title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Data driven Dirichlet sampling on manifolds | This article presents a novel method to sampling on manifolds based on the
Dirichlet distribution. The proposed strategy allows to completely respect the
underlying manifold around which data is observed, and to do massive samplings
with low computational effort. This can be very helpful, for instance, in
neural networks training process, as well as in uncertainty analysis and
stochastic optimization. Due to its simplicity and efficiency, we believe that
the new method has great potential. Three manifolds (two dimensional ring,
Mobius strip and spider geometry) are considered to test the proposed
methodology, and then it is employed to an engineering application, related to
gas seal coefficients.
| stat |
Geometric adaptive Monte Carlo in random environment | Manifold Markov chain Monte Carlo algorithms have been introduced to sample
more effectively from challenging target densities exhibiting multiple modes or
strong correlations. Such algorithms exploit the local geometry of the
parameter space, thus enabling chains to achieve a faster convergence rate when
measured in number of steps. However, acquiring local geometric information can
often increase computational complexity per step to the extent that sampling
from high-dimensional targets becomes inefficient in terms of total
computational time. This paper analyzes the computational complexity of
manifold Langevin Monte Carlo and proposes a geometric adaptive Monte Carlo
sampler aimed at balancing the benefits of exploiting local geometry with
computational cost to achieve a high effective sample size for a given
computational cost. The suggested sampler is a discrete-time stochastic process
in random environment. The random environment allows to switch between local
geometric and adaptive proposal kernels with the help of a schedule. An
exponential schedule is put forward that enables more frequent use of geometric
information in early transient phases of the chain, while saving computational
time in late stationary phases. The average complexity can be manually set
depending on the need for geometric exploitation posed by the underlying model.
| stat |
Study on upper limit of sample sizes for a two-level test in NIST
SP800-22 | NIST SP800-22 is one of the most widely used statistical testing tools for
pseudorandom number generators (PRNGs). This tool consists of 15 tests
(one-level tests) and two additional tests (two-level tests). Each one-level
test provides one or more $p$-values. The two-level tests measure the
uniformity of the obtained $p$-values for a fixed one-level test. One of the
two-level tests categorizes the $p$-values into ten intervals of equal length,
and apply a chi-squared goodness-of-fit test. This two-level test is often more
powerful than one-level tests, but sometimes it rejects even good PRNGs when
the sample size at the second level is too large, since it detects
approximation errors in the computation of $p$-values. In this paper, we
propose a practical upper limit of the sample size in this two-level test, for
each of six tests appeared in SP800-22. These upper limits are derived by the
chi-squared discrepancy between the distribution of the approximated $p$-values
and the uniform distribution $U(0, 1)$. We also computed a "risky" sample size
at the second level for each one-level test. Our experiments show that the
two-level test with the proposed upper limit gives appropriate results, while
using the risky size often rejects even good PRNGs. We also propose another
improvement: to use the exact probability for the ten categories in the
computation of goodness-of-fit at the two-level test. This allows us to
increase the sample size at the second level, and would make the test more
sensitive than the NIST's recommending usage.
| stat |
Generalization Properties of Optimal Transport GANs with Latent
Distribution Learning | The Generative Adversarial Networks (GAN) framework is a well-established
paradigm for probability matching and realistic sample generation. While recent
attention has been devoted to studying the theoretical properties of such
models, a full theoretical understanding of the main building blocks is still
missing. Focusing on generative models with Optimal Transport metrics as
discriminators, in this work we study how the interplay between the latent
distribution and the complexity of the pushforward map (generator) affects
performance, from both statistical and modelling perspectives. Motivated by our
analysis, we advocate learning the latent distribution as well as the
pushforward map within the GAN paradigm. We prove that this can lead to
significant advantages in terms of sample complexity.
| stat |
Approximate maximum likelihood estimators for linear regression with
design matrix uncertainty | In this paper we consider regression problems subject to arbitrary noise in
the operator or design matrix. This characterization appropriately models many
physical phenomena with uncertainty in the regressors. Although the problem has
been studied extensively for ordinary/total least squares, and via models that
implicitly or explicitly assume Gaussianity, less attention has been paid to
improving estimation for regression problems under general uncertainty in the
design matrix. To address difficulties encountered when dealing with
distributions of sums of random variables, we rely on the saddle point method
to estimate densities and form an approximate log-likelihood to maximize. We
show that the proposed method performs favorably against other classical
methods.
| stat |
Sequential Estimation of Nonparametric Correlation using Hermite Series
Estimators | In this article we describe a new Hermite series based sequential estimator
for the Spearman's rank correlation coefficient and provide algorithms
applicable in both the stationary and non-stationary settings. To treat the
non-stationary setting, we introduce a novel, exponentially weighted estimator
for the Spearman's rank correlation, which allows the local nonparametric
correlation of a bivariate data stream to be tracked. To the best of our
knowledge this is the first algorithm to be proposed for estimating a
time-varying Spearman's rank correlation that does not rely on a moving window
approach. We explore the practical effectiveness of the Hermite series based
estimators through real data and simulation studies demonstrating good
practical performance. The simulation studies in particular reveal competitive
performance compared to an existing algorithm. The potential applications of
this work are manifold. The Hermite series based Spearman's rank correlation
estimator can be applied to fast and robust online calculation of correlation
which may vary over time. Possible machine learning applications include,
amongst others, fast feature selection and hierarchical clustering on massive
data sets.
| stat |
Degenerate Feedback Loops in Recommender Systems | Machine learning is used extensively in recommender systems deployed in
products. The decisions made by these systems can influence user beliefs and
preferences which in turn affect the feedback the learning system receives -
thus creating a feedback loop. This phenomenon can give rise to the so-called
"echo chambers" or "filter bubbles" that have user and societal implications.
In this paper, we provide a novel theoretical analysis that examines both the
role of user dynamics and the behavior of recommender systems, disentangling
the echo chamber from the filter bubble effect. In addition, we offer practical
solutions to slow down system degeneracy. Our study contributes toward
understanding and developing solutions to commonly cited issues in the complex
temporal scenario, an area that is still largely unexplored.
| stat |
Prediction of Compression Index of Fine-Grained Soils Using a Gene
Expression Programming Model | In construction projects, estimation of the settlement of fine-grained soils
is of critical importance, and yet is a challenging task. The coefficient of
consolidation for the compression index (Cc) is a key parameter in modeling the
settlement of fine-grained soil layers. However, the estimation of this
parameter is costly, time-consuming, and requires skilled technicians. To
overcome these drawbacks, we aimed to predict Cc through other soil parameters,
i.e., the liquid limit (LL), plastic limit (PL), and initial void ratio (e0).
Using these parameters is more convenient and requires substantially less time
and cost compared to the conventional tests to estimate Cc. This study presents
a novel prediction model for the Cc of fine-grained soils using gene expression
programming (GEP). A database consisting of 108 different data points was used
to develop the model. A closed-form equation solution was derived to estimate
Cc based on LL, PL, and e0. The performance of the developed GEP-based model
was evaluated through the coefficient of determination (R2), the root mean
squared error (RMSE), and the mean average error (MAE). The proposed model
performed better in terms of R2, RMSE, and MAE compared to the other models.
| stat |
Information Bottleneck Theory on Convolutional Neural Networks | Recent years, many researches attempt to open the black box of deep neural
networks and propose a various of theories to understand it. Among them,
Information Bottleneck (IB) theory claims that there are two distinct phases
consisting of fitting phase and compression phase in the course of training.
This statement attracts many attentions since its success in explaining the
inner behavior of feedforward neural networks. In this paper, we employ IB
theory to understand the dynamic behavior of convolutional neural networks
(CNNs) and investigate how the fundamental features such as convolutional layer
width, kernel size, network depth, pooling layers and multi-fully connected
layer have impact on the performance of CNNs. In particular, through a series
of experimental analysis on benchmark of MNIST and Fashion-MNIST, we
demonstrate that the compression phase is not observed in all these cases. This
shows us the CNNs have a rather complicated behavior than feedforward neural
networks.
| stat |
A Framework for General Conformable Fractional Grey System Models: A
Physical Perspective and its Actual Application | Grey system theory is an important mathematical tool for describing uncertain
information in the real world. It has been used to solve the uncertainty
problems specially caused by lack of information. As a novel theory, the theory
can deal with various fields and plays an important role in modeling the small
sample problems. But many modeling mechanisms of grey system need to be
answered, such as why grey accumulation can be successfully applied to grey
prediction model? What is the key role of grey accumulation? Some scholars have
already given answers to a certain extent. In this paper, we explain the role
from the perspective of complex networks. Further, we propose generalized
conformable accumulation and difference, and clarify its physical meaning in
the grey model. We use our newly proposed fractional accumulation and
difference to our generalized conformable fractional grey model, or GCFGM(1,1),
and employ practical cases to verify that GCFGM(1,1) has higher accuracy
compared to traditional models.
| stat |
Unsupervised inference approach to facial attractiveness | The perception of facial beauty is a complex phenomenon depending on many,
detailed and global facial features influencing each other. In the machine
learning community this problem is typically tackled as a problem of supervised
inference. However, it has been conjectured that this approach does not capture
the complexity of the phenomenon. A recent original experiment
(Ib\'a\~nez-Berganza et al., Scientific Reports 9, 8364, 2019) allowed
different human subjects to navigate the face-space and ``sculpt'' their
preferred modification of a reference facial portrait. Here we present an
unsupervised inference study of the set of sculpted facial vectors in that
experiment. We first infer minimal, interpretable, and faithful probabilistic
models (through Maximum Entropy and artificial neural networks) of the
preferred facial variations, that capture the origin of the observed
inter-subject diversity in the sculpted faces. The application of such
generative models to the supervised classification of the gender of the
sculpting subjects, reveals an astonishingly high prediction accuracy. This
result suggests that much relevant information regarding the subjects may
influence (and be elicited from) her/his facial preference criteria, in
agreement with the multiple motive theory of attractiveness proposed in
previous works.
| stat |
A Bivariate Dead Band Process Adjustment Policy | A bivariate extension to Box and Jenkins (1963) feedback adjustment problem
is presented in this paper. The model balances the fixed cost of making an
adjustment, which is assumed independent of the magnitude of the adjustments,
with the cost of running the process off-target, which is assumed quadratic. It
is also assumed that two controllable factors are available to compensate for
the deviations from target of two responses in the presence of a bivariate
IMA(1,1) disturbance. The optimal policy has the form of a "dead band", in
which adjustments are justified only when the predicted process responses
exceed some boundary in $\mathbb{R}^2$. This boundary indicates when the
responses are predicted to be far enough from their targets that an additional
adjustment or intervention in the process is justified. Although originally
developed to control a machine tool, dead band control policies have
application in other areas. For example, they could be used to control a
disease through the application of a drug to a patient depending on the level
of a substance in the body (e.g., diabetes control). This paper presents
analytical formulae for the computation of the loss function that combines
off-target and adjustment costs per time unit. Expressions are derived for the
average adjustment interval and for the scaled mean square deviations from
target. The minimization of the loss function and the practical use of the
resulting dead band adjustment strategy is illustrated with an application to a
semiconductor manufacturing process.
| stat |
Adaptive Step-Length Selection in Gradient Boosting for Generalized
Additive Models for Location, Scale and Shape | Tuning of model-based boosting algorithms relies mainly on the number of
iterations, while the step-length is fixed at a predefined value. For complex
models with several predictors such as Generalized Additive Models for
Location, Scale and Shape (GAMLSS), imbalanced updates of predictors, where
some distribution parameters are updated more frequently than others, can be a
problem that prevents some submodels to be appropriately fitted within a
limited number of boosting iterations. We propose an approach using adaptive
step-length (ASL) determination within a non-cyclical boosting algorithm for
GAMLSS to prevent such imbalance. Moreover, for the important special case of
the Gaussian distribution, we discuss properties of the ASL and derive a
semi-analytical form of the ASL that avoids manual selection of the search
interval and numerical optimization to find the optimal step-length, and
consequently improves computational efficiency. We show competitive behavior of
the proposed approaches compared to penalized maximum likelihood and boosting
with a fixed step-length for GAMLSS models in two simulations and two
applications, in particular for cases of large variance and/or more variables
than observations. In addition, the idea of the ASL is also applicable to other
models with more than one predictor like zero-inflated count model, and brings
up insights into the choice of the reasonable defaults for the step-length in
simpler special case of (Gaussian) additive models.
| stat |
Restricting the Flow: Information Bottlenecks for Attribution | Attribution methods provide insights into the decision-making of machine
learning models like artificial neural networks. For a given input sample, they
assign a relevance score to each individual input variable, such as the pixels
of an image. In this work we adapt the information bottleneck concept for
attribution. By adding noise to intermediate feature maps we restrict the flow
of information and can quantify (in bits) how much information image regions
provide. We compare our method against ten baselines using three different
metrics on VGG-16 and ResNet-50, and find that our methods outperform all
baselines in five out of six settings. The method's information-theoretic
foundation provides an absolute frame of reference for attribution values
(bits) and a guarantee that regions scored close to zero are not necessary for
the network's decision. For reviews: https://openreview.net/forum?id=S1xWh1rYwB
For code: https://github.com/BioroboticsLab/IBA
| stat |
A Flexible and Adaptive Framework for Abstention Under Class Imbalance | In practical applications of machine learning, it is often desirable to
identify and abstain on examples where the model's predictions are likely to be
incorrect. Much of the prior work on this topic focused on out-of-distribution
detection or performance metrics such as top-k accuracy. Comparatively little
attention was given to metrics such as area-under-the-curve or Cohen's Kappa,
which are extremely relevant for imbalanced datasets. Abstention strategies
aimed at top-k accuracy can produce poor results on these metrics when applied
to imbalanced datasets, even when all examples are in-distribution. We propose
a framework to address this gap. Our framework leverages the insight that
calibrated probability estimates can be used as a proxy for the true class
labels, thereby allowing us to estimate the change in an arbitrary metric if an
example were abstained on. Using this framework, we derive computationally
efficient metric-specific abstention algorithms for optimizing the sensitivity
at a target specificity level, the area under the ROC, and the weighted Cohen's
Kappa. Because our method relies only on calibrated probability estimates, we
further show that by leveraging recent work on domain adaptation under label
shift, we can generalize to test-set distributions that may have a different
class imbalance compared to the training set distribution. On various
experiments involving medical imaging, natural language processing, computer
vision and genomics, we demonstrate the effectiveness of our approach. Source
code available at https://github.com/blindauth/abstention. Colab notebooks
reproducing results available at
https://github.com/blindauth/abstention_experiments.
| stat |
Privacy Risks of Securing Machine Learning Models against Adversarial
Examples | The arms race between attacks and defenses for machine learning models has
come to a forefront in recent years, in both the security community and the
privacy community. However, one big limitation of previous research is that the
security domain and the privacy domain have typically been considered
separately. It is thus unclear whether the defense methods in one domain will
have any unexpected impact on the other domain.
In this paper, we take a step towards resolving this limitation by combining
the two domains. In particular, we measure the success of membership inference
attacks against six state-of-the-art defense methods that mitigate the risk of
adversarial examples (i.e., evasion attacks). Membership inference attacks
determine whether or not an individual data record has been part of a model's
training set. The accuracy of such attacks reflects the information leakage of
training algorithms about individual members of the training set. Adversarial
defense methods against adversarial examples influence the model's decision
boundaries such that model predictions remain unchanged for a small area around
each input. However, this objective is optimized on training data. Thus,
individual data records in the training set have a significant influence on
robust models. This makes the models more vulnerable to inference attacks.
To perform the membership inference attacks, we leverage the existing
inference methods that exploit model predictions. We also propose two new
inference methods that exploit structural properties of robust models on
adversarially perturbed data. Our experimental evaluation demonstrates that
compared with the natural training (undefended) approach, adversarial defense
methods can indeed increase the target model's risk against membership
inference attacks.
| stat |
Infinitely Wide Tensor Networks as Gaussian Process | Gaussian Process is a non-parametric prior which can be understood as a
distribution on the function space intuitively. It is known that by introducing
appropriate prior to the weights of the neural networks, Gaussian Process can
be obtained by taking the infinite-width limit of the Bayesian neural networks
from a Bayesian perspective. In this paper, we explore the infinitely wide
Tensor Networks and show the equivalence of the infinitely wide Tensor Networks
and the Gaussian Process. We study the pure Tensor Network and another two
extended Tensor Network structures: Neural Kernel Tensor Network and Tensor
Network hidden layer Neural Network and prove that each one will converge to
the Gaussian Process as the width of each model goes to infinity. (We note here
that Gaussian Process can also be obtained by taking the infinite limit of at
least one of the bond dimensions $\alpha_{i}$ in the product of tensor nodes,
and the proofs can be done with the same ideas in the proofs of the
infinite-width cases.) We calculate the mean function (mean vector) and the
covariance function (covariance matrix) of the finite dimensional distribution
of the induced Gaussian Process by the infinite-width tensor network with a
general set-up. We study the properties of the covariance function and derive
the approximation of the covariance function when the integral in the
expectation operator is intractable. In the numerical experiments, we implement
the Gaussian Process corresponding to the infinite limit tensor networks and
plot the sample paths of these models. We study the hyperparameters and plot
the sample path families in the induced Gaussian Process by varying the
standard deviations of the prior distributions. As expected, the parameters in
the prior distribution namely the hyper-parameters in the induced Gaussian
Process controls the characteristic lengthscales of the Gaussian Process.
| stat |
Partisan Lean of States: Electoral College and Popular Vote | We compare federal election results for each state versus the USA in every
second year from 1992 to 2018, to model partisan lean of each state and its
dependence on the nationwide popular vote. For each state, we model both its
current partisan lean and its rate of change, as well as sensitivity of state
results with respect to the nationwide popular vote, using Bayesian linear
regression. We apply this to simulate the Electoral College outcome in 2020,
given even (equal) nationwide popular vote, as well as 2016, 2008, and 2004
nationwide popular vote. We backtest 2012 and 2016 elections given actual
popular vote. Taking equal popular vote for two major parties, we prove that
the Electoral College is biased towards Republicans.
| stat |
Cronus: Robust and Heterogeneous Collaborative Learning with Black-Box
Knowledge Transfer | Collaborative (federated) learning enables multiple parties to train a model
without sharing their private data, but through repeated sharing of the
parameters of their local models. Despite its advantages, this approach has
many known privacy and security weaknesses and performance overhead, in
addition to being limited only to models with homogeneous architectures. Shared
parameters leak a significant amount of information about the local (and
supposedly private) datasets. Besides, federated learning is severely
vulnerable to poisoning attacks, where some participants can adversarially
influence the aggregate parameters. Large models, with high dimensional
parameter vectors, are in particular highly susceptible to privacy and security
attacks: curse of dimensionality in federated learning. We argue that sharing
parameters is the most naive way of information exchange in collaborative
learning, as they open all the internal state of the model to inference
attacks, and maximize the model's malleability by stealthy poisoning attacks.
We propose Cronus, a robust collaborative machine learning framework. The
simple yet effective idea behind designing Cronus is to control, unify, and
significantly reduce the dimensions of the exchanged information between
parties, through robust knowledge transfer between their black-box local
models. We evaluate all existing federated learning algorithms against
poisoning attacks, and we show that Cronus is the only secure method, due to
its tight robustness guarantee. Treating local models as black-box, reduces the
information leakage through models, and enables us using existing
privacy-preserving algorithms that mitigate the risk of information leakage
through the model's output (predictions). Cronus also has a significantly lower
sample complexity, compared to federated learning, which does not bind its
security to the number of participants.
| stat |
In Search of Lost (Mixing) Time: Adaptive Markov chain Monte Carlo
schemes for Bayesian variable selection with very large p | The availability of data sets with large numbers of variables is rapidly
increasing. The effective application of Bayesian variable selection methods
for regression with these data sets has proved difficult since available Markov
chain Monte Carlo methods do not perform well in typical problem sizes of
interest. The current paper proposes new adaptive Markov chain Monte Carlo
algorithms to address this shortcoming. The adaptive design of these algorithms
exploits the observation that in large $p$ small $n$ settings, the majority of
the $p$ variables will be approximately uncorrelated a posteriori. The
algorithms adaptively build suitable non-local proposals that result in moves
with squared jumping distance significantly larger than standard methods. Their
performance is studied empirically in high-dimensional problems (with both
simulated and actual data) and speedups of up to 4 orders of magnitude are
observed. The proposed algorithms are easily implementable on multi-core
architectures and are well suited for parallel tempering or sequential Monte
Carlo implementations.
| stat |
Assessing the Performance of Diagnostic Classification Models in Small
Sample Contexts with Different Estimation Methods | Fueled by the call for formative assessments, diagnostic classification
models (DCMs) have recently gained popularity in psychometrics. Despite their
potential for providing diagnostic information that aids in classroom
instruction and students' learning, empirical applications of DCMs to classroom
assessments have been highly limited. This is partly because how DCMs with
different estimation methods perform in small sample contexts is not yet
well-explored. Hence, this study aims to investigate the performance of
respondent classification and item parameter estimation with a comprehensive
simulation design that resembles classroom assessments using different
estimation methods. The key findings are the following: (1) although the marked
difference in respondent classification accuracy was not observed among the
maximum likelihood (ML), Bayesian, and nonparametric methods, the Bayesian
method provided slightly more accurate respondent classification in
parsimonious DCMs than the ML method, and in complex DCMs, the ML method
yielded the slightly better result than the Bayesian method; (2) while item
parameter recovery was poor in both Bayesian and ML methods, the Bayesian
method exhibited unstable slip values owing to the multimodality of their
posteriors under complex DCMs, and the ML method produced irregular estimates
that appear to be well-estimated due to a boundary problem under parsimonious
DCMs.
| stat |
Doubly stochastic models for replicated spatio-temporal point processes | This paper proposes a log-linear model for the latent intensity functions of
a replicated spatio-temporal point process. By simultaneously fitting
correlated spatial and temporal Karhunen-Lo\`eve expansions, the model produces
spatial and temporal components that are usually easy to interpret and capture
the most important modes of variation and spatio-temporal correlation of the
process. The asymptotic distribution of the estimators is derived. The finite
sample properties are studied by simulations. As an example of application, we
analyze bike usage patterns on the Divvy bike sharing system of the city of
Chicago.
| stat |
Pairwise likelihood inference for the multivariate ordered probit model | This paper provides a closed form expression for the pairwise score vector
for the multivariate ordered probit model. This result has several implications
in likelihood-based inference. It is indeed used both to speed-up gradient
based optimization routines for point estimation, and to provide a building
block to compute standard errors and confidence intervals by means of the
Godambe matrix.
| stat |
Methods to Compute Prediction Intervals: A Review and New Results | This paper reviews two main types of prediction interval methods under a
parametric framework. First, we describe methods based on an (approximate)
pivotal quantity. Examples include the plug-in, pivotal, and calibration
methods. Then we describe methods based on a predictive distribution (sometimes
derived based on the likelihood). Examples include Bayesian, fiducial, and
direct-bootstrap methods. Several examples involving continuous distributions
along with simulation studies to evaluate coverage probability properties are
provided. We provide specific connections among different prediction interval
methods for the (log-)location-scale family of distributions. This paper also
discusses general prediction interval methods for discrete data, using the
binomial and Poisson distributions as examples. We also overview methods for
dependent data, with application to time series, spatial data, and Markov
random fields, for example.
| stat |
Universal Prediction Band via Semi-Definite Programming | We propose a computationally efficient method to construct nonparametric,
heteroskedastic prediction bands for uncertainty quantification, with or
without any user-specified predictive model. The data-adaptive prediction band
is universally applicable with minimal distributional assumptions, with strong
non-asymptotic coverage properties, and easy to implement using standard convex
programs. Our approach can be viewed as a novel variance interpolation with
confidence and further leverages techniques from semi-definite programming and
sum-of-squares optimization. Theoretical and numerical performances for the
proposed approach for uncertainty quantification are analyzed.
| stat |
Higher-dimensional spatial extremes via single-site conditioning | Currently available models for spatial extremes suffer either from
inflexibility in the dependence structures that they can capture, lack of
scalability to high dimensions, or in most cases, both of these. We present an
approach to spatial extreme value theory based on the conditional multivariate
extreme value model, whereby the limit theory is formed through conditioning
upon the value at a particular site being extreme. The ensuing methodology
allows for a flexible class of dependence structures, as well as models that
can be fitted in high dimensions. To overcome issues of conditioning on a
single site, we suggest a joint inference scheme based on all observation
locations, and implement an importance sampling algorithm to provide spatial
realizations and estimates of quantities conditioning upon the process being
extreme at any of one of an arbitrary set of locations. The modelling approach
is applied to Australian summer temperature extremes, permitting assessment the
spatial extent of high temperature events over the continent.
| stat |
Graph Neural Networks for Model Recommendation using Time Series Data | Time series prediction aims to predict future values to help stakeholders
make proper strategic decisions. This problem is relevant in all industries and
areas, ranging from financial data to demand to forecast. However, it remains
challenging for practitioners to select the appropriate model to use for
forecasting tasks. With this in mind, we present a model architecture based on
Graph Neural Networks to provide model recommendations for time series
forecasting. We validate our approach on three relevant datasets and compare it
against more than sixteen techniques. Our study shows that the proposed method
performs better than target baselines and state of the art, including
meta-learning. The results show the relevancy and suitability of GNN as methods
for model recommendations in time series forecasting.
| stat |
Monte-Carlo acceleration: importance sampling and hybrid dynamic systems | The reliability of a complex industrial system can rarely be assessed
analytically. As system failure is often a rare event, crude Monte-Carlo
methods are prohibitively expensive from a computational point of view. In
order to reduce computation times, variance reduction methods such as
importance sampling can be used. We propose an adaptation of this method for a
class of multi-component dynamical systems. We address a system whose failure
corresponds to a physical variable of the system (temperature, pressure, water
level) entering a critical region. Such systems are common in hydraulic and
nuclear industry. In these systems, the statuses of the components (on, off, or
out-of-order) determine the dynamics of the physical variables, and is altered
both by deterministic feedback mechanisms and random failures or repairs. In
order to deal with this interplay between components status and physical
variables we model trajectory using piecewise deterministic Markovian processes
(PDMP). We show how to adapt the importance sampling method to PDMP, by
introducing a reference measure on the trajectory space, and we present a
biasing strategy for importance sampling. A simulation study compares our
importance sampling method to the crude Monte-Carlo method for a
three-component-system.
| stat |
Monitoring data in R with the lumberjack package | Monitoring data while it is processed and transformed can yield detailed
insight into the dynamics of a (running) production system. The lumberjack
package is a lightweight package allowing users to follow how an R object is
transformed as it is manipulated by R code. The package abstracts all logging
code from the user, who only needs to specify which objects are logged and what
information should be logged. A few default loggers are included with the
package but the package is extensible through user-defined logger objects.
| stat |
Quantile regression on inactivity time | The inactivity time, or lost lifespan specifically for mortality data,
concerns time from occurrence of an event of interest to the current time point
and has recently emerged as a new summary measure for cumulative information
inherent in time-to-event data. This summary measure provides several benefits
over the traditional methods, including more straightforward interpretation yet
less sensitivity to heavy censoring. However, there exists no systematic
modeling approach to inferring the quantile inactivity time in the literature.
In this paper, we propose a regression method for the quantiles of the
inactivity time distribution under right censoring. The consistency and
asymptotic normality of the regression parameters are established. To avoid
estimation of the probability density function of the inactivity time
distribution under censoring, we propose a computationally efficient method for
estimating the variance-covariance matrix of the regression coefficient
estimates. Simulation results are presented to validate the finite sample
properties of the proposed estimators and test statistics. The proposed method
is illustrated with a real dataset from a clinical trial on breast cancer.
| stat |
Statistical Guarantees and Algorithmic Convergence Issues of Variational
Boosting | We provide statistical guarantees for Bayesian variational boosting by
proposing a novel small bandwidth Gaussian mixture variational family. We
employ a functional version of Frank-Wolfe optimization as our variational
algorithm and study frequentist properties of the iterative boosting updates.
Comparisons are drawn to the recent literature on boosting, describing how the
choice of the variational family and the discrepancy measure affect both
convergence and finite-sample statistical properties of the optimization
routine. Specifically, we first demonstrate stochastic boundedness of the
boosting iterates with respect to the data generating distribution. We next
integrate this within our algorithm to provide an explicit convergence rate,
ending with a result on the required number of boosting updates.
| stat |
Global Convergence of Least Squares EM for Demixing Two Log-Concave
Densities | This work studies the location estimation problem for a mixture of two
rotation invariant log-concave densities. We demonstrate that Least Squares EM,
a variant of the EM algorithm, converges to the true location parameter from a
randomly initialized point. We establish the explicit convergence rates and
sample complexity bounds, revealing their dependence on the signal-to-noise
ratio and the tail property of the log-concave distribution. Moreover, we show
that this global convergence property is robust under model mis-specification.
Our analysis generalizes previous techniques for proving the convergence
results for Gaussian mixtures. In particular, we make use of an
angle-decreasing property for establishing global convergence of Least Squares
EM beyond Gaussian settings, as $\ell_2$ distance contraction no longer holds
globally for general log-concave mixtures.
| stat |
Bayesian regression tree models for causal inference: regularization,
confounding, and heterogeneous effects | This paper presents a novel nonlinear regression model for estimating
heterogeneous treatment effects from observational data, geared specifically
towards situations with small effect sizes, heterogeneous effects, and strong
confounding. Standard nonlinear regression models, which may work quite well
for prediction, have two notable weaknesses when used to estimate heterogeneous
treatment effects. First, they can yield badly biased estimates of treatment
effects when fit to data with strong confounding. The Bayesian causal forest
model presented in this paper avoids this problem by directly incorporating an
estimate of the propensity function in the specification of the response model,
implicitly inducing a covariate-dependent prior on the regression function.
Second, standard approaches to response surface modeling do not provide
adequate control over the strength of regularization over effect heterogeneity.
The Bayesian causal forest model permits treatment effect heterogeneity to be
regularized separately from the prognostic effect of control variables, making
it possible to informatively "shrink to homogeneity". We illustrate these
benefits via the reanalysis of an observational study assessing the causal
effects of smoking on medical expenditures as well as extensive simulation
studies.
| stat |
Adaptive Hyper-box Matching for Interpretable Individualized Treatment
Effect Estimation | We propose a matching method for observational data that matches units with
others in unit-specific, hyper-box-shaped regions of the covariate space. These
regions are large enough that many matches are created for each unit and small
enough that the treatment effect is roughly constant throughout. The regions
are found as either the solution to a mixed integer program, or using a (fast)
approximation algorithm. The result is an interpretable and tailored estimate
of a causal effect for each unit.
| stat |
Low incidence rate of COVID-19 undermines confidence in estimation of
the vaccine efficacy | Knowing the true effect size of clinical interventions in randomised clinical
trials is key to informing the public health policies. Vaccine efficacy is
defined in terms of the relative risk or the ratio of two disease risks.
However, only approximate methods are available for estimating the variance of
the relative risk. In this article, we show using a probabilistic model that
uncertainty in the efficacy rate could be underestimated when the disease risk
is low. Factoring in the baseline rate of the disease, we estimate broader
confidence intervals for the efficacy rates of the vaccines recently developed
for COVID-19. We propose new confidence intervals for the relative risk. We
further show that sample sizes required for phase 3 efficacy trials are
routinely underestimated and propose a new method for sample size calculation
where the efficacy is of interest. We also discuss the deleterious effects of
classification bias which is particularly relevant at low disease prevalence.
| stat |
Weak Convergence and Optimal Tuning of the Reversible Jump Algorithm | The reversible jump algorithm is a useful Markov chain Monte Carlo method
introduced by Green (1995) that allows switches between subspaces of differing
dimensionality, and therefore, model selection. Although this method is now
increasingly used in key areas (e.g. biology and finance), it remains a
challenge to implement it. In this paper, we focus on a simple sampling context
in order to obtain theoretical results that lead to an optimal tuning procedure
for the considered reversible jump algorithm, and consequently, to easy
implementation. The key result is the weak convergence of the sequence of
stochastic processes engendered by the algorithm. It represents the main
contribution of this paper as it is, to our knowledge, the first weak
convergence result for the reversible jump algorithm. The sampler updating the
parameters according to a random walk, this result allows to retrieve the
well-known 0.234 rule for finding the optimal scaling. It also leads to an
answer to the question: "with what probability should a parameter update be
proposed comparatively to a model switch at each iteration?"
| stat |
Probabilistic Modeling of Deep Features for Out-of-Distribution and
Adversarial Detection | We present a principled approach for detecting out-of-distribution (OOD) and
adversarial samples in deep neural networks. Our approach consists in modeling
the outputs of the various layers (deep features) with parametric probability
distributions once training is completed. At inference, the likelihoods of the
deep features w.r.t the previously learnt distributions are calculated and used
to derive uncertainty estimates that can discriminate in-distribution samples
from OOD samples. We explore the use of two classes of multivariate
distributions for modeling the deep features - Gaussian and Gaussian mixture -
and study the trade-off between accuracy and computational complexity. We
demonstrate benefits of our approach on image features by detecting OOD images
and adversarially-generated images, using popular DNN architectures on MNIST
and CIFAR10 datasets. We show that more precise modeling of the feature
distributions result in significantly improved detection of OOD and adversarial
samples; up to 12 percentage points in AUPR and AUROC metrics. We further show
that our approach remains extremely effective when applied to video data and
associated spatio-temporal features by detecting adversarial samples on
activity classification tasks using UCF101 dataset, and the C3D network. To our
knowledge, our methodology is the first one reported for reliably detecting
white-box adversarial framing, a state-of-the-art adversarial attack for video
classifiers.
| stat |
On a projection-based class of uniformity tests on the hypersphere | We propose a projection-based class of uniformity tests on the hypersphere
using statistics that integrate, along all possible directions, the weighted
quadratic discrepancy between the empirical cumulative distribution function of
the projected data and the projected uniform distribution. Simple expressions
for several test statistics are obtained for the circle and sphere, and
relatively tractable forms for higher dimensions. Despite its different origin,
the proposed class is shown to be related with the well-studied Sobolev class
of uniformity tests. Our new class proves itself advantageous by allowing to
derive new tests for hyperspherical data that neatly extend the circular tests
by Watson, Ajne, and Rothman, and by introducing the first instance of an
Anderson-Darling-like test for such data. The asymptotic distributions and the
local optimality against certain alternatives of the new tests are obtained. A
simulation study evaluates the theoretical findings and evidences that, for
certain scenarios, the new tests are competitive against previous proposals.
The new tests are employed in three astronomical applications.
| stat |
Scalable Estimation and Inference with Large-scale or Online Survival
Data | With the rapid development of data collection and aggregation technologies in
many scientific disciplines, it is becoming increasingly ubiquitous to conduct
large-scale or online regression to analyze real-world data and unveil
real-world evidence. In such applications, it is often numerically challenging
or sometimes infeasible to store the entire dataset in memory. Consequently,
classical batch-based estimation methods that involve the entire dataset are
less attractive or no longer applicable. Instead, recursive estimation methods
such as stochastic gradient descent that process data points sequentially are
more appealing, exhibiting both numerical convenience and memory efficiency. In
this paper, for scalable estimation of large or online survival data, we
propose a stochastic gradient descent method which recursively updates the
estimates in an online manner as data points arrive sequentially in streams.
Theoretical results such as asymptotic normality and estimation efficiency are
established to justify its validity. Furthermore, to quantify the uncertainty
associated with the proposed stochastic gradient descent estimator and
facilitate statistical inference, we develop a scalable resampling strategy
that specifically caters to the large-scale or online setting. Simulation
studies and a real data application are also provided to assess its performance
and illustrate its practical utility.
| stat |
Mar-Co: a new dependence structure to model match outcomes in football | The approaches commonly used to model the number of goals in a football match
are characterised by strong assumptions about the dependence between the number
of goals scored by the two competing teams and about their marginal
distribution. In this work, we argue that the assumptions traditionally made
are not always based on solid arguments and sometimes they can be hardly
justified. In light of this, we propose a modification of the Dixon and Coles
(1997) model by relaxing the assumption of Poisson-distributed marginal
variables and by introducing an innovative dependence structure. Specifically,
we define the joint distribution of the number of goals scored during a match
by means of thoroughly chosen marginal (Mar-) and conditional distributions
(-Co). The resulting Mar-Co model is able to balance flexibility and conceptual
simplicity. A real data application involving five European leagues suggests
that the introduction of the novel dependence structure allows to capture and
interpret fundamental league-specific dynamics. In terms of betting
performance, the newly introduced Mar-Co model does not perform worse than the
Dixon and Coles one in a traditional framework (i.e. 1-X-2 bet) and it
outperforms the competing model when a more comprehensive dependence structure
is needed (i.e. Under/Over 2.5 bet).
| stat |
Online tensor factorization and CP-dictionary learning for Markovian
data | Online Tensor Factorization (OTF) is a fundamental tool in learning
low-dimensional interpretable features from streaming multi-modal data. While
various algorithmic and theoretical aspects of OTF have been investigated
recently, general convergence guarantee to stationary points of the objective
function without any incoherence or sparsity assumptions is still lacking even
for the i.i.d. case. In this work, we introduce a novel OTF algorithm that
learns a CANDECOMP/PARAFAC (CP) basis from a given stream of tensor-valued data
under general constraints, including nonnegativity constraints that induce
interpretability of learned CP basis. We prove that our algorithm converges
almost surely to the set of stationary points of the objective function under
the hypothesis that the sequence of data tensors is generated by some
underlying Markov chain. Our setting covers the classical i.i.d. case as well
as a wide range of application contexts including data streams generated by
independent or MCMC sampling. Our result closes a gap between OTF and Online
Matrix Factorization in global convergence analysis. Experimentally, we show
that our OTF algorithm converges much faster than standard algorithms for
nonnegative tensor factorization tasks on both synthetic and real-world data.
Also, we demonstrate the utility of our algorithm on a diverse set of examples
from image, video, and time-series data, illustrating how one may learn
qualitatively different CP-dictionaries from the same tensor data by exploiting
the tensor structure in multiple ways.
| stat |
An outlier-robust model averaging approach by Mallows-type criterion | Model averaging is an alternative to model selection for dealing with model
uncertainty, which is widely used and very valuable. However, most of the
existing model averaging methods are proposed based on the least squares loss
function, which could be very sensitive to the presence of outliers in the
data. In this paper, we propose an outlier-robust model averaging approach by
Mallows-type criterion. The key idea is to develop weight choice criteria by
minimising an estimator of the expected prediction error for the function being
convex with an unique minimum, and twice differentiable in expectation, rather
than the expected squared error. The robust loss functions, such as least
absolute deviation and Huber's function, reduce the effects of large residuals
and poor samples. Simulation study and real data analysis are conducted to
demonstrate the finite-sample performance of our estimators and compare them
with other model selection and averaging methods.
| stat |
Volatility Models Applied to Geophysics and High Frequency Financial
Market Data | This work is devoted to the study of modeling geophysical and financial time
series. A class of volatility models with time-varying parameters is presented
to forecast the volatility of time series in a stationary environment. The
modeling of stationary time series with consistent properties facilitates
prediction with much certainty. Using the GARCH and stochastic volatility
model, we forecast one-step-ahead suggested volatility with +/- 2 standard
prediction errors, which is enacted via Maximum Likelihood Estimation. We
compare the stochastic volatility model relying on the filtering technique as
used in the conditional volatility with the GARCH model. We conclude that the
stochastic volatility is a better forecasting tool than GARCH (1, 1), since it
is less conditioned by autoregressive past information.
| stat |
From one environment to many: The problem of replicability of
statistical inferences | Among plausible causes for replicability failure, one that has not received
sufficient attention is the environment in which the research is conducted.
Consisting of the population, equipment, personnel, and various conditions such
as location, time, and weather, the research environment can affect treatments
and outcomes, and changes in the research environment that occur when an
experiment is redone can affect replicability. We examine the extent to which
such changes contribute to replicability failure. Our framework is that of an
initial experiment that generates the data and a follow-up experiment that is
done the same way except for a change in the research environment. We assume
that the initial experiment satisfies the assumptions of the two-sample
t-statistic and that the follow-up experiment is described by a mixed model
which includes environmental parameters. We derive expressions for the effect
that the research environment has on power, sample size selection, p-values,
and confidence levels. We measure the size of the environmental effect with the
environmental effect ratio EER which is the ratio of the standard deviations of
environment by treatment interaction and error. By varying EER, it is possible
to determine conditions that favor replicability and those that do not.
| stat |
Validating Conditional Density Models and Bayesian Inference Algorithms | Conditional density models f(y|x), where x represents a potentially
high-dimensional feature vector, are an integral part of uncertainty
quantification in prediction and Bayesian inference. However, such models can
be difficult to calibrate. While existing validation techniques can determine
whether an approximated conditional density is compatible overall with a data
sample, they lack practical procedures for identifying, localizing, and
interpreting the nature of (statistically significant) discrepancies over the
entire feature space. In this paper, we present more discerning diagnostics
such as (i) the "Local Coverage Test" (LCT), which is able to distinguish an
arbitrarily misspecified model from the true conditional density of the sample,
and (ii) "Amortized Local P-P plots" (ALP), which can quickly provide
interpretable graphical summaries of distributional differences at any location
x in the feature space. Our validation procedures scale to high dimensions, and
can potentially adapt to any type of data at hand. We demonstrate the
effectiveness of LCT and ALP through a simulated experiment and a realistic
application to parameter inference for galaxy images.
| stat |
Data Selection for Short Term load forecasting | Power load forecast with Machine Learning is a fairly mature application of
artificial intelligence and it is indispensable in operation, control and
planning. Data selection techniqies have been hardly used in this application.
However, the use of such techniques could be beneficial provided the assumption
that the data is identically distributed is clearly not true in load
forecasting, but it is cyclostationary. In this work we present a fully
automatic methodology to determine what are the most adequate data to train a
predictor which is based on a full Bayesian probabilistic model. We assess the
performance of the method with experiments based on real publicly available
data recorded from several years in the United States of America.
| stat |
Graph Spectral Feature Learning for Mixed Data of Categorical and
Numerical Type | Feature learning in the presence of a mixed type of variables, numerical and
categorical types, is an important issue for related modeling problems. For
simple neighborhood queries under mixed data space, standard practice is to
consider numerical and categorical variables separately and combining them
based on some suitable distance functions. Alternatives, such as Kernel
learning or Principal Component do not explicitly consider the inter-dependence
structure among the mixed type of variables. In this work, we propose a novel
strategy to explicitly model the probabilistic dependence structure among the
mixed type of variables by an undirected graph. Spectral decomposition of the
graph Laplacian provides the desired feature transformation. The Eigen spectrum
of the transformed feature space shows increased separability and more
prominent clusterability among the observations. The main novelty of our paper
lies in capturing interactions of the mixed feature type in an unsupervised
framework using a graphical model. We numerically validate the implications of
the feature learning strategy
| stat |
Multivariate Signal Modelling with Applications to Inertial Sensor
Calibration | The common approach to inertial sensor calibration for navigation purposes
has been to model the stochastic error signals of individual sensors
independently, whether as components of a single inertial measurement unit
(IMU) in different directions or arrayed in the same direction for redundancy.
These signals usually have an extremely complex spectral structure that is
often described using latent (or composite) models composed by a sum of
underlying models. A large amount of research in this domain has been focused
on the latter aspect through the proposal of various methods that have been
able to improve the estimation of these models both from a computational and a
statistical point of view. However, the separate calibration of the individual
sensors is still unable to take into account the dependence between each of
them which can have an important impact on the precision of the navigation
systems. In this paper, we develop a new approach to simultaneously model both
the individual signals and the dependence between them by studying the quantity
called Wavelet Cross-Covariance and using it to extend the application of the
Generalized Method of Wavelet Moments. This new method can be used in other
settings for time series modeling, especially in cases where the dependence
among signals may be hard to detect. Moreover, in the field of inertial sensor
calibration, this approach can deliver important contributions among which the
possibility to test dependence between sensors, integrate their dependence
within the navigation filter and construct an optimal virtual sensor that can
be used to simplify and improve navigation accuracy. The advantages of this
method and its usefulness for inertial sensor calibration are highlighted
through a simulation study and an applied example with a small array of XSens
MTi-G IMUs.
| stat |
Tight Regret Bounds for Infinite-armed Linear Contextual Bandits | Linear contextual bandit is an important class of sequential decision making
problems with a wide range of applications to recommender systems, online
advertising, healthcare, and many other machine learning related tasks. While
there is a lot of prior research, tight regret bounds of linear contextual
bandit with infinite action sets remain open. In this paper, we address this
open problem by considering the linear contextual bandit with (changing)
infinite action sets. We prove a regret upper bound on the order of
$O(\sqrt{d^2T\log T})\times \text{poly}(\log\log T)$ where $d$ is the domain
dimension and $T$ is the time horizon. Our upper bound matches the previous
lower bound of $\Omega(\sqrt{d^2 T\log T})$ in [Li et al., 2019] up to iterated
logarithmic terms.
| stat |
Normalizing Flows on Tori and Spheres | Normalizing flows are a powerful tool for building expressive distributions
in high dimensions. So far, most of the literature has concentrated on learning
flows on Euclidean spaces. Some problems however, such as those involving
angles, are defined on spaces with more complex geometries, such as tori or
spheres. In this paper, we propose and compare expressive and numerically
stable flows on such spaces. Our flows are built recursively on the dimension
of the space, starting from flows on circles, closed intervals or spheres.
| stat |
Bayesian semiparametric analysis of multivariate continuous responses,
with variable selection | This article presents an approach to Bayesian semiparametric inference for
Gaussian multivariate response regression. We are motivated by various small
and medium dimensional problems from the physical and social sciences. The
statistical challenges revolve around dealing with the unknown mean and
variance functions and in particular, the correlation matrix. To tackle these
problems, we have developed priors over the smooth functions and a Markov chain
Monte Carlo algorithm for inference and model selection. Specifically,
Dirichlet process mixtures of Gaussian distributions are used as the basis for
a cluster-inducing prior over the elements of the correlation matrix. The
smooth, multidimensional means and variances are represented using radial basis
function expansions. The complexity of the model, in terms of variable
selection and smoothness, is then controlled by spike-slab priors. A simulation
study is presented, demonstrating performance as the response dimension
increases. Finally, the model is fit to a number of real world datasets. An R
package, scripts for replicating synthetic and real data examples, and a
detailed description of the MCMC sampler are available in the supplementary
materials online.
| stat |
Adaptive Designs for Optimal Observed Fisher Information | Expected Fisher information can be found a priori and as a result its inverse
is the primary variance approximation used in the design of experiments. This
is in contrast to the common claim that the inverse of observed Fisher
information is a better approximation to the variance of the maximum likelihood
estimator. Observed Fisher information cannot be known a priori; however, if an
experiment is conducted sequentially (in a series of runs) the observed Fisher
information from previous runs is available. In the current work two adaptive
designs are proposed that use the observed Fisher information from previous
runs in the design of the current run.
| stat |
An efficient estimator of the parameters of the Generalized Lambda
Distribution | Estimation of the four generalized lambda distribution parameters is not
straightforward, and available estimators that perform best have large
computation times. In this paper, we introduce a simple two-step estimator of
the parameters that is comparatively very quick to compute and performs well
when compared with other methods. This computational efficiency makes the use
of bootstrapping to obtain interval estimators for the parameters possible.
Simulations are used to assess the performance of the new estimators and
applications to several data sets are included.
| stat |
Linear Convergence of the Subspace Constrained Mean Shift Algorithm:
From Euclidean to Directional Data | This paper studies linear convergence of the subspace constrained mean shift
(SCMS) algorithm, a well-known algorithm for identifying a density ridge
defined by a kernel density estimator. By arguing that the SCMS algorithm is a
special variant of a subspace constrained gradient ascent (SCGA) algorithm with
an adaptive step size, we derive linear convergence of such SCGA algorithm.
While the existing research focuses mainly on density ridges in the Euclidean
space, we generalize density ridges and the SCMS algorithm to directional data.
In particular, we establish the stability theorem of density ridges with
directional data and prove the linear convergence of our proposed directional
SCMS algorithm.
| stat |
Bayesian Computation with Intractable Likelihoods | This article surveys computational methods for posterior inference with
intractable likelihoods, that is where the likelihood function is unavailable
in closed form, or where evaluation of the likelihood is infeasible. We review
recent developments in pseudo-marginal methods, approximate Bayesian
computation (ABC), the exchange algorithm, thermodynamic integration, and
composite likelihood, paying particular attention to advancements in
scalability for large datasets. We also mention R and MATLAB source code for
implementations of these algorithms, where they are available.
| stat |
Statistical Robust Chinese Remainder Theorem for Multiple Numbers | Generalized Chinese Remainder Theorem (CRT) is a well-known approach to solve
ambiguity resolution related problems. In this paper, we study the robust CRT
reconstruction for multiple numbers from a view of statistics. To the best of
our knowledge, it is the first rigorous analysis on the underlying statistical
model of CRT-based multiple parameter estimation. To address the problem, two
novel approaches are established. One is to directly calculate a conditional
maximum a posteriori probability (MAP) estimation of the residue clustering,
and the other is based on a generalized wrapped Gaussian mixture model to
iteratively search for MAP of both estimands and clustering. Residue error
correcting codes are introduced to improve the robustness further. Experimental
results show that the statistical schemes achieve much stronger robustness
compared to state-of-the-art deterministic schemes, especially in heavy-noise
scenarios.
| stat |
Adaptive exponential power distribution with moving estimator for
nonstationary time series | While standard estimation assumes that all datapoints are from probability
distribution of the same fixed parameters $\theta$, we will focus on maximum
likelihood (ML) adaptive estimation for nonstationary time series: separately
estimating parameters $\theta_T$ for each time $T$ based on the earlier values
$(x_t)_{t<T}$ using (exponential) moving ML estimator $\theta_T=\arg\max_\theta
l_T$ for $l_T=\sum_{t<T} \eta^{T-t} \ln(\rho_\theta (x_t))$ and some
$\eta\in(0,1]$. Computational cost of such moving estimator is generally much
higher as we need to optimize log-likelihood multiple times, however, in many
cases it can be made inexpensive thanks to dependencies. We focus on such
example: $\rho(x)\propto \exp(-|(x-\mu)/\sigma|^\kappa/\kappa)$ exponential
power distribution (EPD) family, which covers wide range of tail behavior like
Gaussian ($\kappa=2$) or Laplace ($\kappa=1$) distribution. It is also
convenient for such adaptive estimation of scale parameter $\sigma$ as its
standard ML estimation is $\sigma^\kappa$ being average $\|x-\mu\|^\kappa$. By
just replacing average with exponential moving average:
$(\sigma_{T+1})^\kappa=\eta(\sigma_T)^\kappa +(1-\eta)|x_T-\mu|^\kappa$ we can
inexpensively make it adaptive. It is tested on daily log-return series for
DJIA companies, leading to essentially better log-likelihoods than standard
(static) estimation, with optimal $\kappa$ tails types varying between
companies. Presented general alternative estimation philosophy provides tools
which might be useful for building better models for analysis of nonstationary
time-series.
| stat |
Inter-frequency radio signal quality prediction for handover, evaluated
in 3GPP LTE | Radio resource management in cellular networks is typically based on device
measurements reported to the serving base station. Frequent measuring of signal
quality on available frequencies would allow for highly reliable networks and
optimal connection at all times. However, these measurements are associated
with costs, such as dedicated device time for performing measurements when the
device will be unavailable for communication. To reduce the costs, we consider
predictions of inter-frequency radio quality measurements that are useful to
assess potential inter-frequency handover decisions. In this contribution, we
have considered measurements from a live 3GPP LTE network. We demonstrate that
straightforward applications of the most commonly used machine learning models
are unable to provide high accuracy predictions. Instead, we propose a novel
approach with a duo-threshold for high accuracy decision recommendations. Our
approach leads to class specific prediction accuracies as high as 92% and 95%,
still drastically reducing the need for inter-frequency measurements.
| stat |
Statistical testing in a Linear Probability Space | Imagine that you could calculate of posttest probabilities, i.e. Bayes
theorem with simple addition. This is possible if we stop thinking of
probabilities as ranging from 0 to 1.0. There is a naturally occurring linear
probability space when data are transformed into the logarithm of the odds
ratio (log10 odds). In this space, probabilities are replaced by W (Weight)
where W=log10(probability/(1-probability)). I would like to argue the multiple
benefits of performing statistical testing in a linear probability space: 1)
Statistical testing is accurate in linear probability space but not in other
spaces. 2) Effect size is called Impact (I) and is the difference in means
between two treatments (I=Wmean2-Wmean1). 3) Bayes theorem is simply
Wposttest=Wpretest+Itest. 4) Significance (p value) is replaced by Certainty
(C) which is the W of the p value. Methods to transform data into and out of
linear probability space are described.
| stat |
Trajectory growth lower bounds for random sparse deep ReLU networks | This paper considers the growth in the length of one-dimensional trajectories
as they are passed through deep ReLU neural networks, which, among other
things, is one measure of the expressivity of deep networks. We generalise
existing results, providing an alternative, simpler method for lower bounding
expected trajectory growth through random networks, for a more general class of
weights distributions, including sparsely connected networks. We illustrate
this approach by deriving bounds for sparse-Gaussian, sparse-uniform, and
sparse-discrete-valued random nets. We prove that trajectory growth can remain
exponential in depth with these new distributions, including their sparse
variants, with the sparsity parameter appearing in the base of the exponent.
| stat |
Graphical tests of independence for general distributions | We propose two model-free, permutation-based tests of independence between a
pair of random variables. The tests can be applied to samples from any
bivariate distribution: continuous, discrete or mixture of those, with light
tails or heavy tails, \ldots The tests take advantage of the recent development
of the global envelope tests in the context of spatial statistics. Apart from
the broad applicability of the tests, their main benefit lies in the graphical
interpretation of the test outcome: in case of rejection of the null hypothesis
of independence, the combinations of quantiles in the two marginals are
indicated for which the deviation from independence is significant. This
information can be used to gain more insight into the properties of the
observed data and as a guidance for proposing more complicated models and
hypotheses. We assess the performance of the proposed tests in a simulation
study and compare them to several well-established tests of independence.
Furthermore, we illustrate the use of the tests and the interpretation of the
test outcome in two real datasets consisting of meteorological reports (daily
mean temperature and total daily precipitation, having an atomic component at 0
millimeters) and road accidents reports (type of road and the weather
conditions, both variables having categorical distribution).
| stat |
Distribution-Based Invariant Deep Networks for Learning Meta-Features | Recent advances in deep learning from probability distributions successfully
achieve classification or regression from distribution samples, thus invariant
under permutation of the samples. The first contribution of the paper is to
extend these neural architectures to achieve invariance under permutation of
the features, too. The proposed architecture, called Dida, inherits the NN
properties of universal approximation, and its robustness w.r.t.
Lipschitz-bounded transformations of the input distribution is established. The
second contribution is to empirically and comparatively demonstrate the merits
of the approach on two tasks defined at the dataset level. On both tasks, Dida
learns meta-features supporting the characterization of a (labelled) dataset.
The first task consists of predicting whether two dataset patches are extracted
from the same initial dataset. The second task consists of predicting whether
the learning performance achieved by a hyper-parameter configuration under a
fixed algorithm (ranging in k-NN, SVM, logistic regression and linear
classifier with SGD) dominates that of another configuration, for a dataset
extracted from the OpenML benchmarking suite. On both tasks, Dida outperforms
the state of the art: DSS (Maron et al., 2020) and Dataset2Vec (Jomaa et al.,
2019) architectures, as well as the models based on the hand-crafted
meta-features of the literature.
| stat |
Epidemiological Forecasting with Model Reduction of Compartmental
Models. Application to the COVID-19 pandemic | We propose a forecasting method for predicting epidemiological health series
on a two-week horizon at the regional and interregional resolution. The
approach is based on model order reduction of parametric compartmental models,
and is designed to accommodate small amount of sanitary data. The efficiency of
the method is shown in the case of the prediction of the number of infected and
removed people during the two pandemic waves of COVID-19 in France, which have
taken place approximately between February and November 2020. Numerical results
illustrate the promising potential of the approach.
| stat |
Evaluating Ex Ante Counterfactual Predictions Using Ex Post Causal
Inference | We derive a formal, decision-based method for comparing the performance of
counterfactual treatment regime predictions using the results of experiments
that give relevant information on the distribution of treated outcomes. Our
approach allows us to quantify and assess the statistical significance of
differential performance for optimal treatment regimes estimated from
structural models, extrapolated treatment effects, expert opinion, and other
methods. We apply our method to evaluate optimal treatment regimes for
conditional cash transfer programs across countries where predictions are
generated using data from experimental evaluations in other countries and
pre-program data in the country of interest.
| stat |
Inference for stochastic kinetic models from multiple data sources for
joint estimation of infection dynamics from aggregate reports and virological
data | Influenza and respiratory syncytial virus (RSV) are the leading etiological
agents of seasonal acute respiratory infections (ARI) around the world. Medical
doctors typically base the diagnosis of ARI on patients' symptoms alone and do
not always conduct virological tests necessary to identify individual viruses,
which limits the ability to study the interaction between multiple pathogens
and make public health recommendations. We consider a stochastic kinetic model
(SKM) for two interacting ARI pathogens circulating in a large population and
an empirically motivated background process for infections with other pathogens
causing similar symptoms. An extended marginal sampling approach based on the
Linear Noise Approximation to the SKM integrates multiple data sources and
additional model components. We infer the parameters defining the pathogens'
dynamics and interaction within a Bayesian hierarchical model and explore the
posterior trajectories of infections for each illness based on aggregate
infection reports from six epidemic seasons collected by the state health
department, and a subset of virological tests from a sentinel program at a
general hospital in San Luis Potos\'i, M\'exico. We interpret the results based
on real and simulated data and make recommendations for future data collection
strategies. Supplementary materials and software are provided online.
| stat |
A comparison of statistical and machine learning methods for creating
national daily maps of ambient PM$_{2.5}$ concentration | A typical problem in air pollution epidemiology is exposure assessment for
individuals for which health data are available. Due to the sparsity of
monitoring sites and the limited temporal frequency with which measurements of
air pollutants concentrations are collected (for most pollutants, once every 3
or 6 days), epidemiologists have been moving away from characterizing ambient
air pollution exposure solely using measurements. In the last few years,
substantial research efforts have been placed in developing statistical methods
or machine learning techniques to generate estimates of air pollution at finer
spatial and temporal scales (daily, usually) with complete coverage. Some of
these methods include: geostatistical techniques, such as kriging; spatial
statistical models that use the information contained in air quality model
outputs (statistical downscaling models); linear regression modeling approaches
that leverage the information in GIS covariates (land use regression); or
machine learning methods that mine the information contained in relevant
variables (neural network and deep learning approaches). Although some of these
exposure modeling approaches have been used in several air pollution
epidemiological studies, it is not clear how much the predicted exposures
generated by these methods differ, and which method generates more reliable
estimates. In this paper, we aim to address this gap by evaluating a variety of
exposure modeling approaches, comparing their predictive performance and
computational difficulty. Using PM$_{2.5}$ in year 2011 over the continental
U.S. as case study, we examine the methods' performances across seasons, rural
vs urban settings, and levels of PM$_{2.5}$ concentrations (low, medium, high).
| stat |
Assessment of Heterogeneous Treatment Effect Estimation Accuracy via
Matching | We study the assessment of the accuracy of heterogeneous treatment effect
(HTE) estimation, where the HTE is not directly observable so standard
computation of prediction errors is not applicable. To tackle the difficulty,
we propose an assessment approach by constructing pseudo-observations of the
HTE based on matching. Our contributions are three-fold: first, we introduce a
novel matching distance derived from proximity scores in random forests;
second, we formulate the matching problem as an average minimum-cost flow
problem and provide an efficient algorithm; third, we propose a
match-then-split principle for the assessment with cross-validation. We
demonstrate the efficacy of the assessment approach on synthetic data and data
generated from a real dataset.
| stat |
Home advantage of European major football leagues under COVID-19
pandemic | Since March 2020, the environment surrounding football has changed
dramatically because of the COVID-19 pandemic. After a few months' break,
re-scheduled matches were held behind closed doors without spectators. The main
objective of this study is a quantitative evaluation of ``crowd effects'' on
home advantage, using the results of these closed matches. The proposed
analysis uses pairwise comparison method to reduce the effects caused by the
unbalanced schedule. The following conclusions were drawn from the statistical
hypothesis tests conducted in this study: In four major European leagues, the
home advantage is reduced in closed matches compared to than in the normal
situation, i.e., with spectators. The reduction amounts among leagues were
different. For example, in Germany, the home advantage was negative during the
closed-match period. On the other hand,in England,statistically significant
differences in home advantage were not observed between closed matches and
normal situation.
| stat |
Non-proportional hazards in immuno-oncology: is an old perspective
needed? | A fundamental concept in two-arm non-parametric survival analysis is the
comparison of observed versus expected numbers of events on one of the
treatment arms (the choice of which arm is arbitrary), where the expectation is
taken assuming that the true survival curves in the two arms are identical.
This concept is at the heart of the counting-process theory that provides a
rigorous basis for methods such as the log-rank test. It is natural, therefore,
to maintain this perspective when extending the log-rank test to deal with
non-proportional hazards, for example by considering a weighted sum of the
"observed - expected" terms, where larger weights are given to time periods
where the hazard ratio is expected to favour the experimental treatment. In
doing so, however, one may stumble across some rather subtle issues, related to
the difficulty in ascribing a causal interpretation to hazard ratios, that may
lead to strange conclusions. An alternative approach is to view non-parametric
survival comparisons as permutation tests. With this perspective, one can
easily improve on the efficiency of the log-rank test, whilst thoroughly
controlling the false positive rate. In particular, for the field of
immuno-oncology, where researchers often anticipate a delayed treatment effect,
sample sizes could be substantially reduced without loss of power.
| stat |
On the spatial and temporal shift in the archetypal seasonal temperature
cycle as driven by annual and semi-annual harmonics | Statistical methods are required to evaluate and quantify the uncertainty in
environmental processes, such as land and sea surface temperature, in a
changing climate. Typically, annual harmonics are used to characterize the
variation in the seasonal temperature cycle. However, an often overlooked
feature of the climate seasonal cycle is the semi-annual harmonic, which can
account for a significant portion of the variance of the seasonal cycle and
varies in amplitude and phase across space. Together, the spatial variation in
the annual and semi-annual harmonics can play an important role in driving
processes that are tied to seasonality (e.g., ecological and agricultural
processes). We propose a multivariate spatio-temporal model to quantify the
spatial and temporal change in minimum and maximum temperature seasonal cycles
as a function of the annual and semi-annual harmonics. Our approach captures
spatial dependence, temporal dynamics, and multivariate dependence of these
harmonics through spatially and temporally-varying coefficients. We apply the
model to minimum and maximum temperature over North American for the years 1979
to 2018. Formal model inference within the Bayesian paradigm enables the
identification of regions experiencing significant changes in minimum and
maximum temperature seasonal cycles due to the relative effects of changes in
the two harmonics.
| stat |
Identifiability of causal effects with multiple causes and a binary
outcome | Unobserved confounding presents a major threat to causal inference from
observational studies. Recently, several authors suggest that this problem may
be overcome in a shared confounding setting where multiple treatments are
independent given a common latent confounder. It has been shown that under a
linear Gaussian model for the treatments, the causal effect is not identifiable
without parametric assumptions on the outcome model. In this paper, we show
that the causal effect is indeed identifiable if we assume a general binary
choice model for the outcome with a non-probit link. Our identification
approach is based on the incongruence between Gaussianity of the treatments and
latent confounder, and non-Gaussianity of a latent outcome variable. We further
develop a two-step likelihood-based estimation procedure.
| stat |
Bayesian dynamic variable selection in high dimensions | This paper proposes a variational Bayes algorithm for computationally
efficient posterior and predictive inference in time-varying parameter (TVP)
models. Within this context we specify a new dynamic variable/model selection
strategy for TVP dynamic regression models in the presence of a large number of
predictors. This strategy allows for assessing in individual time periods which
predictors are relevant (or not) for forecasting the dependent variable. The
new algorithm is evaluated numerically using synthetic data and its
computational advantages are established. Using macroeconomic data for the US
we find that regression models that combine time-varying parameters with the
information in many predictors have the potential to improve forecasts of price
inflation over a number of alternative forecasting models.
| stat |
Incompletely observed nonparametric factorial designs with repeated
measurements: A wild bootstrap approach | In many life science experiments or medical studies, subjects are repeatedly
observed and measurements are collected in factorial designs with multivariate
data. The analysis of such multivariate data is typically based on multivariate
analysis of variance (MANOVA) or mixed models, requiring complete data, and
certain assumption on the underlying parametric distribution such as continuity
or a specific covariance structure, e.g., compound symmetry. However, these
methods are usually not applicable when discrete data or even ordered
categorical data are present. In such cases, nonparametric rank-based methods
that do not require stringent distributional assumptions are the preferred
choice. However, in the multivariate case, most rank-based approaches have only
been developed for complete observations. It is the aim of this work is to
develop asymptotic correct procedures that are capable of handling missing
values, allowing for singular covariance matrices and are applicable for
ordinal or ordered categorical data. This is achieved by applying a wild
bootstrap procedure in combination with quadratic form-type test statistics.
Beyond proving their asymptotic correctness, extensive simulation studies
validate their applicability for small samples. Finally, two real data examples
are analyzed.
| stat |
direpack: A Python 3 package for state-of-the-art statistical dimension
reduction methods | The direpack package aims to establish a set of modern statistical dimension
reduction techniques into the Python universe as a single, consistent package.
The dimension reduction methods included resort into three categories:
projection pursuit based dimension reduction, sufficient dimension reduction,
and robust M estimators for dimension reduction. As a corollary, regularized
regression estimators based on these reduced dimension spaces are provided as
well, ranging from classical principal component regression up to sparse
partial robust M regression. The package also contains a set of classical and
robust pre-processing utilities, including generalized spatial signs, as well
as dedicated plotting functionality and cross-validation utilities. Finally,
direpack has been written consistent with the scikit-learn API, such that the
estimators can flawlessly be included into (statistical and/or machine)
learning pipelines in that framework.
| stat |
Population-based Global Optimisation Methods for Learning Long-term
Dependencies with RNNs | Despite recent innovations in network architectures and loss functions,
training RNNs to learn long-term dependencies remains difficult due to
challenges with gradient-based optimisation methods. Inspired by the success of
Deep Neuroevolution in reinforcement learning (Such et al. 2017), we explore
the use of gradient-free population-based global optimisation (PBO) techniques
-- training RNNs to capture long-term dependencies in time-series data. Testing
evolution strategies (ES) and particle swarm optimisation (PSO) on an
application in volatility forecasting, we demonstrate that PBO methods lead to
performance improvements in general, with ES exhibiting the most consistent
results across a variety of architectures.
| stat |
Missing Data Imputation using Optimal Transport | Missing data is a crucial issue when applying machine learning algorithms to
real-world datasets. Starting from the simple assumption that two batches
extracted randomly from the same dataset should share the same distribution, we
leverage optimal transport distances to quantify that criterion and turn it
into a loss function to impute missing data values. We propose practical
methods to minimize these losses using end-to-end learning, that can exploit or
not parametric assumptions on the underlying distributions of values. We
evaluate our methods on datasets from the UCI repository, in MCAR, MAR and MNAR
settings. These experiments show that OT-based methods match or out-perform
state-of-the-art imputation methods, even for high percentages of missing
values.
| stat |
Rank Determination in Tensor Factor Model | Factor model is an appealing and effective analytic tool for high-dimensional
time series, with a wide range of applications in economics, finance and
statistics. One of the fundamental issues in using factor model for time series
in practice is the determination of the number of factors to use. This paper
develops two criteria for such a task for tensor factor models where the signal
part of an observed time series in tensor form assumes a Tucker decomposition
with the core tensor as the factor tensor. The task is to determine the
dimensions of the core tensor. One of the proposed criteria is similar to
information based criteria of model selection, and the other is an extension of
the approaches based on the ratios of consecutive eigenvalues often used in
factor analysis for panel time series. The new criteria are designed to locate
the gap between the true smallest non-zero eigenvalue and the zero eigenvalues
of a functional of the population version of the auto-cross-covariances of the
tensor time series using their sample versions. As sample size and tensor
dimension increase, such a gap increases under regularity conditions, resulting
in consistency of the rank estimator. The criteria are built upon the existing
non-iterative and iterative estimation procedures of tensor factor model,
yielding different performances. We provide sufficient conditions and
convergence rate for the consistency of the criteria as the sample size $T$ and
the dimensions of the observed tensor time series go to infinity. The results
include the vector factor models as special cases, with an additional
convergence rates. The results also include the cases when there exist factors
with different signal strength. In addition, the convergence rates of the
eigenvalue estimators are established. Simulation studies provide promising
finite sample performance for the two criteria.
| stat |
Challenges in Markov chain Monte Carlo for Bayesian neural networks | Markov chain Monte Carlo (MCMC) methods have not been broadly adopted in
Bayesian neural networks (BNNs). This paper initially reviews the main
challenges in sampling from the parameter posterior of a neural network via
MCMC. Such challenges culminate to lack of convergence to the parameter
posterior. Nevertheless, this paper shows that a non-converged Markov chain,
generated via MCMC sampling from the parameter space of a neural network, can
yield via Bayesian marginalization a valuable predictive posterior of the
output of the neural network. Classification examples based on multilayer
perceptrons showcase highly accurate predictive posteriors. The postulate of
limited scope for MCMC developments in BNNs is partially valid; an
asymptotically exact parameter posterior seems less plausible, yet an accurate
predictive posterior is a tenable research avenue.
| stat |
Efficient MCMC Sampling with Dimension-Free Convergence Rate using
ADMM-type Splitting | Performing exact Bayesian inference for complex models is computationally
intractable. Markov chain Monte Carlo (MCMC) algorithms can provide reliable
approximations of the posterior distribution but are expensive for large
datasets and high-dimensional models. A standard approach to mitigate this
complexity consists in using subsampling techniques or distributing the data
across a cluster. However, these approaches are typically unreliable in
high-dimensional scenarios. We focus here on a recent alternative class of MCMC
schemes exploiting a splitting strategy akin to the one used by the celebrated
ADMM optimization algorithm. These methods appear to provide empirically
state-of-the-art performance but their theoretical behavior in high dimension
is currently unknown. In this paper, we propose a detailed theoretical study of
one of these algorithms known as the split Gibbs sampler. Under regularity
conditions, we establish explicit convergence rates for this scheme using Ricci
curvature and coupling ideas. We support our theory with numerical
illustrations.
| stat |
Grabit: Gradient Tree-Boosted Tobit Models for Default Prediction | A frequent problem in binary classification is class imbalance between a
minority and a majority class such as defaults and non-defaults in default
prediction. In this article, we introduce a novel binary classification model,
the Grabit model, which is obtained by applying gradient tree boosting to the
Tobit model. We show how this model can leverage auxiliary data to obtain
increased predictive accuracy for imbalanced data. We apply the Grabit model to
predicting defaults on loans made to Swiss small and medium-sized enterprises
(SME) and obtain a large and significant improvement in predictive performance
compared to other state-of-the-art approaches.
| stat |
Point-process based Bayesian modeling of space-time structures of forest
fire occurrences in Mediterranean France | Due to climate change and human activity, wildfires are expected to become
more frequent and extreme worldwide, causing economic and ecological disasters.
The deployment of preventive measures and operational forecasts can be aided by
stochastic modeling that helps to understand and quantify the mechanisms
governing the occurrence intensity. We here develop a point process framework
for wildfire ignition points observed in the French Mediterranean basin since
1995, and we fit a spatio-temporal log-Gaussian Cox process with monthly
temporal resolution in a Bayesian framework using the integrated nested Laplace
approximation (INLA). Human activity is the main direct cause of wildfires and
is indirectly measured through a number of appropriately defined proxies
related to land-use covariates (urbanization, road network) in our approach,
and we further integrate covariates of climatic and environmental conditions to
explain wildfire occurrences. We include spatial random effects with Mat\'ern
covariance and temporal autoregression at yearly resolution. Two major
methodological challenges are tackled: first, handling and unifying multi-scale
structures in data is achieved through computer-intensive preprocessing steps
with GIS software and kriging techniques; second, INLA-based estimation with
high-dimensional response vectors and latent models is facilitated through
intra-year subsampling, taking into account the occurrence structure of
wildfires.
| stat |
Group Inference in High Dimensions with Applications to Hierarchical
Testing | High-dimensional group inference is an essential part of statistical methods
for analysing complex data sets, including hierarchical testing, tests of
interaction, detection of heterogeneous treatment effects and inference for
local heritability. Group inference in regression models can be measured with
respect to a weighted quadratic functional of the regression sub-vector
corresponding to the group. Asymptotically unbiased estimators of these
weighted quadratic functionals are constructed and a novel procedure using
these estimators for inference is proposed. We derive its asymptotic Gaussian
distribution which enables the construction of asymptotically valid confidence
intervals and tests which perform well in terms of length or power. The
proposed test is computationally efficient even for a large group,
statistically valid for any group size and achieving good power performance for
testing large groups with many small regression coefficients. We apply the
methodology to several interesting statistical problems and demonstrate its
strength and usefulness on simulated and real data.
| stat |
Fast Counting in Machine Learning Applications | We propose scalable methods to execute counting queries in machine learning
applications. To achieve memory and computational efficiency, we abstract
counting queries and their context such that the counts can be aggregated as a
stream. We demonstrate performance and scalability of the resulting approach on
random queries, and through extensive experimentation using Bayesian networks
learning and association rule mining. Our methods significantly outperform
commonly used ADtrees and hash tables, and are practical alternatives for
processing large-scale data.
| stat |
Application of the interacting particle system method to piecewise
deterministic Markov processes used in reliability | Variance reduction methods are often needed for the reliability assessment of
complex industrial systems, we focus on one variance reduction method in a
given context, that is the interacting particle system method (IPS) used on
piecewise deterministic Markov processes (PDMP) for reliability assessment .
The PDMPs are a very large class of processes which benefit from high modeling
capacities, they can model almost any Markovian phenomenon that does not
include diffusion. In reliability assessment, the PDMPs modeling industrial
systems generally involve low jump rates and jump kernels favoring one safe
arrival, we call such model a "concentrated PDMP".
Used on such concentrated PDMPs, the IPS is inefficient and does not always
provide a variance reduction. Indeed, the efficiency of the IPS method relies
on simulating many different trajectories during its propagation steps, but
unfortunately concentrated PDMPs are likely to generate the same deterministic
trajectories over and over. We propose an adaptation of the IPS method called
IPS+M that reduces this phenomenon. The IPS+M consists in modifying the
propagation steps of the IPS, by conditioning the propagation to avoid
generating the same trajectories multiple times. We prove that, compared to the
IPS, the IPS+M method always provides an estimator with a lower variance. We
also carry out a quick simulation study on a two-components system that
confirms this result.
| stat |
Limitless Regression Discontinuity | Conventionally, regression discontinuity analysis contrasts a univariate
regression's limits as its independent variable, $R$, approaches a cut-point,
$c$, from either side. Alternative methods target the average treatment effect
in a small region around $c$, at the cost of an assumption that treatment
assignment, $\mathcal{I}\left[R<c\right]$, is ignorable vis a vis potential
outcomes.
Instead, the method presented in this paper assumes Residual Ignorability,
ignorability of treatment assignment vis a vis detrended potential outcomes.
Detrending is effected not with ordinary least squares but with MM-estimation,
following a distinct phase of sample decontamination. The method's inferences
acknowledge uncertainty in both of these adjustments, despite its applicability
whether $R$ is discrete or continuous; it is uniquely robust to leading
validity threats facing regression discontinuity designs.
| stat |
Discussion of "Functional Models for Time-Varying Random Objects'' by
Dubey and M\"uller | The discussion focuses on metric covariance, a new association measure
between paired random objects in a metric space, developed by Dubey and
M\"uller, and on its relationship with other similar concepts which have
previously appeared in the literature, including distance covariance by
Sz\'ekely et al, as well as its generalisations which rely on the formalism of
reproducing kernel Hilbert spaces (RKHS).
| stat |
Statistical inference for individual fairness | As we rely on machine learning (ML) models to make more consequential
decisions, the issue of ML models perpetuating or even exacerbating undesirable
historical biases (e.g., gender and racial biases) has come to the fore of the
public's attention. In this paper, we focus on the problem of detecting
violations of individual fairness in ML models. We formalize the problem as
measuring the susceptibility of ML models against a form of adversarial attack
and develop a suite of inference tools for the adversarial cost function. The
tools allow auditors to assess the individual fairness of ML models in a
statistically-principled way: form confidence intervals for the worst-case
performance differential between similar individuals and test hypotheses of
model fairness with (asymptotic) non-coverage/Type I error rate control. We
demonstrate the utility of our tools in a real-world case study.
| stat |
Algorithms for Estimating Trends in Global Temperature Volatility | Trends in terrestrial temperature variability are perhaps more relevant for
species viability than trends in mean temperature. In this paper, we develop
methodology for estimating such trends using multi-resolution climate data from
polar orbiting weather satellites. We derive two novel algorithms for
computation that are tailored for dense, gridded observations over both space
and time. We evaluate our methods with a simulation that mimics these data's
features and on a large, publicly available, global temperature dataset with
the eventual goal of tracking trends in cloud reflectance temperature
variability.
| stat |
Relations between networks, regression, partial correlation, and latent
variable model | The Gaussian graphical model (GGM) has become a popular tool for analyzing
networks of psychological variables. In a recent paper in this journal, Forbes,
Wright, Markon, and Krueger (FWMK) voiced the concern that GGMs that are
estimated from partial correlations wrongfully remove the variance that is
shared by its constituents. If true, this concern has grave consequences for
the application of GGMs. Indeed, if partial correlations only capture the
unique covariances, then the data that come from a unidimensional latent
variable model ULVM should be associated with an empty network (no edges), as
there are no unique covariances in a ULVM. We know that this cannot be true,
which suggests that FWMK are missing something with their claim. We introduce a
connection between the ULVM and the GGM and use that connection to prove that
we find a fully-connected and not an empty network associated with a ULVM. We
then use the relation between GGMs and linear regression to show that the
partial correlation indeed does not remove the common variance.
| stat |
Block models for multipartite networks.Applications in ecology and
ethnobiology | Modeling relations between individuals is a classical question in social
sciences, ecology, etc. In order to uncover a latent structure in the data, a
popular approach consists in clustering individuals according to the observed
patterns of interactions. To do so, Stochastic block models (SBM) and Latent
Block models (LBM) are standard tools for clustering the individuals with
respect to their comportment in a unique network. However, when adopting an
integrative point of view, individuals are not involved in a unique network but
are part of several networks, resulting into a potentially complex multipartite
network. In this contribution, we propose a stochastic block model able to
handle multipartite networks, thus supplying a clustering of the individuals
based on their connection behavior in more than one network. Our model is an
extension of the latent block models (LBM) and stochastic block model (SBM).
The parameters -- such as the marginal probabilities of assignment to blocks
and the matrix of probabilities of connections between blocks -- are estimated
through a variational Expectation-Maximization procedure. The numbers of blocks
are chosen with the Integrated Completed Likelihood criterion, a penalized
likelihood criterion. The pertinence of our methodology is illustrated on two
datasets issued from ecology and ethnobiology.
| stat |
A Note on the Estimation Method of Intervention Effects based on
Statistical Decision Theory | In this paper, we deal with the problem of estimating the intervention effect
in the statistical causal analysis using the structural equation model and the
causal diagram. The intervention effect is defined as a causal effect on the
response variable $Y$ when the causal variable $X$ is fixed to a certain value
by an external operation and is defined based on the causal diagram. The
intervention effect is defined as a function of the probability distributions
in the causal diagram, however, generally these probability distributions are
unknown, so it is required to estimate them from data. In other words, the
steps of the estimation of the intervention effect using the causal diagram are
as follows: 1. Estimate the causal diagram from the data, 2. Estimate the
probability distributions in the causal diagram from the data, 3. Calculate the
intervention effect. However, if the problem of estimating the intervention
effect is formulated in the statistical decision theory framework, estimation
with this procedure is not necessarily optimal. In this study, we formulate the
problem of estimating the intervention effect for the two cases, the case where
the causal diagram is known and the case where it is unknown, in the framework
of statistical decision theory and derive the optimal decision method under the
Bayesian criterion. We show the effectiveness of the proposed method through
numerical simulations.
| stat |
Prediction and model-assisted estimation of diameter distributions using
Norwegian national forest inventory and airborne laser scanning data | Diameter at breast height (DBH) distributions offer valuable information for
operational and strategic forest management decisions. We predicted DBH
distributions using Norwegian national forest inventory and airborne laser
scanning data and compared the predictive performances of linear mixed-effects
(PPM), generalized linear-mixed (GLM) and k nearest neighbor (NN) models. While
GLM resulted in smaller prediction errors than PPM, both were clearly
outperformed by NN. We therefore studied the ability of the NN model to improve
the precision of stem frequency estimates by DBH classes in the 8.7 Mha study
area using a model-assisted (MA) estimator suitable for systematic sampling. MA
estimates yielded greater than or approximately equal efficiencies as direct
estimates using field data only. The relative efficiencies (REs) associated
with the MA estimates ranged between 0.95-1.47 and 0.96-1.67 for 2 and 6 cm DBH
class widths, respectively, when dominant tree species were assumed to be
known. The use of a predicted tree species map, instead of the observed
information, decreased the REs by up to 10%.
| stat |
Double Descent Risk and Volume Saturation Effects: A Geometric
Perspective | The appearance of the double-descent risk phenomenon has received growing
interest in the machine learning and statistics community, as it challenges
well-understood notions behind the U-shaped train-test curves. Motivated
through Rissanen's minimum description length (MDL), Balasubramanian's Occam's
Razor, and Amari's information geometry, we investigate how the logarithm of
the model volume: $\log V$, works to extend intuition behind the AIC and BIC
model selection criteria. We find that for the particular model classes of
isotropic linear regression and statistical lattices, the $\log V$ term may be
decomposed into a sum of distinct components, each of which assist in their
explanations of the appearance of this phenomenon. In particular they suggest
why generalization error does not necessarily continue to grow with increasing
model dimensionality.
| stat |
Adaptive Checkpoint Adjoint Method for Gradient Estimation in Neural ODE | Neural ordinary differential equations (NODEs) have recently attracted
increasing attention; however, their empirical performance on benchmark tasks
(e.g. image classification) are significantly inferior to discrete-layer
models. We demonstrate an explanation for their poorer performance is the
inaccuracy of existing gradient estimation methods: the adjoint method has
numerical errors in reverse-mode integration; the naive method directly
back-propagates through ODE solvers, but suffers from a redundantly deep
computation graph when searching for the optimal stepsize. We propose the
Adaptive Checkpoint Adjoint (ACA) method: in automatic differentiation, ACA
applies a trajectory checkpoint strategy which records the forward-mode
trajectory as the reverse-mode trajectory to guarantee accuracy; ACA deletes
redundant components for shallow computation graphs; and ACA supports adaptive
solvers. On image classification tasks, compared with the adjoint and naive
method, ACA achieves half the error rate in half the training time; NODE
trained with ACA outperforms ResNet in both accuracy and test-retest
reliability. On time-series modeling, ACA outperforms competing methods.
Finally, in an example of the three-body problem, we show NODE with ACA can
incorporate physical knowledge to achieve better accuracy. We provide the
PyTorch implementation of ACA:
\url{https://github.com/juntang-zhuang/torch-ACA}.
| stat |
An Online Stochastic Kernel Machine for Robust Signal Classification | We present a novel variation of online kernel machines in which we exploit a
consensus based optimization mechanism to guide the evolution of decision
functions drawn from a reproducing kernel Hilbert space, which efficiently
models the observed stationary process.
| stat |
Test & Roll: Profit-Maximizing A/B Tests | Marketers often use A/B testing as a tool to compare marketing treatments in
a test stage and then deploy the better-performing treatment to the remainder
of the consumer population. While these tests have traditionally been analyzed
using hypothesis testing, we re-frame them as an explicit trade-off between the
opportunity cost of the test (where some customers receive a sub-optimal
treatment) and the potential losses associated with deploying a sub-optimal
treatment to the remainder of the population.
We derive a closed-form expression for the profit-maximizing test size and
show that it is substantially smaller than typically recommended for a
hypothesis test, particularly when the response is noisy or when the total
population is small. The common practice of using small holdout groups can be
rationalized by asymmetric priors. The proposed test design achieves nearly the
same expected regret as the flexible, yet harder-to-implement multi-armed
bandit under a wide range of conditions.
We demonstrate the benefits of the method in three different marketing
contexts -- website design, display advertising and catalog tests -- in which
we estimate priors from past data. In all three cases, the optimal sample sizes
are substantially smaller than for a traditional hypothesis test, resulting in
higher profit.
| stat |
Short-term prediction of Time Series based on bounding techniques | In this paper it is reconsidered the prediction problem in time series
framework by using a new non-parametric approach. Through this reconsideration,
the prediction is obtained by a weighted sum of past observed data. These
weights are obtained by solving a constrained linear optimization problem that
minimizes an outer bound of the prediction error. The innovation is to consider
both deterministic and stochastic assumptions in order to obtain the upper
bound of the prediction error, a tuning parameter is used to balance these
deterministic-stochastic assumptions in order to improve the predictor
performance. A benchmark is included to illustrate that the proposed predictor
can obtain suitable results in a prediction scheme, and can be an interesting
alternative method to the classical non-parametric methods. Besides, it is
shown how this model can outperform the preexisting ones in a short term
forecast.
| stat |
Tree Inference: Response Time in a Binary Multinomial Processing Tree,
Representation and Uniqueness of Parameters | A Multinomial Processing Tree (MPT) is a directed tree with a probability
associated with each arc. Here we consider an additional parameter associated
with each arc, a measure such as the time required to select the arc. MPTs are
often used as models of tasks. Each vertex represents a process and an arc
descending from a vertex represents selection of an outcome of the process. A
source vertex represents processing that begins when a stimulus is presented
and a terminal vertex represents making a response. Responses are partitioned
into classes. An experimental factor selectively influences a vertex if
changing the level of the factor changes parameter values on arcs descending
from that vertex and on no others. Earlier work shows that if each of two
experimental factors selectively influences a different vertex in an arbitrary
MPT it is equivalent for the factors to one of two relatively simple MPTs.
Which of the two applies depends on whether the two selectively influenced
vertices are ordered by the factors or not. A special case, the Standard Binary
Tree for Ordered Processes, arises if the vertices are so ordered and the
factor selectively influencing the first vertex changes parameter values on
only two arcs descending from that vertex. Here we derive necessary and
sufficient conditions for the probability and measure associated with a
particular response class to be accounted for by this special case. Parameter
values are not unique and we give admissible transformations for transforming
one set of parameter values to another. When an experiment with two factors is
conducted, the number of observations and parameters to be estimated depend on
the number of levels of each factor; we provide degrees of freedom.
| stat |
Masked Conditional Neural Networks for Audio Classification | We present the ConditionaL Neural Network (CLNN) and the Masked ConditionaL
Neural Network (MCLNN) designed for temporal signal recognition. The CLNN takes
into consideration the temporal nature of the sound signal and the MCLNN
extends upon the CLNN through a binary mask to preserve the spatial locality of
the features and allows an automated exploration of the features combination
analogous to hand-crafting the most relevant features for the recognition task.
MCLNN has achieved competitive recognition accuracies on the GTZAN and the
ISMIR2004 music datasets that surpass several state-of-the-art neural network
based architectures and hand-crafted methods applied on both datasets.
| stat |
Quantile regression: a penalization approach | Sparse group LASSO (SGL) is a penalization technique used in regression
problems where the covariates have a natural grouped structure and provides
solutions that are both between and within group sparse. In this paper the SGL
is introduced to the quantile regression (QR) framework, and a more flexible
version, the adaptive sparse group LASSO (ASGL), is proposed. This proposal
adds weights to the penalization improving prediction accuracy. Usually,
adaptive weights are taken as a function of the original nonpenalized solution
model. This approach is only feasible in the n > p framework. In this work, a
solution that allows using adaptive weights in high-dimensional scenarios is
proposed. The benefits of this proposal are studied both in synthetic and real
datasets.
| stat |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.