Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_Swarms_on_Continuous_Data.txt | In order to confirm the validity of the model, we applied it to a collection of 244 images, each represented by a vector of 117 MM (Mathematical Morphology) extracted features [19], performing several tests of digital image retrieval. Results show that these images, on a continuous basis, are in many instances classified with rates above 90% (e.g., fig.2), which represent a successful framework for dealing with data in real time, without the overcome of re- training the classifier whenever new data (samples or even new classes) arrives to the system. These results, their comparison with overall data standard classifications, their logic in the recent Data-Mining area [15, 10], as well as the dynamic nature of the algorithm proposed are effusively discussed. |
Processed_Swarms_on_Continuous_Data.txt | probabilities on the lattice to go from cell k to cell i are given by Pik [5] (Eq. 3.2), where the notation j/k indicates the sum over all the pixels j which are in the local neighbourhood of k. Di measures the magnitude of the difference in orientation for the previous direction at time t-1. |
Processed_Swarms_on_Continuous_Data.txt | where n>1 determines the steepness of the threshold (normally n=2, but similar results can be obtained with other values of n>1). Now, at s = q , this probability is exactly ½. Therefore, individuals with a lower value of q are likely to respond to a lower level of stimulus. In order to take account on the number of objects present in one neighborhood, Eq. 3.3, was used (where, n now stands for the number of objects present in one neighborhood, and q = 5), defining c (Eq. 3.4) as the response threshold associated to the number of items present in a 3 x 3 region around r (one specific grid location). Now, in order to take account on the hypothetical similarity between objects, and in each ant action due to this factor, a Euclidean normalized distance d is computed within all the pairs of objects present in that 3 x 3 region around r. Being a and b, a pair of objects, and fa(i), fb(i) their respective feature vectors (being each object defined by F features), then d = (1/dmax).[(1/F).(cid:229) i=1,F(fa(i)-fb(i))2]½. Clearly, this distance d reaches its maximum (=1, since d is normalized by dmax) when two objects are maximally different, and d=0 when they are equally defined by the same F features. Then, d and e (Eqs. 3.5, 3.6), are respectively defined as the response threshold functions associated to the similarity of objects, in case of dropping an object (Eq. 3.5), and picking it up (Eq. 3.6), at site r. Finally, in every action taken by an agent, and in order to deal, and represent different stimulus intensities (number of items and their similarity), present at each site in the environment visited by one ant, the strategy used a composition of the above defined response threshold functions (Eq. 3.4, 3.5 and 3.6). Several composed probabilities were analysed and used as test functions in one preliminar test [17]. The best results (fig.2 – section 2) where achieved with test function #1 below, achieving a classification rate of 94% (out of 4 different were used, as well the LM algorithm for comparison reasons – see [16,17]). However, all data in this case (244 objects corresponding to 244 images represented by 117 MM features each – fig. 2, extracted via [19]) was nourished into the system at once. What happens if we feed them continuously, starting by little groups of data ? This work tries to answer this precise question, which is not only pertinent to real world applications as can induce better results, as was demonstrated by other authors in similar areas [9,2]. For other algorithm details please consult [16,17,18]. |
Processed_Swarms_on_Continuous_Data.txt | Instead of feeding the whole data at once to the system (as in fig. 2 [17]), the idea was to make a partition first of the data in several groups in order to test the behavior of the same algorithm on the same data (as well the final classification rate), but in contexts where the data continuously flows into the system. The 244 images of Portuguese granites [19] were then randomly separated into 6 groups. Five groups of 48 randomly samples and a final one with 4 samp les only. This data was then fed into the system at several time steps. In order to compare results with [17], the total time steps used on this experiment was 1E6 as earlier. The first group of 48 samples (group A) was fed at t=0, the second (B) at t=10000, the third one (C) at t=20000, the fourth one (D) at t=30000, the fifth one (E) at t=40000, and finally the sixth group (F) of only four samples at t=50000. Now, in order to have an idea of the classification rate we measured it at several time steps and have used the same method as in [17], that is k-NNR - Nearest Neighbor Rule – i.e., a label of an unknown item is determined by the label of his first k neighbors; k=3 was used as before. At each time step measured (see fig. 3), 20 % samples (several sub-sets were used – 10 to be precise) were randomly chosen as a test set, using k-NNR and the remaining samples at that time (a sort of training data) to found their labels. Using this method, figure 3 shows the average classification rate at each time step. |
Processed_Swarms_on_Continuous_Data.txt | Figure 3 – Average classification rate achieved via k-NNR (using 10 sub-groups of 20% samples as test and 80% as training) at several time steps (log). |
Processed_Swarms_on_Continuous_Data.txt | benchmark comparison of the above cited methods should be interesting to explore, the ability of this algorithm to perform continuous mappings and the incapacity of the latter to conceive it, tend to difficult any serious comparison. Suitable dynamic benchmarks of data, which can be continously fed into any classifier system should be created, in order to make serious comparisons. |
Processed_Revisiting_Panel_Data_Discrete_Choice_Models_with_.txt | This paper revisits the identification and estimation of a class of semiparametric (distribution- free) panel data binary choice models with lagged dependent variables, exogenous covariates, and entity fixed effects. We provide a novel identification strategy, using an “identification at infinity” argument. In contrast with the celebrated Honor´e and Kyriazidou (2000), our method permits time trends of any form and does not suffer from the “curse of dimen- sionality”. We propose an easily implementable conditional maximum score estimator. The asymptotic properties of the proposed estimator are fully characterized. A small-scale Monte Carlo study demonstrates that our approach performs satisfactorily in finite samples. We illustrate the usefulness of our method by presenting an empirical application to enrollment in private hospital insurance using the Household, Income and Labor Dynamics in Australia (HILDA) Survey data. |
Processed_Spatially_Resolved_STIS_Spectroscopy_of_Betelgeuse.txt | We present spatially resolved spectra observed with the Space Tele- scope Imaging Spectrograph on the Hubble Space Telescope of the upper chromosphere and dust envelope of α Orionis (M2 Iab). In the fall of 2002 a set of five high-resolution near-UV spectra was obtained by scan- ning at intensity peak-up position and four off-limb target positions up to one arcsecond, using a small aperture (200 by 63 mas), to investigate the thermal conditions and flow dynamics in the outer atmosphere of this important nearby cool supergiant star. |
Processed_Spatially_Resolved_STIS_Spectroscopy_of_Betelgeuse.txt | Based on Mg ii h & k, Fe ii λ2716, C ii λ2327, and Al ii ] λ2669 emission lines we provide the first evidence for the presence of warm chromospheric plasma at least 1 arcsecond away from the star at ∼40 R∗ (1 R∗≃700 R⊙). The STIS spectra reveal that Betelgeuse’s upper chromosphere extends far beyond the circumstellar Hα envelope of ∼5 R∗, determined from previous ground-based imaging (Hebden et al. 1987). |
Processed_Spatially_Resolved_STIS_Spectroscopy_of_Betelgeuse.txt | The flux in the broad and self-absorbed resonance lines of Mg ii decreases by a factor of ∼700 compared to the flux at chromospheric disk center. We observe strong asymmetry changes in the Mg ii h and Si i resonance line profiles when scanning off-limb, signaling the outward acceleration of gas outflow in the upper chromosphere. |
Processed_Spatially_Resolved_STIS_Spectroscopy_of_Betelgeuse.txt | From the radial intensity distributions of Fe i and Fe ii emission lines we determine the radial non-LTE iron ionization balance. We compute that the local kinetic gas temperatures of the warm chromospheric gas component in the outer atmosphere exceed 2600 K, when assuming lo- cal gas densities of the cool gas component we determine from radiative transfer models that fit the 9.7 µm silicate dust emission feature. The spatially resolved STIS spectra directly demonstrate that warm chromo- spheric plasma co-exisists with cool gas in Betelgeuse’s circumstellar dust envelope. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | We present the morphological catalog of galaxies in nearby clusters of the WINGS survey (Fasano et al. 2006). The catalog contains a total number of 39923 galaxies, for which we provide the automatic estimates of the morphological type applying the purposely devised tool MORPHOT to the V-band WINGS imaging. For ∼3000 galax- ies we also provide visual estimates of the morphological types. A substantial part of the paper is devoted to the description of the MORPHOT tool, whose application is limited, at least for the moment, to the WINGS imaging only. The approach of the tool to the automation of morphological classification is a non parametric and fully empiri- cal one. In particular, MORPHOT exploits 21 morphological diagnostics, directly and easily computable from the galaxy image, to provide two independent classifications: one based on a Maximum Likelihood (ML), semi-analytical technique, the other one on a Neural Network (NN) machine. A suitably selected sample of ∼1000 visually clas- sified WINGS galaxies is used to calibrate the diagnostics for the ML estimator and as a training set in the NN machine. The final morphological estimator combines the two techniques and proves to be effective both when applied to an additional test sample of ∼1000 visually classified WINGS galaxies and when compared with small samples of SDSS galaxies visually classified by Fukugita et al. (2007) and Nair et al. (2010). Finally, besides the galaxy morphology distribution (corrected for field contamination) in the WINGS clusters, we present the ellipticity (ε), color (B-V) and Sersic index (n) distributions for different morphological types, as well as the morphological fractions as a function of the clustercentric distance (in units of R200). |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | nearby clusters (0.04(cid:54)z(cid:54)0.07), selected from three X-ray flux limited samples compiled from ROSAT All-Sky Survey data (Ebeling et al. 1996, 1998, 2000). The observations in the optical bands have been obtained with the WFC@INT and with the WFI@MPG/ESO-2.2 cameras for the northern and southern clusters, respectively. Follow-ups of the opti- cal WINGS survey include medium-resolution, multi-fiber spectra of ∼6000 galaxies in 48 WINGS clusters (Cava et al. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | A recent, comprehensive review of the various aspects and issues linked to galaxy morphology can be found in Buta (2011). Till a dozen years ago the morphological type esti- mate of galaxies has been obtained just by visual inspec- tion of photografic plates or CCD frames. A few attempts were actually made in the 90’s to obtain automated morpho- logical classification of galaxies (with neural networks and self-organizing maps; Naim et al. 1995, 1997), but they re- mained isolated. In the last decade, the sudden availability of CCD mosaics has made it no longer feasible to conduct mor- phological classifications by eye, since one has often to deal with wide and/or deep fields, each one containing thousands galaxies. This has triggered a number of papers proposing different tools for automatic morphology estimate of large galaxy samples. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | Figure 1. Visual morphology versus Sersic’s index (top panel) and bulge fraction (bottom panel) for the 527 galaxies in common between the MORPHOT calibration sample (see Section 3.1) and the WINGS-GASP2D sample (S´anchez-Janssen et al. 2011). |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | while the first one is a global property that can be described by means of simple analytical laws and leaves mostly aside the problems connected with image texture, S/N ratio and resolution (apart from the convolution with the local PSFs), the second one mainly deals with pixel-scale behaviours and features which, in default of human eyes, make difficult any quantitative description. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | nostic devised by Abraham et al. (1996, concen- tration vs. asymmetry) and the popular CAS (Concen- tration/Asymmetry/clumpinesS) diagnostic set (Conselice 2003), it is worth mentioning those proposed by Abraham et al. (2003, Gini Coefficient) and Lotz et al. (2004, M20 coefficient), Lauger et al. (2005, concentration and asymme- try at different wavelenghts), Yamauchi et al. (2005, concen- tration and coarseness coefficients), Menanteau (2006), van der Wel (2008) and Petty (2009). To the second category belong the tools devised by Odewahn et al. (2002, Fourier analysis of the images), Ball et al. (2004), Goderya et al. (2004), de la Calleja and Fuentes (2004), Kelly and McKay (2004, shapelet analysis), Moore et al. (2006), Scarlata et al. (2007, ZEST), Huertas-Company et al. (2008) and Shamir (2009). A mixed approach (B/T decomposition + non para- metric diagnostics) has been tried by Rahman and Shan- darin (2004), Cheng et al. (2011) and Vikram et al. (2010, PyMorph). |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | The non-parametric approach seems to be more effec- tive than the parametric one in estimating the morphological type of galaxies (Hatziminaoglou et al. 2005) and has been claimed to be even able to compare with visual estimates as far as the intrinsic scatter and the robustness of the re- sults are concerned (Odewahn et al. 2002; Bell et al. 2004; Huertas-Company et al. 2008). However, a common limita- tion of the non-parametric tools available in the literature is the scarce ability of separating S0s from elliptical galax- ies, which is actually an important issue when dealing with galaxy evolution in clusters (Dressler et al. 1997; Fasano et al. 2000; Treu et al. 2003; Postman et al. 2005; Desai et al. 2007; Poggianti et al. 2009). |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | In this paper we describe a new automatic, non- parametric tool for the morphological type estimate of large galaxy samples (MORPHOT Fasano et al. 2007), which is in fact able to separate Es from S0s in a majority of cases. MORPHOT belongs to the second previously mentioned category of non parametric tools. It starts with a set of 21 suitably devised morphological diagnostics, and combines them in two different (independent) ways, thus producing the final morphological type (and the relative confidence in- terval) for each galaxy in a given input catalog. We fine-tune MORPHOT for extensive application to the WINGS cluster sample and present the catalogs of the survey, which contain morphological types of ∼40000 galaxies. We stress that, al- though the basic methodology is robust for any set of digital images of similar spatial resolution and dynamic range, at this stage the tool does not pretend to have a general valid- ity, regardless of the observing conditions (telescope, detec- tor, seeing) and the galaxy sample (redshift). However, we will show that it produces reliable results for the particular purposes of the WINGS survey. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | After the pioneering attempt by Raynolds (1920) to provide a morphological classification of spi- ral nebulae and since the first, definite understand- ing by Edwin Hubble (1925) of the extragalactic na- ture of many nebulae, a number of different classifica- tion schemes have been proposed for the galaxy morphol- ogy. The original Hubble’s sequence (Hubble 1922, spi- rals/elongated/globular/irregulars) was improved by the au- thor himself, first introducing the concept of tuning fork to distinguish between normal and barred disk galaxies (Hub- ble 1926), then defining the S0 morphological type (Hub- ble 1936). Later on, the Hubble system has been refined and completed, introducing spiral types later than Sc (Sd and Sm; de Vaucouleurs 1959), a new type of amorphous galaxies (Irr-II; Holmberg 1950) and ring-based (de Vau- couleurs 1963) or arm-based (Elmegreen et al. 1987) dis- tinctions among disk galaxies. |
Processed_Morphology_of_galaxies_in_the_WINGS_clusters.txt | Radically different classification schemes were proposed by Morgan (1958) and van den Bergh (1959, 1960a,b). The first one links morphology with central concentration of light and stellar populations, also introducing the new cD type. The second one links morphology with total luminosity (lu- minosity classes) and extends the basic Hubble’s scheme (E/Sp/Irr) to the lowest luminosity galaxies (dwarfs). |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | On 1 June 2019, just before 7:30 PM local time, the Desert Fireball Network detected a -9.3 mag- nitude fireball over South Australia near the Western Australia border. The event was observed by six fireball observatories, and lasted for five seconds. One station was nearly directly underneath the trajectory, greatly constraining the trajectory solution. This trajectory’s backward numerical integra- tions indicate that the object originated from the outer main belt with a semi-major axis of 2.75 au. A light curve was also extracted and showed that the body experienced very little fragmentation dur- ing its atmospheric passage. A search campaign was conducted with several Desert Fireball Network team members and other volunteers. One 42 g fragment was recovered within the predicted fall area based on the dark flight model. Based on measurements of short-lived radionuclides, the fragment was confirmed to be a fresh fall. The meteorite, Arpu Kuilpu, has been classified as an H5 ordinary chondrite. This marks the fifth fall recovered in Australia by the Desert Fireball Network, and the smallest meteoroid ((cid:39) 2 kg) to ever survive entry and be recovered as a meteorite. |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | The Desert Fireball Network (DFN) is a system of automated photographic all-sky fireball observatories covering over 2.5 million km2 of Australian outback (Howie et al. 2017a). It is the largest single photographic fireball network in the world. The DFN’s primary objective is to recover meteorite falls using fireball observations. These meteorites with precise orbital information can help us better understand small bodies and the debris they generate in the solar system. The DFN is a partner network of the Global Fireball Observatory (GFO), a collaborative project consisting of 18 international partners from all over the world (Devillepoix et al. 2020). Our observatories are optimized to precisely observe bright meteors (“fireballs”) when they impact the atmosphere, providing invaluable information about the asteroidal debris in the inner solar system (Devillepoix et al. 2019). |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | Over forty meteorite falls in total have been recovered with the assistance of fireball observations (Granvik & Brown 2018; Devillepoix et al. 2020; Colas et al. 2020; Gardiol et al. 2021). The most numerous of the meteorite fall types recovered are H-chondrites. This is somewhat unexpected, as the most common Antarctic meteorite type are L- chondrites, followed by H-chondrites (Binzel et al. 2015). More meteorite fall recoveries can clarify whether this trend is significant. Within this study, we discuss the DFN fireball observations and data reduction, which lead to the recovery of Arpu Kuilpu1, an H5 chondrite. This is the fifth meteorite recovered in Australia by the DFN (Bland et al. 2009; Devillepoix et al. 2018; Dyl et al. 2016; Sansom et al. 2020). |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | In total, six DFN camera systems detected the event internally referenced as DN190601 01 (Tab. 1), using the on-board algorithm of Towner et al. (2020). The location of these are mapped in Figure 1 along with the fireball as observed by each system. |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | The DFN captures long exposure images and obtains timing information from de Bruijn sequence dashes (Howie et al. 2017a,b; Howie et al. 2020). This methodology of encoding a de Bruijn sequence into the shutter frequency allows us to obtain absolute timing of the fireballs without the need of an additional subsystem – drastically reducing the size, cost, and power requirements. |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | Observing conditions during the fireball were ideal, enabling three cameras to observe from over 400 km away. Encoded timing was acquired for five sites, but was not resolvable for DFNEXT051. The best convergence angle between the four observations was a reasonable 36◦. However, one camera, DFNEXT041 Hughes, was nearly directly underneath the fireball trajectory. This observation significantly helped to constrain the fit and was instrumental in recovering the small 42 g stone meteorite. The observations from the six DFN fireball observatories provided 256 data points to fit the trajectory (Figure1). |
Processed_Arpu_Kuilpu:_An_H5_from_the_Outer_Main_Belt.txt | Table 1. Locations of DFN Observatories that obtained photographic records of DN190601 01, and nature of data obtained. Times are relative to first fireball observation at 09:53:04.326 UTC on the 1st of June 2019 (from the Hughes camera). Timing information was unable to be extracted from the observations made from DFNEXT051 - Kanandah. P: Photographic record (long-exposure high resolution image, see Sec. 2.1), V: lossless compressed digital video (30 frames per second, see Sec. 2.2). Ranges are from when the meteoroid was at 65 km altitude. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Jackendoff (2002) posed four challenges that linguistic combinatoriality and rules of language present to theories of brain function. The essence of these problems is the question of how the rapid construction and transformation of the compositional structures that are typically taken to be the domain of symbolic processing. typical connectionist approaches fail to meet these challenges and that the dialogue between linguistic theory and cognitive neuroscience will be relatively unproductive until the importance of these problems is widely recognised and the challenges answered by some technical innovation in connectionist modelling. This paper claims that a little-known family of connectionist models (Vector Symbolic Architectures) are able to meet Jackendoff(cid:146)s challenges. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Introduction Jackendoff (2002) has posed four linguistic challenges for cognitive neuroscience. He holds that language is a mental phenomenon and that linguistic functionality must be neurally instantiated. However, (cid:147)although a great deal is known about functional localization of various aspects of language in the brain (cid:133) nothing at all is known about how neurons instantiate the details of rules of grammar (cid:148) (Jackendoff, 2002, p. 58). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Jackendoff's challenges are an attempt to counter this risk by focussing attention on functionality that he sees as central to all linguistic phenomena. The need for these challenges is shown by the lack of connectionist models able to deal with them successfully and the fact that, despite this lack of success, the challenges (cid:147)have not been widely recognized [as such] in the cognitive neuroscience community(cid:148) (Jackendoff, 2002, p. 58). Feldman (2002) broadcast these challenges to the neural network modelling community via the Connectionists Mailing List. The few responses he received were unable that any standard connectionist techniques would meet Jackendoff(cid:146)s challenges. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The challenges are not exclusively linguistic and are arguably fundamental to all cognition. They are (cid:147)not something that can simply be disregarded by ignoring language, (cid:133) but they certainly come to the fore in dealing with the linguistic phenomena that linguists deal with every day(cid:148) (Jackendoff, 2002, p. 67). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Vector Symbolic Architectures (VSAs) are a little- known class of connectionist models that can directly implement functions usually taken to form the kernel of symbolic processing (Gayler, 1998; Kanerva, 1997; Plate, 1994; Rachkovskij & Kussul 2001). They are an enhancement of tensor product variable binding networks (Smolensky, 1990). Like tensor product networks, VSAs can create and manipulate recursively- structured representations in a natural and direct connectionist training. However, unlike tensor product networks, VSAs afford a practical basis for implementations because they require only fixed dimension vector representations. The fact that VSAs relate directly, without training, to both simple, practical, vector implementations and core symbolic processing functionality suggests that they would provide a fruitful connectionist basis for the implementation of cognitive functionality. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The lack of progress by cognitive neuroscientists may also be due to attempting to solve the wrong problems, because of holding naive views of linguistic phenomena (Jackendoff, 2002, chap. 3, notes 17 & 21). This apparent naivete may be a conscious research strategy, to start simple, with the intent that current solutions will scale up to ultimate needs. However, this strategy must fail if the initially chosen problems are not actually core to the ultimate problems. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The approach taken in this paper is constrained by the required brevity. Jackendoff's analysis of the linguistic challenges and the inadequacy of typical cognitive neuroscience models is accepted at face value. The range and properties of VSAs are discussed briefly. The presentation of VSAs and their comparison with other connectionist approaches is minimal, as both topics are adequately documented elsewhere. Then I deal with each of Jackendoff's challenges in turn, showing how they are answered by VSAs. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Vector Symbolic Architectures The name (cid:147)Vector Symbolic Architectures(cid:148) has been invented to cover a family of related approaches. These approaches are easily implemented as connectionist to algebraic systems and share a commitment operations on distributed representations over high- dimensional vectors. these approaches are descended from Smolensky's (1990) tensor product variable binding networks. Smolensky the demonstrated representation and manipulation of complex nested structures using connectionist methods is possible. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | His tensor product approach relies on algebraic operations with simple connectionist implementations. This results in structural manipulation occurring in a single pass through the system, thus avoiding the need for prolonged learning (unlike multilayer perceptrons with back-propagation). Unfortunately, tensor product implementations are thoroughly impractical because the vector dimensionality increases exponentially with the depth of the structures to be represented. Consequently, tensor product binding has been in subsequent research. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | VSAs (Gayler, 1998; Kanerva, 1997; Plate, 1994; Rachkovskij & Kussul, 2001) retain the advantages of tensor product binding while avoiding the problem of increasing vector dimensionality. In all connectionist systems, entities are represented by vectors. (The activity levels of a set of connectionist units are construed mathematically as a vector.) In tensor product binding, the representation of the association (or binding) of two entities is created as the outer product of the vectors representing the two entities. Thus, if the entity vectors are of dimensionality n, the outer product will be of dimensionality n2. VSAs overcome this problem of increasing dimensionality by applying a function to the n2 elements of the outer product to yield a resultant vector of dimensionality n. Thus all structures, whether atomic or complex, are represented by vectors of the same dimensionality. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The application of VSAs to cognitive problems is arguably more complex the application of conventional multilayer perceptrons with back- propagation. Three separate levels of description are required. These levels are: the vector representation, the representational architecture and the cognitive architecture. This paper, like other papers on VSAs, deals primarily with the vector representation level and only touches on the representational architecture to the minimum extent necessary. The contention is that VSAs examined at these levels are manifestly better suited the connectionist alternatives and that the extension of the analyses through the representational and cognitive architecture levels can be reasonably expected as a consequence of the ongoing VSA research program. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The currently available VSAs employ three types of operation on vectors: a multiplication-like operator, an addition-like operator, and a permutation-like operator. The precise choice for each operator varies by VSA. Gayler the implementations of some VSAs. The multiplication- like operation is used to associate or bind vectors. The addition-like operation is used to superpose vectors or add them to a set. The permutation-like operation is used to quote or protect vectors from the other operations. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | For the sake of concreteness, in examples I will use the MAP (Multiply, Add, Permute) Coding scheme In MAP Coding, described elementwise multiplication is used to implement the binding of vectors; elementwise addition is used to implement and superposition permutation of the elements is used to implement quotation of vectors. These operations can be used to compose, decompose, complex structures without requiring any training (Gayler, 1998). Analogous descriptions of other VSAs can be found in Kanerva (1997), Plate (1994), and Rachkovskij & Kussul (2001). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The description of VSAs has so far been at the vector representation level, being purely in terms of the content of vectors. Corresponding to each operation there is a representational architecture primitive. For example, to implement binding there would be a layer of connectionist units taking two inputs each and computing These architectural primitives are combined in some fixed circuit to yield a complete representational architecture. The cognitive architecture level deals with how a cognitive problem is represented in terms of vector representations and prior knowledge in order to obtain the desired behaviour from a given representational architecture. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | of only a few objects and attributes. Jackendoff then shows that the binding problem is much larger in linguistics. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Jackendoff refers repeatedly to the example sentence (cid:147)The little star's beside a big star(cid:148) (2002, p. 5). He gives a representation of the structure encoded in the sentence on the phonological, syntactic, semantic/ conceptual, and spatial levels (fig. 1.1). My informal count of this structure shows approximately 130 tokens and 160 relations between tokens. The number of bindings involved in the composition of the structure corresponding to this simple sentence is obviously orders of magnitude larger than in the typical visual example of binding. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | There are two other aspects of the binding problem that must be mentioned: novelty and speed. Although the individual components at the lowest level will be familiar, the total composite will frequently be novel. Furthermore, this novel composite must be constructed in the time it takes to comprehend a single sentence. Thus, approaches relying on extensive training to construct each composite structure are implausible. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Response 1 Binding in MAP Coding is implemented as the elementwise product of the vectors to be bound. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | bind(a,b) = a*b Unbinding in VSAs is implemented as the binding of the inverse of the cue to the trace. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | unbind(cue, trace) = bind(inverse(cue), trace) These functions yield the desired behaviour for binding and unbinding in that a cue can be used to retrieve components from a bound composite. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | unbind(a, bind(a,b)) = b unbind(b, bind(a,b)) = a The mathematical details of binding and unbinding in VSAs and the extension to noisy cues and traces can be found in Plate (1994) and Kanerva (1997). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | bind(a,a) = a*a = 1 and no separate inverse function is required. In this case, the multiplication-like operation can be construed as implementing a number of functions (binding, unbinding, and others), depending on the relationship between the vectors being multiplied (Gayler & Wales, 2000). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | blindness of binding to the contents of the vectors to be bound is that the entities to be bound may be composite structures the construction of complex recursive structures (Plate, 1994; Smolensky, 1990). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The vector representation level of VSAs clearly has the capability to deal, in principle, with the binding problem. The representational architecture primitives have the required characteristics of speed and ability to cope with novelty. It is trivially easy to construct a representational architecture that implements a fixed compositional template. However, there is no currently available representational architecture and cognitive architecture that implements fully variable, input driven composition. My current research (Gayler & Wales, 2000) is aimed at this objective. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Connectionist recurrent associative memories (e.g. Anderson, Silverstein, Ritz, & Jones, 1977) work by creating attractors corresponding to each of the items to be remembered. My research aims to create an enhanced that has recurrent associative memory attractors corresponding to novel, valid compositions of familiar items. Such a memory would be able to recognise novel the components and generating mappings between them. A similar approach, not based on VSAs, has already been the visual domain successfully demonstrated (Arathorn, 2002). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Response 2 In VSAs, superposition (implemented by the addition- like operator) is used to represent multiple items in the same representational space (over the same set of connectionist units). the representational architecture level can be used to add items to a superposition and operate on a set of superposed items simultaneously (Kanerva, 1997; Plate, 1994). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | distinct from the big star by virtue of (among other things) the difference in size. This difference can be captured by representing each star with a composite structure that includes the size attribute. If the composite representations are distinct they can be superposed without losing their identities. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | At the vector representation level, superposition is a similarity preserving operation, whereas binding is a similarity destroying operation (where the similarity of two vectors is measured by their dot product). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | little . big = 0 (little * star) . (big * star) = 0 These two vectors (little*star and big*star) may be safely superposed without losing their identities. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | frame-2 * (is-a*star + colour*red) = frame-2*is-a*star + frame-2*colour*red When these two representations are superposed the properties of the stars are kept separate. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | This encoding begs the question of the origin of the frame identity tokens (frame-1 and frame-2). A good solution to this problem is to calculate the frame identity as the permutation of the frame contents (Gayler, 1998). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Construction of a representational architecture to implement the make-frame function is trivial. This architecture consists of only one permutation operator and one binding operator. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Challenge 3: The Problem of Variables The third challenge arises from the productivity of language. Language users are able to recognise and generate an infinite variety of utterances using only finite resources. This productivity is construed as arising from the use of variables; placeholders which may contain arbitrary values (Jackendoff, 2002, p. 64). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | A rule of a grammar can be construed as a template with variables. When those variables are instantiated, the remainder of the template relates their values to each other. Thus, a rule can be seen as a mechanism for constructing or recognising a composite structure, and the variables of the rule as markers for the components to be productively replaced. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The variables in rules are generally (cid:147)typed(cid:148). That is, the set of values with which a variable may be instantiated must be constrained. The type of a variable is the constraint on its possible values. Traditionally, rules are seen as quite distinct from the structures that they operate on and types are annotations on variables in rules. However, this distinction between rules and structures can be erased. In a thoroughly lexicalised grammar, the information in rules is expressed as structural fragments (like the structures operated on by the rules) which are combined by (cid:147)the only procedural rule (cid:133) UNIFY ((cid:145)clip(cid:146) structures together)(cid:148) (Jackendoff, 2002, p. 182). In such a grammar, the compatibility constraints (types on variables) can be captured in the structure of the fragments. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | imperative the programming language interpretation of a variable as a location to hold a value. However, there is an alternative interpretation of the concept of a variable that is based on declarative programming languages and to constraint-based grammar is more congenial formalisms (Copestake, 2002). This interpretation treats variables as targets for substitution. This interpretation can also be implemented with VSAs. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | A substitution may be represented as the binding of the representations to be substituted for each other (Gayler, 2000). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | make-substitution(a, b) = bind(a, b) The substitution may be applied to a structure by binding the substitution with the structure. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | apply-substitution(make-substitution(x,b), bind(x,y)) = (x*b) * (x*y) = b*y = bind(b,y) Given this capability, any component of a structure can act as a variable (a target for substitution). The only constraint is that if the same component has multiple occurrences the structure in substitution is applied identically to all occurrences. apply-substitution(make-substitution(x,b), (x*y + x*z)) = (b*y + b*z) This suggests a style of processing in which only literal episodes are stored and rules arise as statistical those regularities of potential substitutions across episodes, similar to the Data Oriented Processing of Bod and Scha (1997). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Challenge 4: Binding in Working Memory vs Long-Term Memory Jackendoff's fourth challenge concerns the transparency of the boundary between working memory and long- term memory. He argues that linguistic tasks require the same structures to be instantiated in working memory and long-term memory and that the two instantiations should be functionally equivalent (2002, p. 65). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Response 4 Jackendoff has cast this problem as arising from the difference in physical implementation of working memory and long-term memory. I recast this problem as arising from differences at the vector representation In typical connectionist systems, working level. represented as vectors (of memory activations) are items represented as matrices (of synaptic connectivities). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The crucial difference here is not the physical implementation (activations versus connectivities), but the logical form of the representations (vectors versus matrices). Of course, the synaptic matrices can be reinterpreted as vectors, but they will be of different dimensionality to the activation vectors. That is, the working memory items and long-term memory items exist in different, incommensurable representational spaces. This is hardly surprising, as the synaptic connectivities can be construed as relations between working memory items. They are the mechanism by which items in working memory create other items in working memory. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | In VSAs, working memory items and long-term memory items have the same form at the vector representation level. They are both represented by vectors of the same dimensionality. Those vectors can represent items and relationships between items both in working memory and This equivalence of logical form makes the boundary between working memory and long-term memory transparent. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | implementation are in physical reflected not in what can be represented but in the persistence of those representations and their ability to interact with each other. Working memory items are able to interact with each other and items in long-term memory, whereas long-term memory items are only able to interact with each other via items in working memory. This issue has been discussed with respect to a different set of challenges in Gayler and Wales (1998). |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | The issue of speed of learning has two components: speed of binding and speed of laying down a trace. As discussed in response to the first challenge, binding in VSAs arises from the algebraic properties of the operations and occurs in a single pass through an architectural primitive. Thus, associations can be formed in a single trial. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | pass through the architectural primitive that implements the addition-like operation. This could have multiple for example, a direct physical increment levels, or short-term potentiation of the representational units. Long-term persistence is achieved in functionally equivalent ways with different implementations, for example, long-term potentiation of representational units, or increments to the synaptic connectivities. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Conclusion Jackendoff posed four challenges that are generally to relevant functional language. capabilities that are required for language, but arguably not provided by typical connectionist models. VSAs, a little-known family of connectionist architectures, do meet these challenges. Their ability to meet these challenges arises from the algebraic properties of their vector representations. These properties and the fact that they have straightforward physical implementations suggest that VSAs are ideal candidates for cognitive modelling. However, it has to be asked why VSAs are so relatively little-known. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | There are multiple reasons that could be presented, but I will offer only one. While VSAs are very easy to work with at the vector representation level and the level of primitives of the representational architecture, they are very difficult to work with at the level of complete representational architectures and cognitive architectures. Typical connectionist architectures rely on training procedures to achieve their effectiveness. However, VSAs provide no opportunity for training to substitute for architectural effectiveness. That is, good performance depends on good design rather than automated training, and this is a harder research task. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Arathorn, D. W. (2002). Map-seeking circuits in visual cognition: A computational mechanism for biological and machine vision. Stanford, CA, USA: Stanford University Press. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Gayler, R. W., & Wales, R. (2000, February). Multiplicative binding, representation operators and analogical inference. Paper presented at the 5th Australasian Cognitive Conference, Melbourne, Australia. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Jackendoff, R. (2002). Foundations of language: Brain, meaning, grammar, evolution. Oxford, UK: Oxford University Press. |
Processed_Vector_Symbolic_Architectures_answer_Jackendoff's_.txt | Smolensky, P. (1990). Tensor product variable binding and the representation of symbolic structures in connectionist systems. Artificial Intelligence, 46, 159-216. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Abstract—We report on the design and estimated performance of a balloon-borne hard X-ray polarimeter called HX-POL. The experiment uses a combination of Si and Cadmium Zinc Telluride detectors to measure the polarization of 50 keV-400 keV X-rays from cosmic sources through the dependence of the angular distribution of Compton scattered photons on the polarization direction. On a one-day balloon flight, HX-POL would allow us to measure the polarization of bright Crab- like sources for polarization degrees well below 10%. On a longer (15-30 day) flight from Australia or Antarctica, HX-POL would be be able to measure the polarization of bright galactic X-ray sources down to polarization degrees of a few percent. Hard X-ray polarization measurements provide unique venues for the study of particle acceleration processes by compact objects and relativistic outflows. In this paper, we discuss the overall instrument design and performance. Furthermore, we present results from laboratory tests of the Si and CZT detectors. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | probe3 would use coded mask imaging to scrutinize the entire sky every 170 min in the 5 keV to 600 keV energy band. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | This paper focuses on X-ray polarimetry, another very promising and largely unexplored area. X-ray polarimetry would increase the parameter space for the study of compact objects like black holes and neutron stars from two dimensions (time variability and energy spectra) to four dimensions, by adding two qualitatively new parameters: polarization degree and polarization direction [1], [2], [3], [4]. The polarization measurements can be used to identify the processes respon- sible for the observed X-ray emission, to constrain viewing perspectives and opacities in accretion disk systems, and to gain access to the properties of the magnetic field (order and orientation) in the emission regions. Recently, concepts have been developed to measure the polarization of soft X-rays by tracking the photo-effect electrons that tend to be ejected parallel to the electric field vector of the X-rays with gas pixel detectors [5] and with time projection chambers [6]. At >20 keV energies, the Compton effect can be used to measure the polarization, as photons are preferentially scattered in the direction perpendicular to the electric field vector. Early soft X-ray polarization measurements were done with the OSO 8 experiment which used mosaic crystals of graphite to yield polarization-sensitive Bragg reflection of 2.6 keV and 5.2 keV X-rays [7]. The Ge detectors on board the INTEGRAL γ-ray observatory were used to measure the polarization of gamma- rays based on Compton effect polarimetry [8]. Recently, NASA approved the Gravity and Extreme Magnetism SMEX (GEMS) mission as a “small explorer mission”. GEMS will combine grazing incident X-ray mirrors with time projection chamber detectors to measure the polarization of 2-10 keV X- rays, and will be launched in the next decade [9]. The japanese ASTRO-H mission (formerly called “NeXT”) will have a “soft γ-ray telescope” which uses Si and CdTe detector stacks and will have polarization sensitivity in the γ-ray energy regime [10]. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 1. Conceptual design of the HX-POL balloon payload. The polarization is measured with a Si-CZT detector configuration inside a 2 m high collimator- shield assembly. The collimator is pointed towards a source by rotating the gondola and by changing the elevation. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 2. “Exploded view” of the HX-POL detector assembly. A total of 32 Si detectors (each 0.2 cm×10 cm×10 cm, active detector area: 81 cm2) are tiled and stacked to make a 1.6 cm thick detector module with an active detection area of 324 cm2. The Si detectors are positioned “above” one hundred CZT detectors (each 0.5 cm×2 cm×2 cm) with an active detection area of 400 cm2. The ASIC readout of the Si detectors is located at the perimeter of the Si detectors and the readout of the CZT detectors is located below the CZT detectors. The Si-CZT detectors are positioned above a 2 cm×25 cm×25 cm CsI active rear shield. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | HX-POL is a concept of a balloon-borne X-ray polarimeter that can measure the polarization of bright X-ray sources in the 50-400 keV energy range. Sketches of the gondola and the main detector assembly are shown in Figs. 1 and 2, respectively. The main detector assembly is made of Si and CZT detectors with an active detector area of 324 cm2 (Fig. 2). Photons Compton scatter in the low-Z Si and are photo- absorbed in the high-Z CZT detectors. The polarization can be inferred by analyzing the distribution of the azimuthal scat- tering angles. In addition to giving a high fraction of “golden events” with exactly one interaction in a Si detector and one in a CZT detector, the high-Z/low-Z detector combination avoids the background of neutron events which can mimic Compton scattered photons. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | with 64 pixels (pixel pitch: 2.5 mm). The data acquisition is triggered by a >20 keV hit in one of the CZT detectors with or without a coincident hit in the Si detectors. A coincidence unit flags Si and CZT events if the two components triggered within a coincidence window of 2 micro-seconds. HX-POL’s energy resolution includes contributions of the Si and CZT detectors and will be about 6% in the 100 keV-150 keV energy region where the instrument sees most events. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | The detector assembly resides in a 2 m high collimator assembly that shields the detectors and limits their field of view to a ∼ 8.4◦× 8.4◦ region of the sky. Sources are ob- served by alternating between observations of the source (ON- observations), and observations of a nearby background region (OFF-observations) to enable proper background subtraction. The collimator has a cross section that is larger than the cross section of the detector system, resulting in a “flat” angular response and in relaxed pointing requirements. The graded collimator-shield is made of 4 mm lead. Thin tin and copper layers at the inside of the collimator absorb fluorescent X-ray emission from the lead layer. The Si and CZT detectors reside above an active shield (CsI(Na), 2 cm×25 cm×25 cm) that reduces the atmospheric and internal backgrounds (e.g. events initiated by neutron capture in the CZT detectors) [11]. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Polarization measurements are notorious for systematic ef- fects that may mimic a polarization signal. The HX-POL design copes with systematic errors by rotating the entire collimator-detector assembly around the azimuthal axis at a frequency of about 0.05 Hz. After every 360◦ rotation the sense of rotation is reversed. The rest of the balloon payload does not participate in the rotation. The collimator-detector assembly has a mass of 150 kg, and the minimization of torques that adversely affect the pointing of the collimator are an important design consideration. The azimuthal distribution of Compton scattered events can then be analyzed in celestial coordinates (to extract the polarization of the emission) and in detector coordinates (to estimate systematic errors). Analysis of the ON/OFF data can be used to study effects related to the orientation of the X-ray collimator relative to the vertical axis, and relative to the geomagnetic field. The RHESSI experiment was rotated at 0.25 Hz, a property that was exploited for po- larimetry measurements [12], [13]. The laboratory calibration would include tests with polarized and un-polarized beams of gamma-rays located at different positions relative to the collimator-detector assembly. We have started first tests of a Si+CZT prototype experiment with polarized and un-polarized gamma-rays at Washington University. Both tests use a 5 mCi 137Cs source. A tagged beam of polarized gamma-rays is produced by scattering the gamma-rays off a scintillator with photomultiplier readout. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Photo of a 2×2 array of 2 mm thick silicon strip detectors, each Fig. 3. with a geometric area 100 cm2 and an active area of 81 cm2. On the top side and on the right side, the resistors and capacitors for AC-coupling the crossed strips on the two sides of the detectors can be seen. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Ge detectors. An interacting gamma ray deposits energy in the SCSD which leads to charge induced in the electrodes above and below the interaction. The amount of charge collected is directly proportional to the energy deposited in the detector. The strips on each side of the detector are perpendicular to each other (see the biasing components on two sides of the detector in Fig. 3). The interaction location is defined as the overlapping region where strips from each electrode collected the charge [14]. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | HX-POL uses 2 mm thick Si detectors with an area of 10×10 cm2 contacted with crossed anode and cathode strips with a strip pitch of 1.4 mm. The strips on each side of the detector are surrounded by a guard ring structure. The active area of each Si detector is 81 cm2. The eight layers of 2 mm thick Si detectors absorb vertically incident 50-100 keV X- rays with a probability of ∼50%. Figure 3 shows an array of four 2 mm thick Si detectors. The packaging of the Si- detectors in trays with eight layers of Si detectors is still under development. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 4. The upper panel shows the measured room temperature resolution of the NCI-ASIC as a function of the input capacitance and shaping time. The resolutions refer here to Si. The lower panel shows an 241Am energy spectrum obtained with one 2 mm thick, 10 cm × 10 cm Si detector, identical to the detectors that will be used for HX-POL. The detector was cooled to -40◦ C for this measurement. The 59.5 keV FWHM resolution is 2 keV. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | The 2 mm thick Si detectors achieve excellent energy resolution. The energy resolution of one of the HX-POL Si detectors is shown in Fig. 4. At 59.5 keV the energy resolution is 2 keV. The detector was cooled to -40◦ C for this measurement. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 5. Readout electronics for the CZT array. The photograph shows two CZT detectors (each: 0.5×2×2 cm2, 64 pixels) mounted in a plastic holder. The pixels are at the sides of the detectors facing the readout electronics. Inside the plastic holder reside five boards that hold five ASICs. Four ASICs are used to read out the pixels of the two detectors, and one ASIC is used to read out the cathodes. Several pairs of detectors can be daisy chained to form a “detector chain”. The “harvester board” (visible on the left side) can read out up to five detector chains. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | CZT detectors contacted with 64 pixels at a pitch of 2.5 mm. Our previous results obtained with such detectors are described in [23], [24] and references therein. In the 50 keV- 100 keV energy range, photons loose only a small fraction of their energy in Compton interactions, and the low-energy threshold of the HX-POL experiment is determined by the low-energy threshold of the Si detectors and not by the low- energy threshold of the CZT detectors. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Energy spectrum taken with an 0.5 cm thick CZT detector. The Fig. 6. detector was operated at an energy threshold of 50 keV. The 662 keV energy resolution of the particular pixel shown here is 1.8% FWHM. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | and transmits the data serially to the CPU. Data transfers are serial LVDS (Low Voltage Differential Signaling), transmitting at a rate of 3.125 Mbits/sec. Dead time for an event is 125 microseconds, although we expect that this can be reduced to 70 microseconds. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 6 shows a 137Cs energy spectrum taken with an 0.5 cm thick CZT detector and a prototype version of the ASIC- readout. The detector was operated at an energy threshold of 50 keV. The particular channel achieves an 662 keV energy resolution of 1.8% FWHM. Note that the design of the CZT detector array is still evolving. Right now, the detectors are mounted in plastic holders and make contact through spring loaded pogo pins. An HX-POL flight module would use CZT detectors permanently mounted to ceramic chip carriers. We are furthermore working on reducing the size of the readout electronics. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | We studied the performance of the HX-POL experiment with the GEANT 4 package [25] including the low-energy electromagnetic processes package GLECS [26]. The simu- lations assume a Crab spectrum [27] and a ballon flight at a (zenith-angle averaged) atmospheric depth of 3 gr cm−2. For this depth, the transmissivity of the residual atmosphere increases rapidly from 0 to 0.6 in the 20 keV to 60 keV energy range and increases slowly at higher energies. The code simulates the energy resolution of the Si and CZT detectors and the discretization of the location information introduced by the strip pitch of the Si detectors and the pixel pitch of the CZT detectors. Two data samples were simulated: one with a polarization signature (the “ON-data sample”) and one without (the ”OFF data sample”). The analysis of the simulated data uses only information that will also be available experimentally: the signal amplitudes measured in the Si strips and in the CZT pixels. If more than two interactions are recorded (taking into account the spatial resolution of the detectors), only the two highest energy deposits are used to determine the azimuthal scattering angle. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.