distribution is mathematically derived. Here, we prove that a power\nspectrum, which is a component of the test statistic, follows a chi-squared\ndistribution with 2 degrees of freedom. Based on this fact, we propose a test\nwhose reference distribution of the test statistic is mathematically derived.\nFurthermore, the results of testing non-random sequences and several PRNGs\nshowed that the proposed test is more reliable and definitely more sensitive\nthan the present DFT test.\n",
"title": "Randomness Evaluation with the Discrete Fourier Transform Test Based on Exact Analysis of the Reference Distribution"
} | null | null | null | null | true | null | 19930 | null | Default | null | null |
null | {
"abstract": " Word embeddings generated by neural network methods such as word2vec (W2V)\nare well known to exhibit seemingly linear behaviour, e.g. the embeddings of\nanalogy \"woman is to queen as man is to king\" approximately describe a\nparallelogram. This property is particularly intriguing since the embeddings\nare not trained to achieve it. Several explanations have been proposed, but\neach introduces assumptions that do not hold in practice. We derive a\nprobabilistically grounded definition of paraphrasing and show it can be\nre-interpreted as word transformation, a mathematical description of \"$w_x$ is\nto $w_y$\". From these concepts we prove existence of the linear relationship\nbetween W2V-type embeddings that underlies the analogical phenomenon, and\nidentify explicit error terms in the relationship.\n",
"title": "Analogies Explained: Towards Understanding Word Embeddings"
} | null | null | null | null | true | null | 19931 | null | Default | null | null |
null | {
"abstract": " We will say that an Abelian group $\\Gamma$ of order $n$ has the\n$m$-\\emph{zero-sum-partition property} ($m$-\\textit{ZSP-property}) if $m$\ndivides $n$, $m\\geq 2$ and there is a partition of $\\Gamma$ into pairwise\ndisjoint subsets $A_1, A_2,\\ldots , A_t$, such that $|A_i| = m$ and $\\sum_{a\\in\nA_i}a = g_0$ for $1 \\leq i \\leq t$, where $g_0$ is the identity element of\n$\\Gamma$.\nIn this paper we study the $m$-ZSP property of $\\Gamma$. We show that\n$\\Gamma$ has $m$-ZSP if and only if $|\\Gamma|$ is odd or $m\\geq 3$ and $\\Gamma$\nhas more than one involution. We will apply the results to the study of group\ndistance magic graphs as well as to generalized Kotzig arrays.\n",
"title": "Zero sum partition into sets of the same order and its applications"
} | null | null | null | null | true | null | 19932 | null | Default | null | null |
null | {
"abstract": " Cosmologies including strongly Coupled (SC) Dark Energy (DE) and Warm dark\nmatter (SCDEW) are based on a conformally invariant (CI) attractor solution\nmodifying the early radiative expansion. Then, aside of radiation, a kinetic\nfield $\\Phi$ and a DM component account for a stationary fraction, $\\sim 1\\,\n\\%$, of the total energy. Most SCDEW predictions are hardly distinguishable\nfrom LCDM, while SCDEW alleviates quite a few LCDM conceptual problems, as well\nas its difficulties to meet data below the average galaxy scale. The CI\nexpansion begins at the inflation end, when $\\Phi$ (future DE) possibly plays a\nrole in reheating, and ends at the Higgs' scale. Afterwards, a number of viable\noptions is open, allowing for the transition from the CI expansion to the\npresent Universe. In this paper: (i) We show how the attractor is recovered\nwhen the spin degrees of freedom decreases. (ii) We perform a detailed\ncomparison of CMB anisotropy and polarization spectra for SCDEW and LCDM,\nincluding tensor components, finding negligible discrepancies. (iii) Linear\nspectra exhibit a greater parameter dependence at large $k$'s, but are still\nconsistent with data for suitable parameter choices. (iv) We also compare\nprevious simulation results with fresh data on galaxy concentration. Finally,\n(v) we outline numerical difficulties at high $k$. This motivates a second\nrelated paper, where such problems are treated in a quantitative way.\n",
"title": "Strongly Coupled Dark Energy with Warm dark matter vs. LCDM"
} | null | null | null | null | true | null | 19933 | null | Default | null | null |
null | {
"abstract": " Proof schemata are a variant of LK-proofs able to simulate various induction\nschemes in first-order logic by adding so called proof links to the standard\nfirst-order LK-calculus. Proof links allow proofs to reference proofs thus\ngiving proof schemata a recursive structure. Unfortunately, applying reductive\ncut- elimination is non-trivial in the presence of proof links. Borrowing the\nconcept of lazy instantiation from functional programming, we evaluate proof\nlinks locally allowing reductive cut-elimination to proceed past them. Though,\nthis method cannot be used to obtain cut-free proof schemata, we nonetheless\nobtain important results concerning the schematic CERES method, that is a\nmethod of cut-elimination for proof schemata based on resolution. In \"Towards a\nclausal analysis of cut-elimination\", it was shown that reductive\ncut-elimination transforms a given LK-proof in such a way that a subsumption\nrelation holds between the pre- and post-transformation characteristic clause\nsets, i.e. the clause set representing the cut-structure of an LK-proof. Let\nCL(A') be the characteristic clause set of a normal form A' of an LK-proof A\nthat is reached by performing reductive cut-elimination on A without atomic cut\nelimination. Then CL(A') is subsumed by all characteristic clause sets\nextractable from any application of reductive cut-elimination to A. Such a\nnormal form is referred to as an ACNF top and plays an essential role in\nmethods of cut-elimination by resolution. These results can be extended to\nproof schemata through our \"lazy instantiation\" of proof links, and provides an\nessential step toward a complete cut-elimination method for proof schemata.\n",
"title": "Clausal Analysis of First-order Proof Schemata"
} | null | null | null | null | true | null | 19934 | null | Default | null | null |
null | {
"abstract": " This paper examines the task of detecting intensity of emotion from text. We\ncreate the first datasets of tweets annotated for anger, fear, joy, and sadness\nintensities. We use a technique called best--worst scaling (BWS) that improves\nannotation consistency and obtains reliable fine-grained scores. We show that\nemotion-word hashtags often impact emotion intensity, usually conveying a more\nintense emotion. Finally, we create a benchmark regression system and conduct\nexperiments to determine: which features are useful for detecting emotion\nintensity, and, the extent to which two emotions are similar in terms of how\nthey manifest in language.\n",
"title": "Emotion Intensities in Tweets"
} | null | null | null | null | true | null | 19935 | null | Default | null | null |
null | {
"abstract": " Using simulations with a whole-atmosphere chemistry-climate model nudged by\nmeteorological analyses, global satellite observations of nitrogen oxide (NO)\nand water vapour by the Sub-Millimetre Radiometer instrument (SMR), of\ntemperature by the Microwave Limb Sounder (MLS), as well as local radar\nobservations, this study examines the recent major stratospheric sudden warming\naccompanied by an elevated stratopause event (ESE) that occurred in January\n2013. We examine dynamical processes during the ESE, including the role of\nplanetary wave, gravity wave and tidal forcing on the initiation of the descent\nin the mesosphere-lower thermosphere (MLT) and its continuation throughout the\nmesosphere and stratosphere, as well as the impact of model eddy diffusion. We\nanalyse the transport of NO and find the model underestimates the large descent\nof NO compared to SMR observations. We demonstrate that the discrepancy arises\nabruptly in the MLT region at a time when the resolved wave forcing and the\nplanetary wave activity increase, just before the elevated stratopause reforms.\nThe discrepancy persists despite doubling the model eddy diffusion. While the\nsimulations reproduce an enhancement of the semi-diurnal tide following the\nonset of the 2013 SSW, corroborating new meteor radar observations at high\nnorthern latitudes over Trondheim (63.4$^{\\circ}$N), the modelled tidal\ncontribution to the forcing of the mean meridional circulation and to the\ndescent is a small portion of the resolved wave forcing, and lags it by about\nten days.\n",
"title": "Modelling the descent of nitric oxide during the elevated stratopause event of January 2013"
} | null | null | null | null | true | null | 19936 | null | Default | null | null |
null | {
"abstract": " A fundamental question in biology is how organisms integrate sensory and\nsocial evidence to make decisions. However, few models describe how both these\nstreams of information can be combined to optimize choices. Here we develop a\nnormative model for collective decision making in a network of agents\nperforming a two-alternative forced choice task. We assume that rational\n(Bayesian) agents in this network make private measurements, and observe the\ndecisions of their neighbors until they accumulate sufficient evidence to make\nan irreversible choice. As each agent communicates its decision to those\nobserving it, the flow of social information is described by a directed graph.\nThe decision-making process in this setting is intuitive, but can be complex.\nWe describe when and how the absence of a decision of a neighboring agent\ncommunicates social information, and how an agent must marginalize over all\nunobserved decisions. We also show how decision thresholds and network\nconnectivity affect group evidence accumulation, and describe the dynamics of\ndecision making in social cliques. Our model provides a bridge between the\nabstractions used in the economics literature and the evidence accumulator\nmodels used widely in neuroscience and psychology.\n",
"title": "Optimal Evidence Accumulation on Social Networks"
} | null | null | null | null | true | null | 19937 | null | Default | null | null |
null | {
"abstract": " Conventional seismic techniques for detecting the subsurface geologic\nfeatures are challenged by limited data coverage, computational inefficiency,\nand subjective human factors. We developed a novel data-driven geological\nfeature detection approach based on pre-stack seismic measurements. Our\ndetection method employs an efficient and accurate machine-learning detection\napproach to extract useful subsurface geologic features automatically.\nSpecifically, our method is based on kernel ridge regression model. The\nconventional kernel ridge regression can be computationally prohibited because\nof the large volume of seismic measurements. We employ a data reduction\ntechnique in combination with the conventional kernel ridge regression method\nto improve the computational efficiency and reduce memory usage. In particular,\nwe utilize a randomized numerical linear algebra technique, named Nyström\nmethod, to effectively reduce the dimensionality of the feature space without\ncompromising the information content required for accurate detection. We\nprovide thorough computational cost analysis to show efficiency of our new\ngeological feature detection methods. We further validate the performance of\nour new subsurface geologic feature detection method using synthetic surface\nseismic data for 2D acoustic and elastic velocity models. Our numerical\nexamples demonstrate that our new detection method significantly improves the\ncomputational efficiency while maintaining comparable accuracy. Interestingly,\nwe show that our method yields a speed-up ratio on the order of $\\sim10^2$ to\n$\\sim 10^3$ in a multi-core computational environment.\n",
"title": "Efficient Data-Driven Geologic Feature Detection from Pre-stack Seismic Measurements using Randomized Machine-Learning Algorithm"
} | null | null | null | null | true | null | 19938 | null | Default | null | null |
null | {
"abstract": " We present a novel approach to achieve adaptable band structures and\nnon-reciprocal wave propagation by exploring and exploiting the concept of\nmetastable modular metastructures. Through studying the dynamics of wave\npropagation in a chain composed of finite metastable modules, we provide\nexperimental and analysis results on non-reciprocal wave propagation and unveil\nthe underlying mechanisms in accomplishing such unidirectional energy\ntransmission. Utilizing the property adaptation feature afforded via\ntransitioning amongst metastable states, we uncovered an unprecedented bandgap\nreconfiguration characteristic, which enables the adaptivity of wave\npropagation within the metastructure. Overall, this investigation elucidates\nthe rich dynamics attainable by periodicity, nonlinearity, asymmetry, and\nmetastability, and creates a new class of adaptive structural and material\nsystems capable of realizing tunable bandgaps and non-reciprocal wave\ntransmissions.\n",
"title": "Metastable Modular Metastructures for On-Demand Reconfiguration of Band Structures and Non-Reciprocal Wave Propagation"
} | null | null | null | null | true | null | 19939 | null | Default | null | null |
null | {
"abstract": " From medicines to materials, small organic molecules are indispensable for\nhuman well-being. To plan their syntheses, chemists employ a problem solving\ntechnique called retrosynthesis. In retrosynthesis, target molecules are\nrecursively transformed into increasingly simpler precursor compounds until a\nset of readily available starting materials is obtained. Computer-aided\nretrosynthesis would be a highly valuable tool, however, past approaches were\nslow and provided results of unsatisfactory quality. Here, we employ Monte\nCarlo Tree Search (MCTS) to efficiently discover retrosynthetic routes. MCTS\nwas combined with an expansion policy network that guides the search, and an\n\"in-scope\" filter network to pre-select the most promising retrosynthetic\nsteps. These deep neural networks were trained on 12 million reactions, which\nrepresents essentially all reactions ever published in organic chemistry. Our\nsystem solves almost twice as many molecules and is 30 times faster in\ncomparison to the traditional search method based on extracted rules and\nhand-coded heuristics. Finally after a 60 year history of computer-aided\nsynthesis planning, chemists can no longer distinguish between routes generated\nby a computer system and real routes taken from the scientific literature. We\nanticipate that our method will accelerate drug and materials discovery by\nassisting chemists to plan better syntheses faster, and by enabling fully\nautomated robot synthesis.\n",
"title": "Learning to Plan Chemical Syntheses"
} | null | null | null | null | true | null | 19940 | null | Default | null | null |
null | {
"abstract": " Let $\\hat{L}$ be the projective completion of an ample line bundle $L$ over\n$D$, a smooth projective manifold. Hwang-Singer \\cite{HwangS} have constructed\ncomplete CSCK metric on $\\hat{L}\\backslash D$. When the corresponding \\kahler\nform is in the cohomology class of a rational divisor $A$ and when $L$ has\nnegative CSCK metric on $D$, we show that the Kodaira embedding induced by\northonormal basis of the Bergman space of $kA$ is almost balanced. As a\ncorollary, $(\\hat{L},D,cA,0)$ is K-semistable.\n",
"title": "Projective embedding of pairs and logarithmic K-stability"
} | null | null | null | null | true | null | 19941 | null | Default | null | null |
null | {
"abstract": " To ensure that a robot is able to accomplish an extensive range of tasks, it\nis necessary to achieve a flexible combination of multiple behaviors. This is\nbecause the design of task motions suited to each situation would become\nincreasingly difficult as the number of situations and the types of tasks\nperformed by them increase. To handle the switching and combination of multiple\nbehaviors, we propose a method to design dynamical systems based on point\nattractors that accept (i) \"instruction signals\" for instruction-driven\nswitching. We incorporate the (ii) \"instruction phase\" to form a point\nattractor and divide the target task into multiple subtasks. By forming an\ninstruction phase that consists of point attractors, the model embeds a subtask\nin the form of trajectory dynamics that can be manipulated using sensory and\ninstruction signals. Our model comprises two deep neural networks: a\nconvolutional autoencoder and a multiple time-scale recurrent neural network.\nIn this study, we apply the proposed method to manipulate soft materials. To\nevaluate our model, we design a cloth-folding task that consists of four\nsubtasks and three patterns of instruction signals, which indicate the\ndirection of motion. The results depict that the robot can perform the required\ntask by combining subtasks based on sensory and instruction signals. And, our\nmodel determined the relations among these signals using its internal dynamics.\n",
"title": "Motion Switching with Sensory and Instruction Signals by designing Dynamical Systems using Deep Neural Network"
} | null | null | null | null | true | null | 19942 | null | Default | null | null |
null | {
"abstract": " Over the last decade, tremendous strides have been achieved in our\nunderstanding of magnetism in main sequence hot stars. In particular, the\nstatistical occurrence of their surface magnetism has been established (~10%)\nand the field origin is now understood to be fossil. However, fundamental\nquestions remain: how do these fossil fields evolve during the post-main\nsequence phases, and how do they influence the evolution of hot stars from the\nmain sequence to their ultimate demise? Filling the void of known magnetic\nevolved hot (OBA) stars, studying the evolution of their fossil magnetic fields\nalong stellar evolution, and understanding the impact of these fields on the\nangular momentum, rotation, mass loss, and evolution of the star itself, is\ncrucial to answering these questions, with far reaching consequences, in\nparticular for the properties of the precursors of supernovae explosions and\nstellar remnants. In the framework of the BRITE spectropolarimetric survey and\nLIFE project, we have discovered the first few magnetic hot supergiants. Their\nlongitudinal surface magnetic field is very weak but their configuration\nresembles those of main sequence hot stars. We present these first\nobservational results and propose to interpret them at first order in the\ncontext of magnetic flux conservation as the radius of the star expands with\nevolution. We then also consider the possible impact of stellar structure\nchanges along evolution.\n",
"title": "The evolution of magnetic fields in hot stars"
} | null | null | null | null | true | null | 19943 | null | Default | null | null |
null | {
"abstract": " We develop polynomial-time heuristic methods to solve unimodular quadratic\nprograms (UQPs) approximately, which are known to be NP-hard. In the UQP\nframework, we maximize a quadratic function of a vector of complex variables\nwith unit modulus. Several problems in active sensing and wireless\ncommunication applications boil down to UQP. With this motivation, we present\nthree new heuristic methods with polynomial-time complexity to solve the UQP\napproximately. The first method is called dominant-eigenvector-matching; here\nthe solution is picked that matches the complex arguments of the dominant\neigenvector of the Hermitian matrix in the UQP formulation. We also provide a\nperformance guarantee for this method. The second method, a greedy strategy, is\nshown to provide a performance guarantee of (1-1/e) with respect to the optimal\nobjective value given that the objective function possesses a property called\nstring submodularity. The third heuristic method is called row-swap greedy\nstrategy, which is an extension to the greedy strategy and utilizes certain\nproperties of the UQP to provide a better performance than the greedy strategy\nat the expense of an increase in computational complexity. We present numerical\nresults to demonstrate the performance of these heuristic methods, and also\ncompare the performance of these methods against a standard heuristic method\ncalled semidefinite relaxation.\n",
"title": "Polynomial-Time Methods to Solve Unimodular Quadratic Programs With Performance Guarantees"
} | null | null | null | null | true | null | 19944 | null | Default | null | null |
null | {
"abstract": " It is well known that the F test is severly affected by heteroskedasticity in\nunbalanced analysis of covariance (ANCOVA) models. Currently available remedies\nfor such a scenario are either based on heteroskedasticity-consistent\ncovariance matrix estimation (HCCME) or bootstrap techniques. However, the\nHCCME approach tends to be liberal in small samples. Therefore, we propose a\ncombination of HCCME and a wild bootstrap technique. We prove the theoretical\nvalidity of our approach and investigate its performance in an extensive\nsimulation study in comparison to existing procedures. The results indicate\nthat our proposed test remedies all problems of the ANCOVA F test and its\nheteroskedasticityconsistent alternatives. Our test only requires very general\nconditions, thus being applicable in a broad range of real-life settings.\n",
"title": "Can the Wild Bootstrap be Tamed into a General Analysis of Covariance Model?"
} | null | null | [
"Statistics"
]
| null | true | null | 19945 | null | Validated | null | null |
null | {
"abstract": " Feature representations from pre-trained deep neural networks have been known\nto exhibit excellent generalization and utility across a variety of related\ntasks. Fine-tuning is by far the simplest and most widely used approach that\nseeks to exploit and adapt these feature representations to novel tasks with\nlimited data. Despite the effectiveness of fine-tuning, itis often sub-optimal\nand requires very careful optimization to prevent severe over-fitting to small\ndatasets. The problem of sub-optimality and over-fitting, is due in part to the\nlarge number of parameters used in a typical deep convolutional neural network.\nTo address these problems, we propose a simple yet effective regularization\nmethod for fine-tuning pre-trained deep networks for the task of k-shot\nlearning. To prevent overfitting, our key strategy is to cluster the model\nparameters while ensuring intra-cluster similarity and inter-cluster diversity\nof the parameters, effectively regularizing the dimensionality of the parameter\nsearch space. In particular, we identify groups of neurons within each layer of\na deep network that shares similar activation patterns. When the network is to\nbe fine-tuned for a classification task using only k examples, we propagate a\nsingle gradient to all of the neuron parameters that belong to the same group.\nThe grouping of neurons is non-trivial as neuron activations depend on the\ndistribution of the input data. To efficiently search for optimal groupings\nconditioned on the input data, we propose a reinforcement learning search\nstrategy using recurrent networks to learn the optimal group assignments for\neach network layer. Experimental results show that our method can be easily\napplied to several popular convolutional neural networks and improve upon other\nstate-of-the-art fine-tuning based k-shot learning strategies by more than10%\n",
"title": "Efficient K-Shot Learning with Regularized Deep Networks"
} | null | null | null | null | true | null | 19946 | null | Default | null | null |
null | {
"abstract": " Let p be a prime, and k be a field of characteristic p. We investigate the\nalgebra structure and the structure of the cohomology ring for the connected\nHopf algebras of dimension p^3, which appear in the classification obtained in\n[V.C. Nguyen, L.-H. Wang and X.-T. Wang, Classification of connected Hopf\nalgebras of dimension p^3, J. Algebra 424 (2015), 473-505]. The list consists\nof 23 algebras together with two infinite families. We identify the Morita type\nof the algebra, and in almost all cases this is sufficient to clarify the\nstructure of the cohomology ring.\n",
"title": "The cohomology ring of some Hopf algebras"
} | null | null | [
"Mathematics"
]
| null | true | null | 19947 | null | Validated | null | null |
null | {
"abstract": " Recently, a general expression for Eckart-frame Hamilton operators has been\nobtained by the gateway Hamiltonian method ({\\it J. Chem. Phys.} {\\bf 142},\n174107 (2015); {\\it ibid.} {\\bf 143}, 064104 (2015)). The kinetic energy\noperator in this general Hamiltonian is nearly identical with that of the\nEckart-Watson operator even when curvilinear vibrational coordinates are\nemployed. Its different realizations correspond to different methods of\ncalculating Eckart displacements. There are at least two different methods for\ncalculating such displacements: rotation and projection. In this communication\nthe application of Eckart Hamiltonian operators constructed by rotation and\nprojection, respectively, is numerically demonstrated in calculating\nvibrational energy levels. The numerical examples confirm that there is no need\nfor rotation to construct an Eckart ro-vibrational Hamiltonian. The application\nof the gateway method is advantageous even when rotation is used, since it\nobviates the need for differentiation of the matrix rotating into the Eckart\nframe. Simple geometrical arguments explain that there are infinitely many\ndifferent methods for calculating Eckart displacements. The geometrical picture\nalso suggests that a unique Eckart displacement vector may be defined as the\nshortest (mass-weighted) Eckart displacement vector among Eckart displacement\nvectors corresponding to configurations related by rotation. Its length, as\nshown analytically and demonstrated by way of numerical examples, is equal to\nor less than that of the Eckart displacement vector one can obtain by rotation\nto the Eckart frame.\n",
"title": "Eckart ro-vibrational Hamiltonians via the gateway Hamilton operator: theory and practice"
} | null | null | null | null | true | null | 19948 | null | Default | null | null |
null | {
"abstract": " This paper studies an optimal trading problem that incorporates the trader's\nmarket view on the terminal asset price distribution and uninformative noise\nembedded in the asset price dynamics. We model the underlying asset price\nevolution by an exponential randomized Brownian bridge (rBb) and consider\nvarious prior distributions for the random endpoint. We solve for the optimal\nstrategies to sell a stock, call, or put, and analyze the associated delayed\nliquidation premia. We solve for the optimal trading strategies numerically and\ncompare them across different prior beliefs. Among our results, we find that\ndisconnected continuation/exercise regions arise when the trader prescribe a\ntwo-point discrete distribution and double exponential distribution.\n",
"title": "Optimal Timing to Trade Along a Randomized Brownian Bridge"
} | null | null | null | null | true | null | 19949 | null | Default | null | null |
null | {
"abstract": " We explore the relationship between features in the Planck 2015 temperature\nand polarization data, shifts in the cosmological parameters, and features from\ninflation. Residuals in the temperature data at low multipole $\\ell$, which are\nresponsible for the high $H_0\\approx 70$ km s$^{-1}$Mpc$^{-1}$ and low\n$\\sigma_8\\Omega_m^{1/2}$ values from $\\ell<1000$ in power-law $\\Lambda$CDM\nmodels, are better fit to inflationary features with a $1.9\\sigma$ preference\nfor running of the running of the tilt or a stronger $99\\%$ CL local\nsignificance preference for a sharp drop in power around $k=0.004$ Mpc$^{-1}$\nin generalized slow roll and a lower $H_0\\approx 67$ km s$^{-1}$Mpc$^{-1}$. The\nsame in-phase acoustic residuals at $\\ell>1000$ that drive the global $H_0$\nconstraints and appear as a lensing anomaly also favor running parameters which\nallow even lower $H_0$, but not once lensing reconstruction is considered.\nPolarization spectra are intrinsically highly sensitive to these parameter\nshifts, and even more so in the Planck 2015 TE data due to an outlier at $\\ell\n\\approx 165$, which disfavors the best fit $H_0$ $\\Lambda$CDM solution by more\nthan $2\\sigma$, and high $H_0$ value at almost $3\\sigma$. Current polarization\ndata also slightly enhance the significance of a sharp suppression of\nlarge-scale power but leave room for large improvements in the future with\ncosmic variance limited $E$-mode measurements.\n",
"title": "Inflationary Features and Shifts in Cosmological Parameters from Planck 2015 Data"
} | null | null | null | null | true | null | 19950 | null | Default | null | null |
null | {
"abstract": " Estimating individualized treatment rules is a central task for personalized\nmedicine. [zhao2012estimating] and [zhang2012robust] proposed outcome weighted\nlearning to estimate individualized treatment rules directly through maximizing\nthe expected outcome without modeling the response directly. In this paper, we\nextend the outcome weighted learning to right censored survival data without\nrequiring either an inverse probability of censoring weighting or a\nsemiparametric modeling of the censoring and failure times as done in\n[zhao2015doubly]. To accomplish this, we take advantage of the tree based\napproach proposed in [zhu2012recursively] to nonparametrically impute the\nsurvival time in two different ways. The first approach replaces the reward of\neach individual by the expected survival time, while in the second approach\nonly the censored observations are imputed by their conditional expected\nfailure times. We establish consistency and convergence rates for both\nestimators. In simulation studies, our estimators demonstrate improved\nperformance compared to existing methods. We also illustrate the proposed\nmethod on a phase III clinical trial of non-small cell lung cancer.\n",
"title": "Tree based weighted learning for estimating individualized treatment rules with censored data"
} | null | null | null | null | true | null | 19951 | null | Default | null | null |
null | {
"abstract": " Fine-tuning in physics and cosmology is often used as evidence that a theory\nis incomplete. For example, the parameters of the standard model of particle\nphysics are \"unnaturally\" small (in various technical senses), which has driven\nmuch of the search for physics beyond the standard model. Of particular\ninterest is the fine-tuning of the universe for life, which suggests that our\nuniverse's ability to create physical life forms is improbable and in need of\nexplanation, perhaps by a multiverse. This claim has been challenged on the\ngrounds that the relevant probability measure cannot be justified because it\ncannot be normalized, and so small probabilities cannot be inferred. We show\nhow fine-tuning can be formulated within the context of Bayesian theory testing\n(or \\emph{model selection}) in the physical sciences. The normalizability\nproblem is seen to be a general problem for testing any theory with free\nparameters, and not a unique problem for fine-tuning. Physical theories in fact\navoid such problems in one of two ways. Dimensional parameters are bounded by\nthe Planck scale, avoiding troublesome infinities, and we are not compelled to\nassume that dimensionless parameters are distributed uniformly, which avoids\nnon-normalizability.\n",
"title": "Fine-Tuning in the Context of Bayesian Theory Testing"
} | null | null | null | null | true | null | 19952 | null | Default | null | null |
null | {
"abstract": " Experiments using nuclei to probe new physics beyond the Standard Model, such\nas neutrinoless $\\beta\\beta$ decay searches testing whether neutrinos are their\nown antiparticle, and direct detection experiments aiming to identify the\nnature of dark matter, require accurate nuclear physics input for optimizing\ntheir discovery potential and for a correct interpretation of their results.\nThis demands a detailed knowledge of the nuclear structure relevant for these\nprocesses. For instance, neutrinoless $\\beta\\beta$ decay nuclear matrix\nelements are very sensitive to the nuclear correlations in the initial and\nfinal nuclei, and the spin-dependent nuclear structure factors of dark matter\nscattering depend on the subtle distribution of the nuclear spin among all\nnucleons. In addition, nucleons are composite and strongly interacting, which\nimplies that many-nucleon processes are necessary for a correct description of\nnuclei and their interactions. It is thus crucial that theoretical studies and\nexperimental analyses consider $\\beta$ decays and dark matter interactions with\na coupling to two nucleons, called two-nucleon currents.\n",
"title": "Nuclear physics insights for new-physics searches using nuclei: Neutrinoless $ββ$ decay and dark matter direct detection"
} | null | null | null | null | true | null | 19953 | null | Default | null | null |
null | {
"abstract": " Machine Learning (ML) models are applied in a variety of tasks such as\nnetwork intrusion detection or Malware classification. Yet, these models are\nvulnerable to a class of malicious inputs known as adversarial examples. These\nare slightly perturbed inputs that are classified incorrectly by the ML model.\nThe mitigation of these adversarial inputs remains an open problem. As a step\ntowards understanding adversarial examples, we show that they are not drawn\nfrom the same distribution than the original data, and can thus be detected\nusing statistical tests. Using thus knowledge, we introduce a complimentary\napproach to identify specific inputs that are adversarial. Specifically, we\naugment our ML model with an additional output, in which the model is trained\nto classify all adversarial inputs. We evaluate our approach on multiple\nadversarial example crafting methods (including the fast gradient sign and\nsaliency map methods) with several datasets. The statistical test flags sample\nsets containing adversarial inputs confidently at sample sizes between 10 and\n100 data points. Furthermore, our augmented model either detects adversarial\nexamples as outliers with high accuracy (> 80%) or increases the adversary's\ncost - the perturbation added - by more than 150%. In this way, we show that\nstatistical properties of adversarial examples are essential to their\ndetection.\n",
"title": "On the (Statistical) Detection of Adversarial Examples"
} | null | null | null | null | true | null | 19954 | null | Default | null | null |
null | {
"abstract": " We introduce the fraud de-anonymization problem, that goes beyond fraud\ndetection, to unmask the human masterminds responsible for posting search rank\nfraud in online systems. We collect and study search rank fraud data from\nUpwork, and survey the capabilities and behaviors of 58 search rank fraudsters\nrecruited from 6 crowdsourcing sites. We propose Dolos, a fraud\nde-anonymization system that leverages traits and behaviors extracted from\nthese studies, to attribute detected fraud to crowdsourcing site fraudsters,\nthus to real identities and bank accounts. We introduce MCDense, a min-cut\ndense component detection algorithm to uncover groups of user accounts\ncontrolled by different fraudsters, and leverage stylometry and deep learning\nto attribute them to crowdsourcing site profiles. Dolos correctly identified\nthe owners of 95% of fraudster-controlled communities, and uncovered fraudsters\nwho promoted as many as 97.5% of fraud apps we collected from Google Play. When\nevaluated on 13,087 apps (820,760 reviews), which we monitored over more than 6\nmonths, Dolos identified 1,056 apps with suspicious reviewer groups. We report\northogonal evidence of their fraud, including fraud duplicates and fraud\nre-posts.\n",
"title": "Search Rank Fraud De-Anonymization in Online Systems"
} | null | null | null | null | true | null | 19955 | null | Default | null | null |
null | {
"abstract": " The types of instability in the interacting binary stars are reviewed. The\nproject \"Inter-Longitude Astronomy\" is a series of smaller projects on concrete\nstars or groups of stars. It has no special funds, and is supported from\nresources and grants of participating organizations, when informal working\ngroups are created. Totally we studied 1900+ variable stars of different types.\nThe characteristic timescale is from seconds to decades and (extrapolating)\neven more. The monitoring of the first star of our sample AM Her was initiated\nby Prof. V.P. Tsesevich (1907-1983). Since more than 358 ADS papers were\npublished. Some highlights of our photometric and photo-polarimetric monitoring\nand mathematical modelling of interacting binary stars of different types are\npresented: classical, asynchronous, intermediate polars and magnetic dwarf\nnovae (DO Dra) with 25 timescales corresponding to different physical\nmechanisms and their combinations (part \"Polar\"); negative and positive\nsuperhumpers in nova-like and many dwarf novae stars (\"Superhumper\"); eclipsing\n\"non-magnetic\" cataclysmic variables; symbiotic systems (\"Symbiosis\");\nsuper-soft sources (SSS, QR And); spotted (and not spotted) eclipsing variables\nwith (and without) evidence for a current mass transfer (\"Eclipser\") with a\nspecial emphasis on systems with a direct impact of the stream into the gainer\nstar's atmosphere, or V361 Lyr-type stars. Other parts of the ILA project are\n\"Stellar Bell\" (interesting pulsating variables of different types and periods\n- M, SR, RV Tau, RR Lyr, Delta Sct) and \"Novice\"(=\"New Variable\") discoveries\nand classification with a subsequent monitoring for searching and studying\npossible multiple components of variability. Special mathematical methods have\nbeen developed to create a set of complementary software for statistically\noptimal modelling of variable stars of different types.\n",
"title": "Instabilities in Interacting Binary Stars"
} | null | null | null | null | true | null | 19956 | null | Default | null | null |
null | {
"abstract": " The interaction blockade phenomenon isolates the motion of a single quantum\nparticle within a multi-particle system, in particular for coherent\noscillations in and out of a region affected by the blockade mechanism. For\nidentical quantum particles with Bose statistics, the presence of the other\nparticles is still felt by a bosonic stimulation factor $\\sqrt{N}$ that speeds\nup the coherent oscillations, where $N$ is the number of bosons. Here we\npropose an experiment to observe this enhancement factor with a small number of\nbosonic atoms. The proposed protocol realises an asymmetric double well\npotential with multiple optical tweezer laser beams. The ability to adjust bias\nindependently of the coherent coupling between the wells allows the potential\nto be loaded with different particle numbers while maintaining the resonance\ncondition needed for coherent oscillations. Numerical simulations with up to\nthree bosons in a realistic potential generated by three optical tweezers\npredict that the relevant avoided level crossing can be probed and the expected\nbosonic enhancement factor observed.\n",
"title": "Interaction blockade for bosons in an asymmetric double well"
} | null | null | null | null | true | null | 19957 | null | Default | null | null |
null | {
"abstract": " SPIDER is a balloon-borne instrument designed to map the polarization of the\nmillimeter-wave sky at large angular scales. SPIDER targets the B-mode\nsignature of primordial gravitational waves in the cosmic microwave background\n(CMB), with a focus on mapping a large sky area with high fidelity at multiple\nfrequencies. SPIDER's first longduration balloon (LDB) flight in January 2015\ndeployed a total of 2400 antenna-coupled Transition Edge Sensors (TESs) at 90\nGHz and 150 GHz. In this work we review the design and in-flight performance of\nthe SPIDER instrument, with a particular focus on the measured performance of\nthe detectors and instrument in a space-like loading and radiation environment.\nSPIDER's second flight in December 2018 will incorporate payload upgrades and\nnew receivers to map the sky at 285 GHz, providing valuable information for\ncleaning polarized dust emission from CMB maps.\n",
"title": "SPIDER: CMB polarimetry from the edge of space"
} | null | null | null | null | true | null | 19958 | null | Default | null | null |
null | {
"abstract": " Wind shear measured by Doppler tracking of the Huygens probe is evaluated,\nand found to be within the range anticipated by pre-flight assessments (namely\nless than two times the Brunt-Vaisala frequency). The strongest large-scale\nshear encountered was ~5 m/s/km, a level associated with 'Light' turbulence in\nterrestrial aviation. Near-surface winds (below 4km) have small-scale\nfluctuations of ~0.2 m/s , indicated both by probe tilt and Doppler tracking,\nand the characteristics of the fluctuation, of interest for future missions to\nTitan, can be reproduced with a simple autoregressive (AR(1)) model. The\nturbulent dissipation rate at an altitude of ~500m is found to be 16 cm2/sec3,\nwhich may be a useful benchmark for atmospheric circulation models.\n",
"title": "Wind Shear and Turbulence on Titan : Huygens Analysis"
} | null | null | null | null | true | null | 19959 | null | Default | null | null |
null | {
"abstract": " Relativistic protocols have been proposed to overcome some impossibility\nresults in classical and quantum cryptography. In such a setting, one takes the\nlocation of honest players into account, and uses the fact that information\ncannot travel faster than the speed of light to limit the abilities of\ndishonest agents. For example, various relativistic bit commitment protocols\nhave been proposed. Although it has been shown that bit commitment is\nsufficient to construct oblivious transfer and thus multiparty computation,\ncomposing specific relativistic protocols in this way is known to be insecure.\nA composable framework is required to perform such a modular security analysis\nof construction schemes, but no known frameworks can handle models of\ncomputation in Minkowski space.\nBy instantiating the systems model from the Abstract Cryptography framework\nwith Causal Boxes, we obtain such a composable framework, in which messages are\nassigned a location in Minkowski space (or superpositions thereof). This allows\nus to analyse relativistic protocols and to derive novel possibility and\nimpossibility results. We show that (1) coin flipping can be constructed from\nthe primitive channel with delay, (2) biased coin flipping, bit commitment and\nchannel with delay are all impossible without further assumptions, and (3) it\nis impossible to improve a channel with delay. Note that the impossibility\nresults also hold in the computational and bounded storage settings. This\nimplies in particular non-composability of all proposed relativistic bit\ncommitment protocols, of bit commitment in the bounded storage model, and of\nbiased coin flipping.\n",
"title": "Composable security in relativistic quantum cryptography"
} | null | null | [
"Computer Science"
]
| null | true | null | 19960 | null | Validated | null | null |
null | {
"abstract": " We prove that all eigenstates of many-body localized symmetry protected\ntopological systems with time reversal symmetry have four-fold degenerate\nentanglement spectra in the thermodynamic limit. To that end, we employ unitary\nquantum circuits where the number of sites the gates act on grows linearly with\nthe system size. We find that the corresponding matrix product operator\nrepresentation has similar local symmetries as matrix product ground states of\nsymmetry protected topological phases. Those local symmetries give rise to a\n$\\mathbb{Z}_2$ topological index, which is robust against arbitrary\nperturbations so long as they do not break time reversal symmetry or drive the\nsystem out of the fully many-body localized phase.\n",
"title": "Tensor networks demonstrate the robustness of localization and symmetry protected topological phases"
} | null | null | null | null | true | null | 19961 | null | Default | null | null |
null | {
"abstract": " We study the formation of massive black holes in the first star clusters. We\nfirst locate star-forming gas clouds in proto-galactic haloes of $\\gtrsim\n\\!10^7\\,{\\rm M}_{\\odot}$ in cosmological hydrodynamics simulations and use them\nto generate the initial conditions for star clusters with masses of $\\sim\n\\!10^5\\,{\\rm M}_{\\odot}$. We then perform a series of direct-tree hybrid\n$N$-body simulations to follow runaway stellar collisions in the dense star\nclusters. In all the cluster models except one, runaway collisions occur within\na few million years, and the mass of the central, most massive star reaches\n$\\sim \\!400-1900\\,{\\rm M}_{\\odot}$. Such very massive stars collapse to leave\nintermediate-mass black holes (IMBHs). The diversity of the final masses may be\nattributed to the differences in a few basic properties of the host haloes such\nas mass, central gas velocity dispersion, and mean gas density of the central\ncore. Finally, we derive the IMBH mass to cluster mass ratios, and compare them\nwith the observed black hole to bulge mass ratios in the present-day Universe.\n",
"title": "Formation of Intermediate-Mass Black Holes through Runaway Collisions in the First Star Clusters"
} | null | null | null | null | true | null | 19962 | null | Default | null | null |
null | {
"abstract": " We study an optimization-based approach to con- struct a mean-reverting\nportfolio of assets. Our objectives are threefold: (1) design a portfolio that\nis well-represented by an Ornstein-Uhlenbeck process with parameters estimated\nby maximum likelihood, (2) select portfolios with desirable characteristics of\nhigh mean reversion and low variance, and (3) select a parsimonious portfolio,\ni.e. find a small subset of a larger universe of assets that can be used for\nlong and short positions. We present the full problem formulation, a\nspecialized algorithm that exploits partial minimization, and numerical\nexamples using both simulated and empirical price data.\n",
"title": "Mean Reverting Portfolios via Penalized OU-Likelihood Estimation"
} | null | null | null | null | true | null | 19963 | null | Default | null | null |
null | {
"abstract": " We show how third-party web trackers can deanonymize users of\ncryptocurrencies. We present two distinct but complementary attacks. On most\nshopping websites, third party trackers receive information about user\npurchases for purposes of advertising and analytics. We show that, if the user\npays using a cryptocurrency, trackers typically possess enough information\nabout the purchase to uniquely identify the transaction on the blockchain, link\nit to the user's cookie, and further to the user's real identity. Our second\nattack shows that if the tracker is able to link two purchases of the same user\nto the blockchain in this manner, it can identify the user's entire cluster of\naddresses and transactions on the blockchain, even if the user employs\nblockchain anonymity techniques such as CoinJoin. The attacks are passive and\nhence can be retroactively applied to past purchases. We discuss several\nmitigations, but none are perfect.\n",
"title": "When the cookie meets the blockchain: Privacy risks of web payments via cryptocurrencies"
} | null | null | null | null | true | null | 19964 | null | Default | null | null |
null | {
"abstract": " Virtualization technologies have evolved along with the development of\ncomputational environments since virtualization offered needed features at that\ntime such as isolation, accountability, resource allocation, resource fair\nsharing and so on. Novel processor technologies bring to commodity computers\nthe possibility to emulate diverse environments where a wide range of\ncomputational scenarios can be run. Along with processors evolution, system\ndevelopers have created different virtualization mechanisms where each new\ndevelopment enhanced the performance of previous virtualized environments.\nRecently, operating system-based virtualization technologies captured the\nattention of communities abroad (from industry to academy and research) because\ntheir important improvements on performance area.\nIn this paper, the features of three container-based operating systems\nvirtualization tools (LXC, Docker and Singularity) are presented. LXC, Docker,\nSingularity and bare metal are put under test through a customized single node\nHPL-Benchmark and a MPI-based application for the multi node testbed. Also the\ndisk I/O performance, Memory (RAM) performance, Network bandwidth and GPU\nperformance are tested for the COS technologies vs bare metal. Preliminary\nresults and conclusions around them are presented and discussed.\n",
"title": "Performance Evaluation of Container-based Virtualization for High Performance Computing Environments"
} | null | null | null | null | true | null | 19965 | null | Default | null | null |
null | {
"abstract": " We give an integral formula for the total $Q^\\prime$-curvature of a\nthree-dimensional CR manifold with positive CR Yamabe constant and nonnegative\nPaneitz operator. Our derivation includes a relationship between the Green's\nfunctions of the CR Laplacian and the $P^\\prime$-operator.\n",
"title": "An integral formula for the $Q$-prime curvature in 3-dimensional CR geometry"
} | null | null | null | null | true | null | 19966 | null | Default | null | null |
null | {
"abstract": " The Internet of Things (IoT) is continuously growing to connect billions of\nsmart devices anywhere and anytime in an Internet-like structure, which enables\na variety of applications, services and interactions between human and objects.\nIn the future, the smart devices are supposed to be able to autonomously\ndiscover a target device with desired features and generate a set of entirely\nnew services and applications that are not supervised or even imagined by human\nbeings. The pervasiveness of smart devices, as well as the heterogeneity of\ntheir design and functionalities, raise a major concern: How can a smart device\nefficiently discover a desired target device? In this paper, we propose a\nSocial-Aware and Distributed (SAND) scheme that achieves a fast, scalable and\nefficient device discovery in the IoT. The proposed SAND scheme adopts a novel\ndevice ranking criteria that measures the device's degree, social relationship\ndiversity, clustering coefficient and betweenness. Based on the device ranking\ncriteria, the discovery request can be guided to travel through critical\ndevices that stand at the major intersections of the network, and thus quickly\nreach the desired target device by contacting only a limited number of\nintermediate devices. With the help of such an intelligent device discovery as\nSAND, the IoT devices, as well as other computing facilities, software and data\non the Internet, can autonomously establish new social connections with each\nother as human being do. They can formulate self-organized computing groups to\nperform required computing tasks, facilitate a fusion of a variety of computing\nservice, network service and data to generate novel applications and services,\nevolve from the individual aritificial intelligence to the collaborative\nintelligence, and eventually enable the birth of a robot society.\n",
"title": "Intelligent Device Discovery in the Internet of Things - Enabling the Robot Society"
} | null | null | null | null | true | null | 19967 | null | Default | null | null |
null | {
"abstract": " Modelling information cascades over online social networks is important in\nfields from marketing to civil unrest prediction, however the underlying\nnetwork structure strongly affects the probability and nature of such cascades.\nEven with simple cascade dynamics the probability of large cascades are almost\nentirely dictated by network properties, with well-known networks such as\nErdos-Renyi and Barabasi-Albert producing wildly different cascades from the\nsame model. Indeed, the notion of 'superspreaders' has arisen to describe\nhighly influential nodes promoting global cascades in a social network. Here we\nuse a simple model of global cascades to show that the presence of locality in\nthe network increases the probability of a global cascade due to the increased\nvulnerability of connecting nodes. Rather than 'super-spreaders', we find that\nthe presence of these highly connected 'super-blockers' in heavy-tailed\nnetworks in fact reduces the probability of global cascades, while promoting\ninformation spread when targeted as the initial spreader.\n",
"title": "Super-blockers and the effect of network structure on information cascades"
} | null | null | null | null | true | null | 19968 | null | Default | null | null |
null | {
"abstract": " Living cells exhibit multi-mode transport that switches between an active,\nself-propelled motion and a seemingly passive, random motion. Cellular\ndecision-making over transport mode switching is a stochastic process that\ndepends on the dynamics of the intracellular chemical network regulating the\ncell migration process. Here, we propose a theory and an exactly solvable model\nof multi-mode active matter. Our exact model study shows that the reversible\ntransition between a passive mode and an active mode is the origin of the\nanomalous, super-Gaussian transport dynamics, which has been observed in\nvarious experiments for multi-mode active matter. We also present the\ngeneralization of our model to encompass complex multi-mode matter with\narbitrary internal state chemical dynamics and internal state dependent\ntransport dynamics.\n",
"title": "Super-Gaussian, super-diffusive transport of multi-mode active matter"
} | null | null | null | null | true | null | 19969 | null | Default | null | null |
null | {
"abstract": " We define various height functions for motives over number fields. We compare\nthese height functions with classical height functions on algebraic varieties,\nand also with analogous height functions for variations of Hodge structures on\ncurves over C. These comparisons provide new questions on motives over number\nfields.\n",
"title": "Height functions for motives"
} | null | null | null | null | true | null | 19970 | null | Default | null | null |
null | {
"abstract": " This paper studies the dimension effect of the linear discriminant analysis\n(LDA) and the regularized linear discriminant analysis (RLDA) classifiers for\nlarge dimensional data where the observation dimension $p$ is of the same order\nas the sample size $n$. More specifically, built on properties of the Wishart\ndistribution and recent results in random matrix theory, we derive explicit\nexpressions for the asymptotic misclassification errors of LDA and RLDA\nrespectively, from which we gain insights of how dimension affects the\nperformance of classification and in what sense. Motivated by these results, we\npropose adjusted classifiers by correcting the bias brought by the unequal\nsample sizes. The bias-corrected LDA and RLDA classifiers are shown to have\nsmaller misclassification rates than LDA and RLDA respectively. Several\ninteresting examples are discussed in detail and the theoretical results on\ndimension effect are illustrated via extensive simulation studies.\n",
"title": "On the dimension effect of regularized linear discriminant analysis"
} | null | null | null | null | true | null | 19971 | null | Default | null | null |
null | {
"abstract": " Fundamental questions on the nature of matter and energy have found answers\nthanks to the use of particle accelerators. Societal applications, such as\ncancer treatment or cancer imaging, illustrate the impact of accelerators in\nour current life. Today, accelerators use metallic cavities that sustain\nelectricfields with values limited to about 100 MV/m. Because of their ability\nto support extreme accelerating gradients, the plasma medium has recently been\nproposed for future cavity-like accelerating structures. This contribution\nhighlights the tremendous evolution of plasma accelerators driven by either\nlaser or particle beams that allow the production of high quality particle\nbeams with a degree of tunability and a set of parameters that make them very\npertinent for many applications.\n",
"title": "Plasma Wake Accelerators: Introduction and Historical Overview"
} | null | null | null | null | true | null | 19972 | null | Default | null | null |
null | {
"abstract": " The main aim of this paper is to prove $R$-triviality for simple, simply\nconnected algebraic groups with Tits index $E_{8,2}^{78}$ or $E_{7,1}^{78}$,\ndefined over a field $k$ of arbitrary characteristic. Let $G$ be such a group.\nWe prove that there exists a quadratic extension $K$ of $k$ such that $G$ is\n$R$-trivial over $K$, i.e., for any extension $F$ of $K$, $G(F)/R=\\{1\\}$, where\n$G(F)/R$ denotes the group of $R$-equivalence classes in $G(F)$, in the sense\nof Manin (see \\cite{M}). As a consequence, it follows that the variety $G$ is\nretract $K$-rational and that the Kneser-Tits conjecture holds for these groups\nover $K$. Moreover, $G(L)$ is projectively simple as an abstract group for any\nfield extension $L$ of $K$. In their monograph (\\cite{TW}) J. Tits and Richard\nWeiss conjectured that for an Albert division algebra $A$ over a field $k$, its\nstructure group $Str(A)$ is generated by scalar homotheties and its\n$U$-operators. This is known to be equivalent to the Kneser-Tits conjecture for\ngroups with Tits index $E_{8,2}^{78}$. We settle this conjecture for Albert\ndivision algebras which are first constructions, in affirmative. These results\nare obtained as corollaries to the main result, which shows that if $A$ is an\nAlbert division algebra which is a first construction and $\\Gamma$ its\nstructure group, i.e., the algebraic group of the norm similarities of $A$,\nthen $\\Gamma(F)/R=\\{1\\}$ for any field extension $F$ of $k$, i.e., $\\Gamma$ is\n$R$-trivial.\n",
"title": "$R$-triviality of some exceptional groups"
} | null | null | null | null | true | null | 19973 | null | Default | null | null |
null | {
"abstract": " We present near-infrared interferometry of the carbon-rich asymptotic giant\nbranch (AGB) star R Sculptoris.\nThe visibility data indicate a broadly circular resolved stellar disk with a\ncomplex substructure. The observed AMBER squared visibility values show drops\nat the positions of CO and CN bands, indicating that these lines form in\nextended layers above the photosphere. The AMBER visibility values are best fit\nby a model without a wind. The PIONIER data are consistent with the same model.\nWe obtain a Rosseland angular diameter of 8.9+-0.3 mas, corresponding to a\nRosseland radius of 355+-55 Rsun, an effective temperature of 2640+-80 K, and a\nluminosity of log L/Lsun=3.74+-0.18. These parameters match evolutionary tracks\nof initial mass 1.5+-0.5 Msun and current mass 1.3+-0.7 Msun. The reconstructed\nPIONIER images exhibit a complex structure within the stellar disk including a\ndominant bright spot located at the western part of the stellar disk. The spot\nhas an H-band peak intensity of 40% to 60% above the average intensity of the\nlimb-darkening-corrected stellar disk. The contrast between the minimum and\nmaximum intensity on the stellar disk is about 1:2.5.\nOur observations are broadly consistent with predictions by dynamic\natmosphere and wind models, although models with wind appear to have a\ncircumstellar envelope that is too extended compared to our observations. The\ndetected complex structure within the stellar disk is most likely caused by\ngiant convection cells, resulting in large-scale shock fronts, and their\neffects on clumpy molecule and dust formation seen against the photosphere at\ndistances of 2-3 stellar radii.\n",
"title": "Aperture synthesis imaging of the carbon AGB star R Sculptoris: Detection of a complex structure and a dominating spot on the stellar disk"
} | null | null | null | null | true | null | 19974 | null | Default | null | null |
null | {
"abstract": " Long Short-Term Memory (LSTM) has achieved state-of-the-art performances on a\nwide range of tasks. Its outstanding performance is guaranteed by the long-term\nmemory ability which matches the sequential data perfectly and the gating\nstructure controlling the information flow. However, LSTMs are prone to be\nmemory-bandwidth limited in realistic applications and need an unbearable\nperiod of training and inference time as the model size is ever-increasing. To\ntackle this problem, various efficient model compression methods have been\nproposed. Most of them need a big and expensive pre-trained model which is a\nnightmare for resource-limited devices where the memory budget is strictly\nlimited. To remedy this situation, in this paper, we incorporate the Sparse\nEvolutionary Training (SET) procedure into LSTM, proposing a novel model dubbed\nSET-LSTM. Rather than starting with a fully-connected architecture, SET-LSTM\nhas a sparse topology and dramatically fewer parameters in both phases,\ntraining and inference. Considering the specific architecture of LSTMs, we\nreplace the LSTM cells and embedding layers with sparse structures and further\non, use an evolutionary strategy to adapt the sparse connectivity to the data.\nAdditionally, we find that SET-LSTM can provide many different good\ncombinations of sparse connectivity to substitute the overparameterized\noptimization problem of dense neural networks. Evaluated on four sentiment\nanalysis classification datasets, the results demonstrate that our proposed\nmodel is able to achieve usually better performance than its fully connected\ncounterpart while having less than 4\\% of its parameters.\n",
"title": "Intrinsically Sparse Long Short-Term Memory Networks"
} | null | null | null | null | true | null | 19975 | null | Default | null | null |
null | {
"abstract": " I welcome the contribution from Falessi et al. [1] hereafter referred to as\nF++ , and the ensuing debate. Experimentation is an important tool within\nempirical software engineering, so how we select participants is clearly a\nrelevant question. Moreover as F++ point out, the question is considerably more\nnuanced than the simple dichotomy it might appear to be at first sight.\nThis commentary is structured as follows. In Section 2 I briefly summarise\nthe arguments of F++ and comment on their approach. Next, in Section 3, I take\na step back to consider the nature of representativeness in inferential\narguments and the need for careful definition. Then I give three examples of\nusing different types of participant to consider impact. I conclude by arguing,\nlargely in agreement with F++, that the question of whether student\nparticipants are representative or not depends on the target population.\nHowever, we need to give careful consideration to defining that population and,\nin particular, not to overlook the representativeness of tasks and environment.\nThis is facilitated by explicit description of the target populations.\n",
"title": "Inferencing into the void: problems with implicit populations Comments on `Empirical software engineering experts on the use of students and professionals in experiments'"
} | null | null | [
"Computer Science"
]
| null | true | null | 19976 | null | Validated | null | null |
null | {
"abstract": " Recent advances in computer vision-in the form of deep neural networks-have\nmade it possible to query increasing volumes of video data with high accuracy.\nHowever, neural network inference is computationally expensive at scale:\napplying a state-of-the-art object detector in real time (i.e., 30+ frames per\nsecond) to a single video requires a $4000 GPU. In response, we present\nNoScope, a system for querying videos that can reduce the cost of neural\nnetwork video analysis by up to three orders of magnitude via\ninference-optimized model search. Given a target video, object to detect, and\nreference neural network, NoScope automatically searches for and trains a\nsequence, or cascade, of models that preserves the accuracy of the reference\nnetwork but is specialized to the target video and are therefore far less\ncomputationally expensive. NoScope cascades two types of models: specialized\nmodels that forego the full generality of the reference model but faithfully\nmimic its behavior for the target video and object; and difference detectors\nthat highlight temporal differences across frames. We show that the optimal\ncascade architecture differs across videos and objects, so NoScope uses an\nefficient cost-based optimizer to search across models and cascades. With this\napproach, NoScope achieves two to three order of magnitude speed-ups\n(265-15,500x real-time) on binary classification tasks over fixed-angle webcam\nand surveillance video while maintaining accuracy within 1-5% of\nstate-of-the-art neural networks.\n",
"title": "NoScope: Optimizing Neural Network Queries over Video at Scale"
} | null | null | [
"Computer Science"
]
| null | true | null | 19977 | null | Validated | null | null |
null | {
"abstract": " Recent research implies that training and inference of deep neural networks\n(DNN) can be computed with low precision numerical representations of the\ntraining/test data, weights and gradients without a general loss in accuracy.\nThe benefit of such compact representations is twofold: they allow a\nsignificant reduction of the communication bottleneck in distributed DNN\ntraining and faster neural network implementations on hardware accelerators\nlike FPGAs. Several quantization methods have been proposed to map the original\n32-bit floating point problem to low-bit representations. While most related\npublications validate the proposed approach on a single DNN topology, it\nappears to be evident, that the optimal choice of the quantization method and\nnumber of coding bits is topology dependent. To this end, there is no general\ntheory available, which would allow users to derive the optimal quantization\nduring the design of a DNN topology. In this paper, we present a quantization\ntool box for the TensorFlow framework. TensorQuant allows a transparent\nquantization simulation of existing DNN topologies during training and\ninference. TensorQuant supports generic quantization methods and allows\nexperimental evaluation of the impact of the quantization on single layers as\nwell as on the full topology. In a first series of experiments with\nTensorQuant, we show an analysis of fix-point quantizations of popular CNN\ntopologies.\n",
"title": "TensorQuant - A Simulation Toolbox for Deep Neural Network Quantization"
} | null | null | null | null | true | null | 19978 | null | Default | null | null |
null | {
"abstract": " The probability simplex is the set of all probability distributions on a\nfinite set and is the most fundamental object in the finite probability theory.\nIn this paper we give a characterization of statistical models on finite sets\nwhich are statistically equivalent to probability simplexes in terms of\n$\\alpha$-families including exponential families and mixture families. The\nsubject has a close relation to some fundamental aspects of information\ngeometry such as $\\alpha$-connections and autoparallelity.\n",
"title": "Information-geometrical characterization of statistical models which are statistically equivalent to probability simplexes"
} | null | null | null | null | true | null | 19979 | null | Default | null | null |
null | {
"abstract": " We present a novel condition, which we term the net- work nullspace property,\nwhich ensures accurate recovery of graph signals representing massive\nnetwork-structured datasets from few signal values. The network nullspace\nproperty couples the cluster structure of the underlying network-structure with\nthe geometry of the sampling set. Our results can be used to design efficient\nsampling strategies based on the network topology.\n",
"title": "The Network Nullspace Property for Compressed Sensing of Big Data over Networks"
} | null | null | null | null | true | null | 19980 | null | Default | null | null |
null | {
"abstract": " Every linear system of partial differential equations (PDEs) admits a scaling\nsymmetry in its dependent variables. In conjunction with other admitted\nsymmetries of linear type, the associated invariant solution condition poses a\nlinear eigenvalue problem. If this problem is structured such that the spectral\ntheorem applies, then the general solution of the considered linear PDE system\nis obtained by summing or integrating the invariant eigenfunctions (modes) over\nall eigenvalues, depending on whether the spectrum of the operator is discrete\nor continuous. By first studying the 1-D diffusion equation as a demonstrating\nexample, this method is then applied to a relevant 2-D problem from\nhydrodynamic stability analysis. The aim of this study is to draw attention to\nthe following two independent facts that need to be addressed in future studies\nwhen constructing solutions for linear PDEs with the method of Lie-symmetries:\n(i) Although each new symmetry leads to a mathematically different spectral\ndecomposition, they may all be physically redundant to standard ones and do not\nreveal a new physical mechanism behind the overall considered dynamical\nprocess, as incorrectly asserted, for example, in the recent studies by the\ngroup of Oberlack et al. Hence, with regard to linear stability analysis, no\nphysically \"new\" or more \"general\" modes are generated by this method than the\nones already established. (ii) Next to the eigenvalue parameters, each single\nmode can also acquire non-system parameters, depending on the choice of its\nunderlying symmetry. These symmetry-induced parameters, however, are all\nphysically irrelevant, since their effect on a single mode will cancel when\nconsidering all modes collectively. In particular, the collective action of all\nsingle modes is identical for all symmetry-based decompositions and thus\nindistinguishable when considering the full physical fields.\n",
"title": "On physically redundant and irrelevant features when applying Lie-group symmetry analysis to hydrodynamic stability analysis"
} | null | null | null | null | true | null | 19981 | null | Default | null | null |
null | {
"abstract": " Convolutional neural networks (CNNs) have been successfully applied on both\ndiscriminative and generative modeling for music-related tasks. For a\nparticular task, the trained CNN contains information representing the decision\nmaking or the abstracting process. One can hope to manipulate existing music\nbased on this 'informed' network and create music with new features\ncorresponding to the knowledge obtained by the network. In this paper, we\npropose a method to utilize the stored information from a CNN trained on\nmusical genre classification task. The network was composed of three\nconvolutional layers, and was trained to classify five-second song clips into\nfive different genres. After training, randomly selected clips were modified by\nmaximizing the sum of outputs from the network layers. In addition to the\npotential of such CNNs to produce interesting audio transformation, more\ninformation about the network and the original music could be obtained from the\nanalysis of the generated features since these features indicate how the\nnetwork 'understands' the music.\n",
"title": "Transforming Musical Signals through a Genre Classifying Convolutional Neural Network"
} | null | null | null | null | true | null | 19982 | null | Default | null | null |
null | {
"abstract": " Dominant approaches to action detection can only provide sub-optimal\nsolutions to the problem, as they rely on seeking frame-level detections, to\nlater compose them into \"action tubes\" in a post-processing step. With this\npaper we radically depart from current practice, and take a first step towards\nthe design and implementation of a deep network architecture able to classify\nand regress whole video subsets, so providing a truly optimal solution of the\naction detection problem. In this work, in particular, we propose a novel deep\nnet framework able to regress and classify 3D region proposals spanning two\nsuccessive video frames, whose core is an evolution of classical region\nproposal networks (RPNs). As such, our 3D-RPN net is able to effectively encode\nthe temporal aspect of actions by purely exploiting appearance, as opposed to\nmethods which heavily rely on expensive flow maps. The proposed model is\nend-to-end trainable and can be jointly optimised for action localisation and\nclassification in a single step. At test time the network predicts\n\"micro-tubes\" encompassing two successive frames, which are linked up into\ncomplete action tubes via a new algorithm which exploits the temporal encoding\nlearned by the network and cuts computation time by 50%. Promising results on\nthe J-HMDB-21 and UCF-101 action detection datasets show that our model does\noutperform the state-of-the-art when relying purely on appearance.\n",
"title": "AMTnet: Action-Micro-Tube Regression by End-to-end Trainable Deep Architecture"
} | null | null | null | null | true | null | 19983 | null | Default | null | null |
null | {
"abstract": " This work introduces a novel estimation method, called LOVE, of the entries\nand structure of a loading matrix A in a sparse latent factor model X = AZ + E,\nfor an observable random vector X in Rp, with correlated unobservable factors Z\n\\in RK, with K unknown, and independent noise E. Each row of A is scaled and\nsparse. In order to identify the loading matrix A, we require the existence of\npure variables, which are components of X that are associated, via A, with one\nand only one latent factor. Despite the fact that the number of factors K, the\nnumber of the pure variables, and their location are all unknown, we only\nrequire a mild condition on the covariance matrix of Z, and a minimum of only\ntwo pure variables per latent factor to show that A is uniquely defined, up to\nsigned permutations. Our proofs for model identifiability are constructive, and\nlead to our novel estimation method of the number of factors and of the set of\npure variables, from a sample of size n of observations on X. This is the first\nstep of our LOVE algorithm, which is optimization-free, and has low\ncomputational complexity of order p2. The second step of LOVE is an easily\nimplementable linear program that estimates A. We prove that the resulting\nestimator is minimax rate optimal up to logarithmic factors in p. The model\nstructure is motivated by the problem of overlapping variable clustering,\nubiquitous in data science. We define the population level clusters as groups\nof those components of X that are associated, via the sparse matrix A, with the\nsame unobservable latent factor, and multi-factor association is allowed.\nClusters are respectively anchored by the pure variables, and form overlapping\nsub-groups of the p-dimensional random vector X. The Latent model approach to\nOVErlapping clustering is reflected in the name of our algorithm, LOVE.\n",
"title": "Adaptive Estimation in Structured Factor Models with Applications to Overlapping Clustering"
} | null | null | null | null | true | null | 19984 | null | Default | null | null |
null | {
"abstract": " We consider the Cauchy problem in R^n for some types of damped wave\nequations. We derive asymptotic profiles of solutions with weighted\nL^{1,1}(R^n) initial data by employing a simple method introduced by the first\nauthor. The obtained results will include regularity loss type estimates, which\nare essentially new in this kind of equations.\n",
"title": "Asymptotic profile of solutions for some wave equations with very strong structural damping"
} | null | null | null | null | true | null | 19985 | null | Default | null | null |
null | {
"abstract": " Unconventional d-wave superconductors with pair-breaking edges are predicted\nto have ground states with spontaneously broken time-reversal and translational\nsymmetries. We use the quasiclassical theory of superconductivity to\ndemonstrate that such phases can exist at any single pair-breaking facet. This\nimplies that a greater variety of systems, not necessarily mesoscopic in size,\nshould be unstable to such symmetry breaking. The density of states averaged\nover the facet displays a broad peak centered at zero energy, which is\nconsistent with experimental findings of a broad zero-bias conductance peak\nwith a temperature-independent width at low temperatures.\n",
"title": "Spontaneous generation of fractional vortex-antivortex pairs at single edges of high-Tc superconductors"
} | null | null | null | null | true | null | 19986 | null | Default | null | null |
null | {
"abstract": " Neuromorphic computing has come to refer to a variety of brain-inspired\ncomputers, devices, and models that contrast the pervasive von Neumann computer\narchitecture. This biologically inspired approach has created highly connected\nsynthetic neurons and synapses that can be used to model neuroscience theories\nas well as solve challenging machine learning problems. The promise of the\ntechnology is to create a brain-like ability to learn and adapt, but the\ntechnical challenges are significant, starting with an accurate neuroscience\nmodel of how the brain works, to finding materials and engineering\nbreakthroughs to build devices to support these models, to creating a\nprogramming framework so the systems can learn, to creating applications with\nbrain-like capabilities. In this work, we provide a comprehensive survey of the\nresearch and motivations for neuromorphic computing over its history. We begin\nwith a 35-year review of the motivations and drivers of neuromorphic computing,\nthen look at the major research areas of the field, which we define as\nneuro-inspired models, algorithms and learning approaches, hardware and\ndevices, supporting systems, and finally applications. We conclude with a broad\ndiscussion on the major research topics that need to be addressed in the coming\nyears to see the promise of neuromorphic computing fulfilled. The goals of this\nwork are to provide an exhaustive review of the research conducted in\nneuromorphic computing since the inception of the term, and to motivate further\nwork by illuminating gaps in the field where new research is needed.\n",
"title": "A Survey of Neuromorphic Computing and Neural Networks in Hardware"
} | null | null | null | null | true | null | 19987 | null | Default | null | null |
null | {
"abstract": " We study junctions of Wilson lines in refined SU(N) Chern-Simons theory and\ntheir local relations. We focus on junctions of Wilson lines in antisymmetric\nand symmetric powers of the fundamental representation and propose a set of\nlocal relations which realize one-parameter deformations of quantum groups\n$\\dot{U}_{q}(\\mathfrak{sl}_{m})$ and $\\dot{U}_{q}(\\mathfrak{sl}_{n|m})$.\n",
"title": "Junctions of refined Wilson lines and one-parameter deformation of quantum groups"
} | null | null | null | null | true | null | 19988 | null | Default | null | null |
null | {
"abstract": " Stochastic gradient algorithms are more and more studied since they can deal\nefficiently and online with large samples in high dimensional spaces. In this\npaper, we first establish a Central Limit Theorem for these estimates as well\nas for their averaged version in general Hilbert spaces. Moreover, since having\nthe asymptotic normality of estimates is often unusable without an estimation\nof the asymptotic variance, we introduce a new recursive algorithm for\nestimating this last one, and we establish its almost sure rate of convergence\nas well as its rate of convergence in quadratic mean. Finally, two examples\nconsisting in estimating the parameters of the logistic regression and\nestimating geometric quantiles are given.\n",
"title": "Online estimation of the asymptotic variance for averaged stochastic gradient algorithms"
} | null | null | null | null | true | null | 19989 | null | Default | null | null |
null | {
"abstract": " The location of the terrestrial magnetopause (MP) and it's subsolar stand-off\ndistance depends not only on the solar wind dynamic pressure and the\ninterplanetary magnetic field (IMF), both of which play a crucial role in\ndetermining it's shape, but also on the nature of the processes involved in the\ninteraction between the solar wind and the magnetosphere. The stand-off\ndistance of the earth's MP and bow shock (BS) also define the extent of\nterrestrial magnetic fields into near-earth space on the sunward side and have\nimportant consequences for space weather. However, asymmetries due to the\ndirection of the IMF are hard to account for, making it nearly impossible to\nfavour any specific model over the other in estimating the extent of the MP or\nBS. Thus, both numerical and empirical models have been used and compared to\nestimate the BS and MP stand-off distances as well as the MP shape, in the\nperiod Jan. 1975-Dec. 2016, covering solar cycles 21-24. The computed MP and BS\nstand-off distances have been found to be increasing steadily over the past two\ndecades, since ~1995, spanning solar cycles 23 and 24. The increasing trend is\nconsistent with earlier reported studies of a long term and steady decline in\nsolar polar magnetic fields and solar wind micro-turbulence levels. The present\nstudy, thus, highlights the response of the terrestrial magnetosphere to the\nlong term global changes in both solar and solar wind activity, through a\ndetailed study of the extent and shape of the terrestrial MP and BS over the\npast four solar cycles, a period spanning the last four decades.\n",
"title": "The response of the terrestrial bow shock and magnetopause of the long term decline in solar polar fields"
} | null | null | null | null | true | null | 19990 | null | Default | null | null |
null | {
"abstract": " Using a membrane-driven diamond anvil cell and both ac magnetic\nsusceptibility and electrical resistivity measurements, we have characterized\nthe superconducting phase diagram of elemental barium to pressures as high as\n65 GPa. We have determined the superconducting properties of the recently\ndiscovered Ba-VI crystal structure, which can only be accessed via the\napplication of pressure at low temperature. We find that Ba-VI exhibits a\nmaximum Tc near 8 K, which is substantially higher than the maximum Tc found\nwhen pressure is applied at room temperature.\n",
"title": "Superconductivity of barium-VI synthesized via compression at low temperatures"
} | null | null | null | null | true | null | 19991 | null | Default | null | null |
null | {
"abstract": " Everyday robotics are challenged to deal with autonomous product handling in\napplications like logistics or retail, possibly causing damage on the items\nduring manipulation. Traditionally, most approaches try to minimize physical\ninteraction with goods. However, we propose to take into account any unintended\nmotion of objects in the scene and to learn manipulation strategies in a\nself-supervised way which minimize the potential damage. The presented approach\nconsists of a planning method that determines the optimal sequence to\nmanipulate a number of objects in a scene with respect to possible damage by\nsimulating interaction and hence anticipating scene dynamics. The planned\nmanipulation sequences are taken as input to a machine learning process which\ngeneralizes to new, unseen scenes in the same application scenario. This\nlearned manipulation strategy is continuously refined in a self-supervised\noptimization cycle dur- ing load-free times of the system. Such a\nsimulation-in-the-loop setup is commonly known as mental simulation and allows\nfor efficient, fully automatic generation of training data as opposed to\nclassical supervised learning approaches. In parallel, the generated\nmanipulation strategies can be deployed in near-real time in an anytime\nfashion. We evaluate our approach on one industrial scenario (autonomous\ncontainer unloading) and one retail scenario (autonomous shelf replenishment).\n",
"title": "Self-Supervised Damage-Avoiding Manipulation Strategy Optimization via Mental Simulation"
} | null | null | null | null | true | null | 19992 | null | Default | null | null |
null | {
"abstract": " Malware is constantly adapting in order to avoid detection. Model based\nmalware detectors, such as SVM and neural networks, are vulnerable to so-called\nadversarial examples which are modest changes to detectable malware that allows\nthe resulting malware to evade detection. Continuous-valued methods that are\nrobust to adversarial examples of images have been developed using saddle-point\noptimization formulations. We are inspired by them to develop similar methods\nfor the discrete, e.g. binary, domain which characterizes the features of\nmalware. A specific extra challenge of malware is that the adversarial examples\nmust be generated in a way that preserves their malicious functionality. We\nintroduce methods capable of generating functionally preserved adversarial\nmalware examples in the binary domain. Using the saddle-point formulation, we\nincorporate the adversarial examples into the training of models that are\nrobust to them. We evaluate the effectiveness of the methods and others in the\nliterature on a set of Portable Execution~(PE) files. Comparison prompts our\nintroduction of an online measure computed during training to assess general\nexpectation of robustness.\n",
"title": "Adversarial Deep Learning for Robust Detection of Binary Encoded Malware"
} | null | null | null | null | true | null | 19993 | null | Default | null | null |
null | {
"abstract": " We construct examples of finite covers of punctured surfaces where the first\nrational homology is not spanned by lifts of simple closed curves. More\ngenerally, for any set $\\mathcal{O} \\subset F_n$ which is contained in the\nunion of finitely many $Aut(F_n)$-orbits, we construct finite-index normal\nsubgroups of $F_n$ whose first rational homology is not spanned by powers of\nelements of $\\mathcal{O}$. These examples answer questions of Farb-Hensel,\nLooijenga, and Marche. We also show that the quotient of $Out(F_n)$ by the\nsubgroup generated by kth powers of transvections often contains infinite order\nelements, strengthening a result of Bridson-Vogtmann saying that it is often\ninfinite. Finally, for any set $\\mathcal{O} \\subset F_n$ which is contained in\nthe union of finitely many $Aut(F_n)$-orbits, we construct integral linear\nrepresentations of free groups that have infinite image and map all elements of\n$\\mathcal{O}$ to torsion elements.\n",
"title": "Simple closed curves, finite covers of surfaces, and power subgroups of Out(F_n)"
} | null | null | null | null | true | null | 19994 | null | Default | null | null |
null | {
"abstract": " Training convolutional networks (CNN's) that fit on a single GPU with\nminibatch stochastic gradient descent has become effective in practice.\nHowever, there is still no effective method for training large CNN's that do\nnot fit in the memory of a few GPU cards, or for parallelizing CNN training. In\nthis work we show that a simple hard mixture of experts model can be\nefficiently trained to good effect on large scale hashtag (multilabel)\nprediction tasks. Mixture of experts models are not new (Jacobs et. al. 1991,\nCollobert et. al. 2003), but in the past, researchers have had to devise\nsophisticated methods to deal with data fragmentation. We show empirically that\nmodern weakly supervised data sets are large enough to support naive\npartitioning schemes where each data point is assigned to a single expert.\nBecause the experts are independent, training them in parallel is easy, and\nevaluation is cheap for the size of the model. Furthermore, we show that we can\nuse a single decoding layer for all the experts, allowing a unified feature\nembedding space. We demonstrate that it is feasible (and in fact relatively\npainless) to train far larger models than could be practically trained with\nstandard CNN architectures, and that the extra capacity can be well used on\ncurrent datasets.\n",
"title": "Hard Mixtures of Experts for Large Scale Weakly Supervised Vision"
} | null | null | null | null | true | null | 19995 | null | Default | null | null |
null | {
"abstract": " Identifying transport pathways in fractured rock is extremely challenging as\nflow is often organized in a few fractures that occupy a very small portion of\nthe rock volume. We demonstrate that saline tracer experiments combined with\nsingle-hole ground penetrating radar (GPR) reflection imaging can be used to\nmonitor saline tracer movement within mm-aperture fractures. A dipole tracer\ntest was performed in a granitic aquifer by injecting a saline solution in a\nknown fracture, while repeatedly acquiring single-hole GPR sections in the\npumping borehole located 6 m away. The final depth-migrated difference sections\nmake it possible to identify consistent temporal changes over a 30 m depth\ninterval at locations corresponding to fractures previously imaged in GPR\nsections acquired under natural flow and tracer-free conditions. The experiment\nallows determining the dominant flow paths of the injected tracer and the\nvelocity (0.4-0.7 m/min) of the tracer front.\n",
"title": "Single-hole GPR reflection imaging of solute transport in a granitic aquifer"
} | null | null | null | null | true | null | 19996 | null | Default | null | null |
null | {
"abstract": " We analyse archival CGRO-BATSE X-ray flux and spin frequency measurements of\nGX 1+4 over a time span of 3000 days. We systematically search for time\ndependent variations of torque luminosity correlation. Our preliminary results\nindicate that the correlation shifts from being positive to negative on time\nscales of few 100 days.\n",
"title": "Episodic Torque-Luminosity Correlations and Anticorrelations of GX 1+4"
} | null | null | null | null | true | null | 19997 | null | Default | null | null |
null | {
"abstract": " We investigate some basic applications of Fractional Calculus (FC) to\nNewtonian mechanics. After a brief review of FC, we consider a possible\ngeneralization of Newton's second law of motion and apply it to the case of a\nbody subject to a constant force. In our second application of FC to Newtonian\ngravity, we consider a generalized fractional gravitational potential and\nderive the related circular orbital velocities. This analysis might be used as\na tool to model galactic rotation curves, in view of the dark matter problem.\nBoth applications have a pedagogical value in connecting fractional calculus to\nstandard mechanics and can be used as a starting point for a more advanced\ntreatment of fractional mechanics.\n",
"title": "Applications of Fractional Calculus to Newtonian Mechanics"
} | null | null | [
"Physics"
]
| null | true | null | 19998 | null | Validated | null | null |
null | {
"abstract": " We present ~0.4 resolution images of CO(3-2) and associated continuum\nemission from the gas-bearing debris disk around the nearby A star 49 Ceti,\nobserved with the Atacama Large Millimeter/Submillimeter Array (ALMA). We\nanalyze the ALMA visibilities in tandem with the broad-band spectral energy\ndistribution to measure the radial surface density profiles of dust and gas\nemission from the system. The dust surface density decreases with radius\nbetween ~100 and 310 au, with a marginally significant enhancement of surface\ndensity at a radius of ~110 au. The SED requires an inner disk of small grains\nin addition to the outer disk of larger grains resolved by ALMA. The gas disk\nexhibits a surface density profile that increases with radius, contrary to most\nprevious spatially resolved observations of circumstellar gas disks. While ~80%\nof the CO flux is well described by an axisymmetric power-law disk in Keplerian\nrotation about the central star, residuals at ~20% of the peak flux exhibit a\ndeparture from axisymmetry suggestive of spiral arms or a warp in the gas disk.\nThe radial extent of the gas disk (~220 au) is smaller than that of the dust\ndisk (~300 au), consistent with recent observations of other gas-bearing debris\ndisks. While there are so far only three broad debris disks with well\ncharacterized radial dust profiles at millimeter wavelengths, 49 Ceti's disk\nshows a markedly different structure from two radially resolved gas-poor debris\ndisks, implying that the physical processes generating and sculpting the gas\nand dust are fundamentally different.\n",
"title": "Radial Surface Density Profiles of Gas and Dust in the Debris Disk around 49 Ceti"
} | null | null | null | null | true | null | 19999 | null | Default | null | null |
null | {
"abstract": " We investigate equilibrium properties, including structure of the order\nparameter, superflow patterns, and thermodynamics of low-temperature surface\nphases of layered d_{x^2-y^2}-wave superconductors in magnetic field. At zero\nexternal magnetic field, time-reversal symmetry and continuous translational\nsymmetry along the edge are broken spontaneously in a second order phase\ntransition at a temperature $T^*\\approx 0.18 T_c$, where $T_c$ is the\nsuperconducting transition temperature. At the phase transition there is a jump\nin the specific heat that scales with the ratio between the edge length $D$ and\nlayer area ${\\cal A}$ as $(D\\xi_0/{\\cal A})\\Delta C_d$, where $\\Delta C_d$ is\nthe jump in the specific heat at the d-wave superconducting transition and\n$\\xi_0$ is the superconducting coherence length. The phase with broken symmetry\nis characterized by a gauge invariant superfluid momentum ${\\bf p}_s$ that\nforms a non-trivial planar vector field with a chain of sources and sinks along\nthe edges with a period of approximately $12\\xi_0$, and saddle point\ndisclinations in the interior. To find out the relative importance of\ntime-reversal and translational symmetry breaking we apply an external field\nthat breaks time-reversal symmetry explicitly. We find that the phase\ntransition into the state with the non-trivial ${\\bf p}_s$ vector field keeps\nits main signatures, and is still of second order. In the external field, the\nsaddle point disclinations are pushed towards the edges, and thereby a chain of\nedge motifs are formed, where each motif contains a source, a sink, and a\nsaddle point. Due to a competing paramagnetic response at the edges, the phase\ntransition temperature $T^*$ is slowly suppressed with increasing magnetic\nfield strength, but the phase with broken symmetry survives into the mixed\nstate.\n",
"title": "Spontaneously broken translational symmetry at edges of high-temperature superconductors: thermodynamics in magnetic field"
} | null | null | [
"Physics"
]
| null | true | null | 20000 | null | Validated | null | null |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.