title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Learning step sizes for unfolded sparse coding | Sparse coding is typically solved by iterative optimization techniques, such
as the Iterative Shrinkage-Thresholding Algorithm (ISTA). Unfolding and
learning weights of ISTA using neural networks is a practical way to accelerate
estimation. In this paper, we study the selection of adapted step sizes for
ISTA. We show that a simple step size strategy can improve the convergence rate
of ISTA by leveraging the sparsity of the iterates. However, it is impractical
in most large-scale applications. Therefore, we propose a network architecture
where only the step sizes of ISTA are learned. We demonstrate that for a large
class of unfolded algorithms, if the algorithm converges to the solution of the
Lasso, its last layers correspond to ISTA with learned step sizes. Experiments
show that our method is competitive with state-of-the-art networks when the
solutions are sparse enough.
| stat |
Importance Gaussian Quadrature | Importance sampling (IS) and numerical integration methods are usually
employed for approximating moments of complicated target distributions. In its
basic procedure, the IS methodology randomly draws samples from a proposal
distribution and weights them accordingly, accounting for the mismatch between
the target and proposal. In this work, we present a general framework of
numerical integration techniques inspired by the IS methodology. The framework
can also be seen as an incorporation of deterministic rules into IS methods,
reducing the error of the estimators by several orders of magnitude in several
problems of interest. The proposed approach extends the range of applicability
of the Gaussian quadrature rules. For instance, the IS perspective allows us to
use Gauss-Hermite rules in problems where the integrand is not involving a
Gaussian distribution, and even more, when the integrand can only be evaluated
up to a normalizing constant, as it is usually the case in Bayesian inference.
The novel perspective makes use of recent advances on the multiple IS (MIS) and
adaptive (AIS) literatures, and incorporates it to a wider numerical
integration framework that combines several numerical integration rules that
can be iteratively adapted. We analyze the convergence of the algorithms and
provide some representative examples showing the superiority of the proposed
approach in terms of performance.
| stat |
Conditional As-If Analyses in Randomized Experiments | The injunction to `analyze the way you randomize' is well-known to
statisticians since Fisher advocated for randomization as the basis of
inference. Yet even those convinced by the merits of randomization-based
inference seldom follow this injunction to the letter. Bernoulli randomized
experiments are often analyzed as completely randomized experiments, and
completely randomized experiments are analyzed as if they had been stratified;
more generally, it is not uncommon to analyze an experiment as if it had been
randomized differently. This paper examines the theoretical foundation behind
this practice within a randomization-based framework. Specifically, we ask when
is it legitimate to analyze an experiment randomized according to one design as
if it had been randomized according to some other design. We show that a
sufficient condition for this type of analysis to be valid is that the design
used for analysis be derived from the original design by an appropriate form of
conditioning. We use our theory to justify certain existing methods, question
others, and finally suggest new methodological insights such as conditioning on
approximate covariate balance.
| stat |
Models as Approximations I: Consequences Illustrated with Linear
Regression | In the early 1980s Halbert White inaugurated a "model-robust'' form of
statistical inference based on the "sandwich estimator'' of standard error.
This estimator is known to be "heteroskedasticity-consistent", but it is less
well-known to be "nonlinearity-consistent'' as well. Nonlinearity, however,
raises fundamental issues because in its presence regressors are not ancillary,
hence can't be treated as fixed.
The consequences are deep: (1)~population slopes need to be re-interpreted as
statistical functionals obtained from OLS fits to largely arbitrary joint
$\xy$~distributions; (2)~the meaning of slope parameters needs to be rethought;
(3)~the regressor distribution affects the slope parameters; (4)~randomness of
the regressors becomes a source of sampling variability in slope estimates;
(5)~inference needs to be based on model-robust standard errors, including
sandwich estimators or the $\xy$~bootstrap. In theory, model-robust and
model-trusting standard errors can deviate by arbitrary magnitudes either way.
In practice, significant deviations between them can be detected with a
diagnostic test.
| stat |
Interaction between two exposures: determining odds ratios and
confidence intervals for risk estimates | In epidemiological research, it is common to investigate the interaction
between risk factors for an outcome such as a disease and hence to estimate the
risk associated with being exposed for either or both of two risk factors under
investigation. Interactions can be estimated both on the additive and
multiplicative scale using the same regression model. We here present a review
for calculating interaction and estimating the risk and confidence interval of
two exposures using a single regression model and the relationship between
measures, particularly the standard error for the combined exposure risk group.
| stat |
Weighted Cox regression for the prediction of heterogeneous patient
subgroups | An important task in clinical medicine is the construction of risk prediction
models for specific subgroups of patients based on high-dimensional molecular
measurements such as gene expression data. Major objectives in modeling
high-dimensional data are good prediction performance and feature selection to
find a subset of predictors that are truly associated with a clinical outcome
such as a time-to-event endpoint. In clinical practice, this task is
challenging since patient cohorts are typically small and can be heterogeneous
with regard to their relationship between predictors and outcome. When data of
several subgroups of patients with the same or similar disease are available,
it is tempting to combine them to increase sample size, such as in multicenter
studies. However, heterogeneity between subgroups can lead to biased results
and subgroup-specific effects may remain undetected. For this situation, we
propose a penalized Cox regression model with a weighted version of the Cox
partial likelihood that includes patients of all subgroups but assigns them
individual weights based on their subgroup affiliation. Patients who are likely
to belong to the subgroup of interest obtain higher weights in the
subgroup-specific model. Our proposed approach is evaluated through simulations
and application to real lung cancer cohorts. Simulation results demonstrate
that our model can achieve improved prediction and variable selection accuracy
over standard approaches.
| stat |
Minimax Pareto Fairness: A Multi Objective Perspective | In this work we formulate and formally characterize group fairness as a
multi-objective optimization problem, where each sensitive group risk is a
separate objective. We propose a fairness criterion where a classifier achieves
minimax risk and is Pareto-efficient w.r.t. all groups, avoiding unnecessary
harm, and can lead to the best zero-gap model if policy dictates so. We provide
a simple optimization algorithm compatible with deep neural networks to satisfy
these constraints. Since our method does not require test-time access to
sensitive attributes, it can be applied to reduce worst-case classification
errors between outcomes in unbalanced classification problems. We test the
proposed methodology on real case-studies of predicting income, ICU patient
mortality, skin lesions classification, and assessing credit risk,
demonstrating how our framework compares favorably to other approaches.
| stat |
Confidence Intervals for Seroprevalence | This paper concerns the construction of confidence intervals in standard
seroprevalence surveys. In particular, we discuss methods for constructing
confidence intervals for the proportion of individuals in a population infected
with a disease using a sample of antibody test results and measurements of the
test's false positive and false negative rates. We begin by documenting erratic
behavior in the coverage probabilities of standard Wald and percentile
bootstrap intervals when applied to this problem. We then consider two
alternative sets of intervals constructed with test inversion. The first set of
intervals are approximate, using either asymptotic or bootstrap approximation
to the finite-sample distribution of a chosen test statistic. We consider
several choices of test statistic, including maximum likelihood estimators and
generalized likelihood ratio statistics. We show with simulation that, at
empirically relevant parameter values and sample sizes, the coverage
probabilities for these intervals are close to their nominal level and are
approximately equi-tailed. The second set of intervals are shown to contain the
true parameter value with probability at least equal to the nominal level, but
can be conservative in finite samples. To conclude, we outline the application
of the methods that we consider to several related problems, and we provide a
set of practical recommendations.
| stat |
A Loss-Based Prior for Gaussian Graphical Models | Gaussian graphical models play an important role in various areas such as
genetics, finance, statistical physics and others. They are a powerful
modelling tool which allows one to describe the relationships among the
variables of interest. From the Bayesian perspective, there are two sources of
randomness: one is related to the multivariate distribution and the quantities
that may parametrise the model, the other has to do with the underlying graph,
$G$, equivalent to describing the conditional independence structure of the
model under consideration. In this paper, we propose a prior on G based on two
loss components. One considers the loss in information one would incur in
selecting the wrong graph, while the second penalises for large number of
edges, favouring sparsity. We illustrate the prior on simulated data and on
real datasets, and compare the results with other priors on $G$ used in the
literature. Moreover, we present a default choice of the prior as well as
discuss how it can be calibrated so as to reflect available prior information.
| stat |
Approximation Properties of Variational Bayes for Vector Autoregressions | Variational Bayes (VB) is a recent approximate method for Bayesian inference.
It has the merit of being a fast and scalable alternative to Markov Chain Monte
Carlo (MCMC) but its approximation error is often unknown. In this paper, we
derive the approximation error of VB in terms of mean, mode, variance,
predictive density and KL divergence for the linear Gaussian multi-equation
regression. Our results indicate that VB approximates the posterior mean
perfectly. Factors affecting the magnitude of underestimation in posterior
variance and mode are revealed. Importantly, We demonstrate that VB estimates
predictive densities accurately.
| stat |
Improved $q$-values for discrete uniform and homogeneous tests: a
comparative study | Large scale discrete uniform and homogeneous $P$-values often arise in
applications with multiple testing. For example, this occurs in genome wide
association studies whenever a nonparametric one-sample (or two-sample) test is
applied throughout the gene loci. In this paper we consider $q$-values for such
scenarios based on several existing estimators for the proportion of true null
hypothesis, $\pi_0$, which take the discreteness of the $P$-values into
account. The theoretical guarantees of the several approaches with respect to
the estimation of $\pi_0$ and the false discovery rate control are reviewed.
The performance of the discrete $q$-values is investigated through intensive
Monte Carlo simulations, including location, scale and omnibus nonparametric
tests, and possibly dependent $P$-values. The methods are applied to genetic
and financial data for illustration purposes too. Since the particular
estimator of $\pi_0$ used to compute the $q$-values may influence the power,
relative advantages and disadvantages of the reviewed procedures are discussed.
Practical recommendations are given.
| stat |
Posterior Ratio Estimation of Latent Variables | Density Ratio Estimation has attracted attention from the machine learning
community due to its ability to compare the underlying distributions of two
datasets. However, in some applications, we want to compare distributions of
random variables that are \emph{inferred} from observations. In this paper, we
study the problem of estimating the ratio between two posterior probability
density functions of a latent variable. Particularly, we assume the posterior
ratio function can be well-approximated by a parametric model, which is then
estimated using observed information and prior samples. We prove the
consistency of our estimator and the asymptotic normality of the estimated
parameters as the number of prior samples tending to infinity. Finally, we
validate our theories using numerical experiments and demonstrate the
usefulness of the proposed method through some real-world applications.
| stat |
Efficiency Assessment of Approximated Spatial Predictions for Large
Datasets | Due to the well-known computational showstopper of the exact Maximum
Likelihood Estimation (MLE) for large geospatial observations, a variety of
approximation methods have been proposed in the literature, which usually
require tuning certain inputs. For example, the recently developed Tile
Low-Rank approximation (TLR) method involves many tuning parameters, including
numerical accuracy. To properly choose the tuning parameters, it is crucial to
adopt a meaningful criterion for the assessment of the prediction efficiency
with different inputs, which the most commonly-used Mean Square Prediction
Error (MSPE) criterion and the Kullback-Leibler Divergence criterion cannot
fully describe. In this paper, we present three other criteria, the Mean Loss
of Efficiency (MLOE), Mean Misspecification of the Mean Square Error (MMOM),
and Root mean square MOM (RMOM), and show numerically that, in comparison with
the common MSPE criterion and the Kullback-Leibler Divergence criterion, our
criteria are more informative, and thus more adequate to assess the loss of the
prediction efficiency by using the approximated or misspecified covariance
models. Hence, our suggested criteria are more useful for the determination of
tuning parameters for sophisticated approximation methods of spatial model
fitting. To illustrate this, we investigate the trade-off between the execution
time, estimation accuracy, and prediction efficiency for the TLR method with
extensive simulation studies and suggest proper settings of the TLR tuning
parameters. We then apply the TLR method to a large spatial dataset of soil
moisture in the area of the Mississippi River basin, and compare the TLR with
the Gaussian predictive process and the composite likelihood method, showing
that our suggested criteria can successfully be used to choose the tuning
parameters that can keep the estimation or the prediction accuracy in
applications.
| stat |
On a length-biased Birnbaum-Saunders regression model applied to
meteorological data | The length-biased Birnbaum-Saunders distribution is both useful and practical
for environmental sciences. In this paper, we initially derive some new
properties for the length-biased Birnbaum-Saunders distribution, showing that
one of its parameters is the mode and that it is bimodal. We then introduce a
new regression model based on this distribution. We implement use the maximum
likelihood method for parameter estimation, approach interval estimation and
consider three types of residuals. An elaborate Monte Carlo study is carried
out for evaluating the performance of the likelihood-based estimates, the
confidence intervals and the empirical distribution of the residuals. Finally,
we illustrate the proposed regression model with the use of a real
meteorological data set.
| stat |
B-PINNs: Bayesian Physics-Informed Neural Networks for Forward and
Inverse PDE Problems with Noisy Data | We propose a Bayesian physics-informed neural network (B-PINN) to solve both
forward and inverse nonlinear problems described by partial differential
equations (PDEs) and noisy data. In this Bayesian framework, the Bayesian
neural network (BNN) combined with a PINN for PDEs serves as the prior while
the Hamiltonian Monte Carlo (HMC) or the variational inference (VI) could serve
as an estimator of the posterior. B-PINNs make use of both physical laws and
scattered noisy measurements to provide predictions and quantify the aleatoric
uncertainty arising from the noisy data in the Bayesian framework. Compared
with PINNs, in addition to uncertainty quantification, B-PINNs obtain more
accurate predictions in scenarios with large noise due to their capability of
avoiding overfitting. We conduct a systematic comparison between the two
different approaches for the B-PINN posterior estimation (i.e., HMC or VI),
along with dropout used for quantifying uncertainty in deep neural networks.
Our experiments show that HMC is more suitable than VI for the B-PINNs
posterior estimation, while dropout employed in PINNs can hardly provide
accurate predictions with reasonable uncertainty. Finally, we replace the BNN
in the prior with a truncated Karhunen-Lo\`eve (KL) expansion combined with HMC
or a deep normalizing flow (DNF) model as posterior estimators. The KL is as
accurate as BNN and much faster but this framework cannot be easily extended to
high-dimensional problems unlike the BNN based framework.
| stat |
Delayed acceptance ABC-SMC | Approximate Bayesian computation (ABC) is now an established technique for
statistical inference used in cases where the likelihood function is
computationally expensive or not available. It relies on the use of a~model
that is specified in the form of a~simulator, and approximates the likelihood
at a~parameter value $\theta$ by simulating auxiliary data sets $x$ and
evaluating the distance of $x$ from the true data $y$. However, ABC is not
computationally feasible in cases where using the simulator for each $\theta$
is very expensive. This paper investigates this situation in cases where
a~cheap, but approximate, simulator is available. The approach is to employ
delayed acceptance Markov chain Monte Carlo (MCMC) within an ABC sequential
Monte Carlo (SMC) sampler in order to, in a~first stage of the kernel, use the
cheap simulator to rule out parts of the parameter space that are not worth
exploring, so that the ``true'' simulator is only run (in the second stage of
the kernel) where there is a~reasonable chance of accepting proposed values of
$\theta$. We show that this approach can be used quite automatically, with few
tuning parameters. Applications to stochastic differential equation models and
latent doubly intractable distributions are presented.
| stat |
Sparse Representations of Positive Functions via Projected Pseudo-Mirror
Descent | We consider the problem of expected risk minimization when the population
loss is strongly convex and the target domain of the decision variable is
required to be nonnegative, motivated by the settings of maximum likelihood
estimation (MLE) and trajectory optimization. We restrict focus to the case
that the decision variable belongs to a nonparametric Reproducing Kernel
Hilbert Space (RKHS). To solve it, we consider stochastic mirror descent that
employs (i) pseudo-gradients and (ii) projections. Compressive projections are
executed via kernel orthogonal matching pursuit (KOMP), and overcome the fact
that the vanilla RKHS parameterization grows unbounded with time. Moreover,
pseudo-gradients are needed, e.g., when stochastic gradients themselves define
integrals over unknown quantities that must be evaluated numerically, as in
estimating the intensity parameter of an inhomogeneous Poisson Process, and
multi-class kernel logistic regression with latent multi-kernels. We establish
tradeoffs between accuracy of convergence in mean and the projection budget
parameter under constant step-size and compression budget, as well as
non-asymptotic bounds on the model complexity. Experiments demonstrate that we
achieve state-of-the-art accuracy and complexity tradeoffs for inhomogeneous
Poisson Process intensity estimation and multi-class kernel logistic
regression.
| stat |
An analytic theory of shallow networks dynamics for hinge loss
classification | Neural networks have been shown to perform incredibly well in classification
tasks over structured high-dimensional datasets. However, the learning dynamics
of such networks is still poorly understood. In this paper we study in detail
the training dynamics of a simple type of neural network: a single hidden layer
trained to perform a classification task. We show that in a suitable mean-field
limit this case maps to a single-node learning problem with a time-dependent
dataset determined self-consistently from the average nodes population. We
specialize our theory to the prototypical case of a linearly separable dataset
and a linear hinge loss, for which the dynamics can be explicitly solved. This
allow us to address in a simple setting several phenomena appearing in modern
networks such as slowing down of training dynamics, crossover between rich and
lazy learning, and overfitting. Finally, we asses the limitations of mean-field
theory by studying the case of large but finite number of nodes and of training
samples.
| stat |
Simulator Calibration under Covariate Shift with Kernels | We propose a novel calibration method for computer simulators, dealing with
the problem of covariate shift. Covariate shift is the situation where input
distributions for training and test are different, and ubiquitous in
applications of simulations. Our approach is based on Bayesian inference with
kernel mean embedding of distributions, and on the use of an
importance-weighted reproducing kernel for covariate shift adaptation. We
provide a theoretical analysis for the proposed method, including a novel
theoretical result for conditional mean embedding, as well as empirical
investigations suggesting its effectiveness in practice. The experiments
include calibration of a widely used simulator for industrial manufacturing
processes, where we also demonstrate how the proposed method may be useful for
sensitivity analysis of model parameters.
| stat |
A Latent Slice Sampling Algorithm | In this paper we introduce a new sampling algorithm which has the potential
to be adopted as a universal replacement to the Metropolis--Hastings algorithm.
It is related to the slice sampler, and motivated by an algorithm which is
applicable to discrete probability distributions %which can be viewed as an
alternative to the Metropolis--Hastings algorithm in this setting, which
obviates the need for a proposal distribution, in that is has no accept/reject
component. This paper looks at the continuous counterpart. A latent variable
combined with a slice sampler and a shrinkage procedure applied to uniform
density functions creates a highly efficient sampler which can generate random
variables from very high dimensional distributions as a single block.
| stat |
A Fast, Scalable, and Calibrated Computer Model Emulator: An Empirical
Bayes Approach | Mathematical models implemented on a computer have become the driving force
behind the acceleration of the cycle of scientific processes. This is because
computer models are typically much faster and economical to run than physical
experiments. In this work, we develop an empirical Bayes approach to
predictions of physical quantities using a computer model, where we assume that
the computer model under consideration needs to be calibrated and is
computationally expensive. We propose a Gaussian process emulator and a
Gaussian process model for the systematic discrepancy between the computer
model and the underlying physical process. This allows for closed-form and
easy-to-compute predictions given by a conditional distribution induced by the
Gaussian processes. We provide a rigorous theoretical justification of the
proposed approach by establishing posterior consistency of the estimated
physical process. The computational efficiency of the methods is demonstrated
in an extensive simulation study and a real data example. The newly established
approach makes enhanced use of computer models both from practical and
theoretical standpoints.
| stat |
Using Deep Networks for Scientific Discovery in Physiological Signals | Deep neural networks (DNN) have shown remarkable success in the
classification of physiological signals. In this study we propose a method for
examining to what extent does a DNN's performance rely on rediscovering
existing features of the signals, as opposed to discovering genuinely new
features. Moreover, we offer a novel method of "removing" a hand-engineered
feature from the network's hypothesis space, thus forcing it to try and learn
representations which are different from known ones, as a method of scientific
exploration. We then build on existing work in the field of interpretability,
specifically class activation maps, to try and infer what new features the
network has learned. We demonstrate this approach using ECG and EEG signals.
With respect to ECG signals we show that for the specific task of classifying
atrial fibrillation, DNNs are likely rediscovering known features. We also show
how our method could be used to discover new features, by selectively removing
some ECG features and "rediscovering" them. We further examine how could our
method be used as a tool for examining scientific hypotheses. We simulate this
scenario by looking into the importance of eye movements in classifying sleep
from EEG. We show that our tool can successfully focus a researcher's attention
by bringing to light patterns in the data that would be hidden otherwise.
| stat |
Urban Space Insights Extraction using Acoustic Histogram Information | Urban data mining can be identified as a highly potential area that can
enhance the smart city services towards better sustainable development
especially in the urban residential activity tracking. While existing human
activity tracking systems have demonstrated the capability to unveil the hidden
aspects of citizens' behavior, they often come with a high implementation cost
and require a large communication bandwidth. In this paper, we study the
implementation of low-cost analogue sound sensors to detect outdoor activities
and estimate the raining period in an urban residential area. The analogue
sound sensors are transmitted to the cloud every 5 minutes in histogram format,
which consists of sound data sampled every 100ms (10Hz). We then use wavelet
transformation (WT) and principal component analysis (PCA) to generate a more
robust and consistent feature set from the histogram. After that, we performed
unsupervised clustering and attempt to understand the individual
characteristics of each cluster to identify outdoor residential activities. In
addition, on-site validation has been conducted to show the effectiveness of
our approach.
| stat |
Causal Imputation via Synthetic Interventions | Consider the problem of determining the effect of a compound on a specific
cell type. To answer this question, researchers traditionally need to run an
experiment applying the drug of interest to that cell type. This approach is
not scalable: given a large number of different actions (compounds) and a large
number of different contexts (cell types), it is infeasible to run an
experiment for every action-context pair. In such cases, one would ideally like
to predict the outcome for every pair while only having to perform experiments
on a small subset of pairs. This task, which we label "causal imputation", is a
generalization of the causal transportability problem. To address this
challenge, we extend the recently introduced synthetic interventions (SI)
estimator to handle more general data sparsity patterns. We prove that, under a
latent factor model, our estimator provides valid estimates for the causal
imputation task. We motivate this model by establishing a connection to the
linear structural causal model literature. Finally, we consider the prominent
CMAP dataset in predicting the effects of compounds on gene expression across
cell types. We find that our estimator outperforms standard baselines, thus
confirming its utility in biological applications.
| stat |
Bayesian Inference of a Dependent Competing Risk Data | Analysis of competing risks data plays an important role in the lifetime data
analysis. Recently Feizjavadian and Hashemi (Computational Statistics and Data
Analysis, vol. 82, 19-34, 2015) provided a classical inference of a competing
risks data set using four-parameter Marshall-Olkin bivariate Weibull
distribution when the failure of an unit at a particular time point can happen
due to more than one cause. The aim of this paper is to provide the Bayesian
analysis of the same model based on a very flexible Gamma-Dirichlet prior on
the scale parameters. It is observed that the Bayesian inference has certain
advantages over the classical inference in this case. We provide the Bayes
estimates of the unknown parameters and the associated highest posterior
density credible intervals based on Gibbs sampling technique. We further
consider the Bayesian inference of the model parameters assuming partially
ordered Gamma-Dirichlet prior on the scale parameters when one cause is more
severe than the other cause. We have extended the results for different
censoring schemes also.
| stat |
Profiling Compliers in Instrumental Variables Designs | Instrumental variable (IV) analyses are becoming common in health services
research and epidemiology. IV analyses can be used both to analyze randomized
trials with noncompliance and as a form of natural experiment. In these
analyses, investigators often adopt a monotonicity assumption, which implies
that the relevant effect only applies to a subset of the study population known
as compliers. Since the estimated effect is not the average treatment effect of
the study population, it is important to compare the characteristics of
compliers and non-compliers. Profiling compliers and non-compliers is necessary
to understand what subpopulation the researcher is making inferences about, and
an important first step in evaluating the external validity (or lack thereof)
of the IV estimate for compliers. Here, we discuss the assumptions necessary
for profiling, which are weaker than the assumptions necessary for identifying
the local average treatment effect if the instrument is randomly assigned. We
then outline a simple and general method to characterize compliers and
noncompliers using baseline covariates. Next, we extend current methods by
deriving standard errors for these estimates. We demonstrate these methods
using an IV known as tendency to operate (TTO) from health services research.
| stat |
MONK -- Outlier-Robust Mean Embedding Estimation by Median-of-Means | Mean embeddings provide an extremely flexible and powerful tool in machine
learning and statistics to represent probability distributions and define a
semi-metric (MMD, maximum mean discrepancy; also called N-distance or energy
distance), with numerous successful applications. The representation is
constructed as the expectation of the feature map defined by a kernel. As a
mean, its classical empirical estimator, however, can be arbitrary severely
affected even by a single outlier in case of unbounded features. To the best of
our knowledge, unfortunately even the consistency of the existing few
techniques trying to alleviate this serious sensitivity bottleneck is unknown.
In this paper, we show how the recently emerged principle of median-of-means
can be used to design estimators for kernel mean embedding and MMD with
excessive resistance properties to outliers, and optimal sub-Gaussian deviation
bounds under mild assumptions.
| stat |
Defence Against the Modern Arts: the Curse of Statistics -- FRStat | For several decades, legal and scientific scholars have argued that
conclusions from forensic examinations should be supported by statistical data
and reported within a probabilistic framework. Multiple models have been
proposed to quantify the probative value of forensic evidence. Unfortunately,
several of these models rely on ad-hoc strategies that are not scientifically
sound. The opacity of the technical jargon that is used to present these models
and their results and the complexity of the techniques involved make it very
difficult for the untrained user to separate the wheat from the chaff. This
series of paper is intended to help forensic scientists and lawyers recognise
issues in tools proposed to interpret the results of forensic examinations.
This paper focuses on the tool proposed by the Latent Print Branch of the U.S.
Defense Forensic Science Center (DFSC) and called FRStat. In this paper, I
explore the compatibility of the results outputted by FRStat with the language
used by the DFCS to report the conclusions of their fingerprint examinations,
as well as the appropriateness of the statistical modelling underpinning the
tool and the validation of its performance.
| stat |
Exploring the roles of local mobility patterns, socioeconomic
conditions, and lockdown policies in shaping the patterns of COVID-19 spread | The COVID-19 crisis has shown that we can only prevent the risk of mass
contagion through timely, large-scale, coordinated, and decisive actions.
However, frequently the models used by experts [from whom decision-makers get
their main advice] focus on a single perspective [for example, the
epidemiological one] and do not consider many of the multiple forces that
affect the COVID-19 outbreak patterns. The epidemiological, socioeconomic, and
human mobility context of COVID-19 can be considered as a complex adaptive
system. So, these interventions (for example, lock-downs) could have many
and/or unexpected ramifications. This situation makes it difficult to
understand the overall effect produced by any public policy measure and,
therefore, to assess its real effectiveness and convenience. By using mobile
phone data, socioeconomic data, and COVID-19 cases data recorded throughout the
pandemic development, we aim to understand and explain [make sense of] the
observed heterogeneous regional patterns of contagion across time and space. We
will also consider the causal effects produced by confinement policies by
developing data-based models to explore, simulate, and estimate these policies'
effectiveness. We intend to develop a methodology to assess and improve public
policies' effectiveness associated with the fight against the pandemic,
emphasizing its convenience, the precise time of its application, and
extension. The contributions of this work can be used regardless of the region.
The only likely impediment is the availability of the appropriate data.
| stat |
Analyzing the State of COVID-19: Real-time Visual Data Analysis,
Short-Term Forecasting, and Risk Factor Identification | The COVID-19 outbreak was initially reported in Wuhan, China, and it has been
declared as a Public Health Emergency of International Concern (PHEIC) on 30
January 2020 by WHO. It has now spread to over 180 countries, and it has
gradually evolved into a worldwide pandemic, endangering the state of global
public health and becoming a serious threat to the global community. To combat
and prevent the spread of the disease, all individuals should be well-informed
of the rapidly changing state of COVID-19. To accomplish this objective, I have
built a website to analyze and deliver the latest state of the disease and
relevant analytical insights. The website is designed to cater to the general
audience, and it aims to communicate insights through various straightforward
and concise data visualizations that are supported by sound statistical
methods, accurate data modeling, state-of-the-art natural language processing
techniques, and reliable data sources. This paper discusses the major
methodologies which are utilized to generate the insights displayed on the
website, which include an automatic data ingestion pipeline, normalization
techniques, moving average computation, ARIMA time-series forecasting, and
logistic regression models. In addition, the paper highlights key discoveries
that have been derived in regard to COVID-19 using the methodologies.
| stat |
Bayesian estimation of probabilistic sensitivity measures | Computer experiments are becoming increasingly important in scientific
investigations. In the presence of uncertainty, analysts employ probabilistic
sensitivity methods to identify the key-drivers of change in the quantities of
interest. Simulation complexity, large dimensionality and long running times
may force analysts to make statistical inference at small sample sizes. Methods
designed to estimate probabilistic sensitivity measures at relatively low
computational costs are attracting increasing interest. We propose a fully
Bayesian approach to the estimation of probabilistic sensitivity measures based
on a one-sample design. We discuss, first, new estimators based on placing
piecewise constant priors on the conditional distributions of the output given
each input, by partitioning the input space. We then present two alternatives,
based on Bayesian non-parametric density estimation, which bypass the need for
predefined partitions. In all cases, the Bayesian paradigm guarantees the
quantification of uncertainty in the estimation process through the posterior
distribution over the sensitivity measures, without requiring additional
simulator evaluations. The performance of the proposed methods is compared to
that of traditional point estimators in a series of numerical experiments
comprising synthetic but challenging simulators, as well as a realistic
application. $\textit{An Updated Version of the Manuscript is Forthcoming in
Statistics and Computing.}$
| stat |
A global approach for learning sparse Ising models | We consider the problem of learning the link parameters as well as the
structure of a binary-valued pairwise Markov model. Under sparsity assumption,
we propose a method based on $l_1$- regularized logistic regression, which
estimate globally the whole set of edges and link parameters. Unlike the more
recent methods discussed in literature that learn the edges and the
corresponding link parameters one node at a time, in this work we propose a
method that learns all the edges and corresponding link parameters
simultaneously for all nodes. The idea behind this proposal is to exploit the
reciprocal information of the nodes between each other during the estimation
process. Numerical experiments highlight the advantage of this technique and
confirm the intuition behind it.
| stat |
On resampling methods for model assessment in penalized and unpenalized
logistic regression | Penalized logistic regression methods are frequently used to investigate the
relationship between a binary outcome and a set of explanatory variables. The
model performance can be assessed by measures such as the concordance statistic
(c-statistic), the discrimination slope and the Brier score. Often, data
resampling techniques, e.g. crossvalidation, are employed to correct for
optimism in these model performance criteria. Especially with small samples or
a rare binary outcome variable, leave-one-out crossvalidation is a popular
choice. Using simulations and a real data example, we compared the effect of
different resampling techniques on the estimation of c-statistics,
discrimination slopes and Brier scores for three estimators of logistic
regression models, including the maximum likelihood and two maximum
penalized-likelihood estimators. Our simulation study confirms earlier studies
reporting that leave-one-out crossvalidated c-statistics can be strongly biased
towards zero. In addition, our study reveals that this bias is more pronounced
for estimators shrinking predicted probabilities towards the observed event
rate, such as ridge regression. Leave-one-out crossvalidation also provided
pessimistic estimates of the discrimination slope but nearly unbiased estimates
of the Brier score. We recommend to use leave-pair-out crossvalidation,
five-fold crossvalidation with repetition, the enhanced or the .632+ bootstrap
to estimate c-statistics and leave-pair-out or five-fold crossvalidation to
estimate discrimination slopes.
| stat |
A Bayesian approach to deconvolution in well test analysis | In petroleum well test analysis, deconvolution is used to obtain information
about the reservoir system. This information is contained in the response
function, which can be estimated by solving an inverse problem in the pressure
and flow rate measurements. Our Bayesian approach to this problem is based upon
a parametric physical model of reservoir behaviour, derived from the solution
for fluid flow in a general class of reservoirs. This permits joint parametric
Bayesian inference for both the reservoir parameters and the true pressure and
rate values, which is essential due to the typical levels of observation error.
Using a set of flexible priors for the reservoir parameters to restrict the
solution space to physical behaviours, samples from the posterior are generated
using MCMC. Summaries and visualisations of the reservoir parameters'
posterior, response, and true pressure and rate values can be produced,
interpreted, and model selection can be performed. The method is validated
through a synthetic application, and applied to a field data set. The results
are comparable to the state of the art solution, but through our method we gain
access to system parameters, we can incorporate prior knowledge that excludes
non-physical results, and we can quantify parameter uncertainty.
| stat |
Introducing Bayesian Analysis with $\text{m&m's}^\circledR$: an
active-learning exercise for undergraduates | We present an active-learning strategy for undergraduates that applies
Bayesian analysis to candy-covered chocolate $\text{m&m's}^\circledR$. The
exercise is best suited for small class sizes and tutorial settings, after
students have been introduced to the concepts of Bayesian statistics. The
exercise takes advantage of the non-uniform distribution of
$\text{m&m's}^\circledR~$ colours, and the difference in distributions made at
two different factories. In this paper, we provide the intended learning
outcomes, lesson plan and step-by-step guide for instruction, and open-source
teaching materials. We also suggest an extension to the exercise for the
graduate-level, which incorporates hierarchical Bayesian analysis.
| stat |
On Learnability under General Stochastic Processes | Statistical learning theory under independent and identically distributed
(iid) sampling and online learning theory for worst case individual sequences
are two of the best developed branches of learning theory. Statistical learning
under general non-iid stochastic processes is less mature. We provide two
natural notions of learnability of a function class under a general stochastic
process. We show that both notions are in fact equivalent to online
learnability. Our results are sharpest in the binary classification setting but
we also show that similar results continue to hold in the regression setting.
| stat |
Gaussian Process Landmarking on Manifolds | As a means of improving analysis of biological shapes, we propose an
algorithm for sampling a Riemannian manifold by sequentially selecting points
with maximum uncertainty under a Gaussian process model. This greedy strategy
is known to be near-optimal in the experimental design literature, and appears
to outperform the use of user-placed landmarks in representing the geometry of
biological objects in our application. In the noiseless regime, we establish an
upper bound for the mean squared prediction error (MSPE) in terms of the number
of samples and geometric quantities of the manifold, demonstrating that the
MSPE for our proposed sequential design decays at a rate comparable to the
oracle rate achievable by any sequential or non-sequential optimal design; to
our knowledge this is the first result of this type for sequential experimental
design. The key is to link the greedy algorithm to reduced basis methods in the
context of model reduction for partial differential equations. We expect this
approach will find additional applications in other fields of research.
| stat |
Spatially Clustered Varying Coefficient Model | In various applications with large spatial regions, the relationship between
the response variable and the covariates is expected to exhibit complex spatial
patterns. We propose a spatially clustered varying coefficient model, where the
regression coefficients are allowed to vary smoothly within each cluster but
change abruptly across the boundaries of adjacent clusters, and we develop a
unified approach for simultaneous coefficient estimation and cluster
identification. The varying coefficients are approximated by penalized splines,
and the clusters are identified through a fused concave penalty on differences
in neighboring locations, where the spatial neighbors are specified by the
minimum spanning tree (MST). The optimization is solved efficiently based on
the alternating direction method of multipliers, utilizing the sparsity
structure from MST. Furthermore, we establish the oracle property of the
proposed method considering the structure of MST. Numerical studies show that
the proposed method can efficiently incorporate spatial neighborhood
information and automatically detect possible spatially clustered patterns in
the regression coefficients. An empirical study in oceanography illustrates
that the proposed method is promising to provide informative results.
| stat |
Non-Asymptotic Analysis of Monte Carlo Tree Search | In this work, we consider the popular tree-based search strategy within the
framework of reinforcement learning, the Monte Carlo Tree Search (MCTS), in the
context of infinite-horizon discounted cost Markov Decision Process (MDP).
While MCTS is believed to provide an approximate value function for a given
state with enough simulations, the claimed proof in the seminal works is
incomplete. This is due to the fact that the variant, the Upper Confidence
Bound for Trees (UCT), analyzed in prior works utilizes "logarithmic" bonus
term for balancing exploration and exploitation within the tree-based search,
following the insights from stochastic multi-arm bandit (MAB) literature. In
effect, such an approach assumes that the regret of the underlying recursively
dependent non-stationary MABs concentrates around their mean exponentially in
the number of steps, which is unlikely to hold as pointed out in literature,
even for stationary MABs. As the key contribution of this work, we establish
polynomial concentration property of regret for a class of non-stationary MAB.
This in turn establishes that the MCTS with appropriate polynomial rather than
logarithmic bonus term in UCB has the claimed property. Using this as a
building block, we argue that MCTS, combined with nearest neighbor supervised
learning, acts as a "policy improvement" operator: it iteratively improves
value function approximation for all states, due to combining with supervised
learning, despite evaluating at only finitely many states. In effect, we
establish that to learn an $\varepsilon$ approximation of the value function
with respect to $\ell_\infty$ norm, MCTS combined with nearest neighbor
requires a sample size scaling as
$\widetilde{O}\big(\varepsilon^{-(d+4)}\big)$, where $d$ is the dimension of
the state space. This is nearly optimal due to a minimax lower bound of
$\widetilde{\Omega}\big(\varepsilon^{-(d+2)}\big)$.
| stat |
Improvement of the cross-entropy method in high dimension through a
one-dimensional projection without gradient estimation | Rare event probability estimation is an important topic in reliability
analysis. Stochastic methods, such as importance sampling, have been developed
to estimate such probabilities but they often fail in high dimension. In this
paper, we propose a simple cross-entropy-based importance sampling algorithm to
improve rare event estimation in high dimension. We consider the cross-entropy
method with Gaussian auxiliary distributions and we suggest to update the
Gaussian covariance matrix only in a one-dimensional subspace. The idea is to
project in the one dimensional subspace spanned by the Gaussian mean, which
gives an influential direction for the variance estimation. This approach does
not require any additional simulation budget compared to the basic
cross-entropy algorithm and greatly improves its performance in high dimension,
as we show on different numerical test cases.
| stat |
Use of High Dimensional Modeling for automatic variables selection: the
best path algorithm | This paper presents a new algorithm for automatic variables selection. In
particular, using the Graphical Models properties it is possible to develop a
method that can be used in the contest of large dataset. The advantage of this
algorithm is that can be combined with different forecasting models. In this
research we have used the OLS method and we have compared the result with the
LASSO method.
| stat |
CoinPress: Practical Private Mean and Covariance Estimation | We present simple differentially private estimators for the mean and
covariance of multivariate sub-Gaussian data that are accurate at small sample
sizes. We demonstrate the effectiveness of our algorithms both theoretically
and empirically using synthetic and real-world datasets---showing that their
asymptotic error rates match the state-of-the-art theoretical bounds, and that
they concretely outperform all previous methods. Specifically, previous
estimators either have weak empirical accuracy at small sample sizes, perform
poorly for multivariate data, or require the user to provide strong a priori
estimates for the parameters.
| stat |
A Matrix Chernoff Bound for Markov Chains and Its Application to
Co-occurrence Matrices | We prove a Chernoff-type bound for sums of matrix-valued random variables
sampled via a regular (aperiodic and irreducible) finite Markov chain.
Specially, consider a random walk on a regular Markov chain and a Hermitian
matrix-valued function on its state space. Our result gives exponentially
decreasing bounds on the tail distributions of the extreme eigenvalues of the
sample mean matrix. Our proof is based on the matrix expander (regular
undirected graph) Chernoff bound [Garg et al. STOC '18] and scalar
Chernoff-Hoeffding bounds for Markov chains [Chung et al. STACS '12].
Our matrix Chernoff bound for Markov chains can be applied to analyze the
behavior of co-occurrence statistics for sequential data, which have been
common and important data signals in machine learning. We show that given a
regular Markov chain with $n$ states and mixing time $\tau$, we need a
trajectory of length $O(\tau (\log{(n)}+\log{(\tau)})/\epsilon^2)$ to achieve
an estimator of the co-occurrence matrix with error bound $\epsilon$. We
conduct several experiments and the experimental results are consistent with
the exponentially fast convergence rate from theoretical analysis. Our result
gives the first bound on the convergence rate of the co-occurrence matrix and
the first sample complexity analysis in graph representation learning.
| stat |
A statistical interpretation of spectral embedding: the generalised
random dot product graph | A generalisation of a latent position network model known as the random dot
product graph is considered. We show that, whether the normalised Laplacian or
adjacency matrix is used, the vector representations of nodes obtained by
spectral embedding, using the largest eigenvalues by magnitude, provide
strongly consistent latent position estimates with asymptotically Gaussian
error, up to indefinite orthogonal transformation. The mixed membership and
standard stochastic block models constitute special cases where the latent
positions live respectively inside or on the vertices of a simplex, crucially,
without assuming the underlying block connectivity probability matrix is
positive-definite. Estimation via spectral embedding can therefore be achieved
by respectively estimating this simplicial support, or fitting a Gaussian
mixture model. In the latter case, the use of $K$-means (with Euclidean
distance), as has been previously recommended, is suboptimal and for
identifiability reasons unsound. Indeed, Euclidean distances and angles are not
preserved under indefinite orthogonal transformation, and we show stochastic
block model examples where such quantities vary appreciably. Empirical
improvements in link prediction (over the random dot product graph), as well as
the potential to uncover richer latent structure (than posited under the mixed
membership or standard stochastic block models) are demonstrated in a
cyber-security example.
| stat |
A Bayesian Phylogenetic Hidden Markov Model for B Cell Receptor Sequence
Analysis | The human body is able to generate a diverse set of high affinity antibodies,
the soluble form of B cell receptors (BCRs), that bind to and neutralize
invading pathogens. The natural development of BCRs must be understood in order
to design vaccines for highly mutable pathogens such as influenza and HIV. BCR
diversity is induced by naturally occurring combinatorial "V(D)J"
rearrangement, mutation, and selection processes. Most current methods for BCR
sequence analysis focus on separately modeling the above processes. Statistical
phylogenetic methods are often used to model the mutational dynamics of BCR
sequence data, but these techniques do not consider all the complexities
associated with B cell diversification such as the V(D)J rearrangement process.
In particular, standard phylogenetic approaches assume the DNA bases of the
progenitor (or "naive") sequence arise independently and according to the same
distribution, ignoring the complexities of V(D)J rearrangement. In this paper,
we introduce a novel approach to Bayesian phylogenetic inference for BCR
sequences that is based on a phylogenetic hidden Markov model (phylo-HMM). This
technique not only integrates a naive rearrangement model with a phylogenetic
model for BCR sequence evolution but also naturally accounts for uncertainty in
all unobserved variables, including the phylogenetic tree, via posterior
distribution sampling.
| stat |
Bridging Breiman's Brook: From Algorithmic Modeling to Statistical
Learning | In 2001, Leo Breiman wrote of a divide between "data modeling" and
"algorithmic modeling" cultures. Twenty years later this division feels far
more ephemeral, both in terms of assigning individuals to camps, and in terms
of intellectual boundaries. We argue that this is largely due to the "data
modelers" incorporating algorithmic methods into their toolbox, particularly
driven by recent developments in the statistical understanding of Breiman's own
Random Forest methods. While this can be simplistically described as "Breiman
won", these same developments also expose the limitations of the
prediction-first philosophy that he espoused, making careful statistical
analysis all the more important. This paper outlines these exciting recent
developments in the random forest literature which, in our view, occurred as a
result of a necessary blending of the two ways of thinking Breiman originally
described. We also ask what areas statistics and statisticians might currently
overlook.
| stat |
A Nonparametric Test of Dependence Based on Ensemble of Decision Trees | In this paper, a robust non-parametric measure of statistical dependence, or
correlation, between two random variables is presented. The proposed
coefficient is a permutation-like statistic that quantifies how much the
observed sample S_n : {(X_i , Y_i), i = 1 . . . n} is discriminable from the
permutated sample ^S_nn : {(X_i , Y_j), i, j = 1 . . . n}, where the two
variables are independent. The extent of discriminability is determined using
the predictions for the, interchangeable, leave-out sample from training an
aggregate of decision trees to discriminate between the two samples without
materializing the permutated sample. The proposed coefficient is
computationally efficient, interpretable, invariant to monotonic
transformations, and has a well-approximated distribution under independence.
Empirical results show the proposed method to have a high power for detecting
complex relationships from noisy data.
| stat |
Stochastic Stein Discrepancies | Stein discrepancies (SDs) monitor convergence and non-convergence in
approximate inference when exact integration and sampling are intractable.
However, the computation of a Stein discrepancy can be prohibitive if the Stein
operator - often a sum over likelihood terms or potentials - is expensive to
evaluate. To address this deficiency, we show that stochastic Stein
discrepancies (SSDs) based on subsampled approximations of the Stein operator
inherit the convergence control properties of standard SDs with probability 1.
Along the way, we establish the convergence of Stein variational gradient
descent (SVGD) on unbounded domains, resolving an open question of Liu (2017).
In our experiments with biased Markov chain Monte Carlo (MCMC) hyperparameter
tuning, approximate MCMC sampler selection, and stochastic SVGD, SSDs deliver
comparable inferences to standard SDs with orders of magnitude fewer likelihood
evaluations.
| stat |
Sliced-Wasserstein Flows: Nonparametric Generative Modeling via Optimal
Transport and Diffusions | By building upon the recent theory that established the connection between
implicit generative modeling (IGM) and optimal transport, in this study, we
propose a novel parameter-free algorithm for learning the underlying
distributions of complicated datasets and sampling from them. The proposed
algorithm is based on a functional optimization problem, which aims at finding
a measure that is close to the data distribution as much as possible and also
expressive enough for generative modeling purposes. We formulate the problem as
a gradient flow in the space of probability measures. The connections between
gradient flows and stochastic differential equations let us develop a
computationally efficient algorithm for solving the optimization problem. We
provide formal theoretical analysis where we prove finite-time error guarantees
for the proposed algorithm. To the best of our knowledge, the proposed
algorithm is the first nonparametric IGM algorithm with explicit theoretical
guarantees. Our experimental results support our theory and show that our
algorithm is able to successfully capture the structure of different types of
data distributions.
| stat |
Impossibility of Partial Recovery in the Graph Alignment Problem | Random graph alignment refers to recovering the underlying vertex
correspondence between two random graphs with correlated edges. This can be
viewed as an average-case and noisy version of the well-known NP-hard graph
isomorphism problem. For the correlated Erd\"os-R\'enyi model, we prove an
impossibility result for partial recovery in the sparse regime, with constant
average degree and correlation, as well as a general bound on the maximal
reachable overlap. Our bound is tight in the noiseless case (the graph
isomorphism problem) and we conjecture that it is still tight with noise. Our
proof technique relies on a careful application of the probabilistic method to
build automorphisms between tree components of a subcritical Erd\"os-R\'enyi
graph.
| stat |
Adaptive Experimental Design for Efficient Treatment Effect Estimation | The goal of many scientific experiments including A/B testing is to estimate
the average treatment effect (ATE), which is defined as the difference between
the expected outcomes of two or more treatments. In this paper, we consider a
situation where an experimenter can assign a treatment to research subjects
sequentially. In adaptive experimental design, the experimenter is allowed to
change the probability of assigning a treatment using past observations for
estimating the ATE efficiently. However, with this approach, it is difficult to
apply a standard statistical method to construct an estimator because the
observations are not independent and identically distributed. We thus propose
an algorithm for efficient experiments with estimators constructed from
dependent samples. We also introduce a sequential testing framework using the
proposed estimator. To justify our proposed approach, we provide finite and
infinite sample analyses. Finally, we experimentally show that the proposed
algorithm exhibits preferable performance.
| stat |
Hierarchical Expert Networks for Meta-Learning | The goal of meta-learning is to train a model on a variety of learning tasks,
such that it can adapt to new problems within only a few iterations. Here we
propose a principled information-theoretic model that optimally partitions the
underlying problem space such that specialized expert decision-makers solve the
resulting sub-problems. To drive this specialization we impose the same kind of
information processing constraints both on the partitioning and the expert
decision-makers. We argue that this specialization leads to efficient
adaptation to new tasks. To demonstrate the generality of our approach we
evaluate three meta-learning domains: image classification, regression, and
reinforcement learning.
| stat |
Influence Estimation for Generative Adversarial Networks | Identifying harmful instances, whose absence in a training dataset improves
model performance, is important for building better machine learning models.
Although previous studies have succeeded in estimating harmful instances under
supervised settings, they cannot be trivially extended to generative
adversarial networks (GANs). This is because previous approaches require that
(1) the absence of a training instance directly affects the loss value and that
(2) the change in the loss directly measures the harmfulness of the instance
for the performance of a model. In GAN training, however, neither of the
requirements is satisfied. This is because, (1) the generator's loss is not
directly affected by the training instances as they are not part of the
generator's training steps, and (2) the values of GAN's losses normally do not
capture the generative performance of a model. To this end, (1) we propose an
influence estimation method that uses the Jacobian of the gradient of the
generator's loss with respect to the discriminator's parameters (and vice
versa) to trace how the absence of an instance in the discriminator's training
affects the generator's parameters, and (2) we propose a novel evaluation
scheme, in which we assess harmfulness of each training instance on the basis
of how GAN evaluation metric (e.g., inception score) is expect to change due to
the removal of the instance. We experimentally verified that our influence
estimation method correctly inferred the changes in GAN evaluation metrics.
Further, we demonstrated that the removal of the identified harmful instances
effectively improved the model's generative performance with respect to various
GAN evaluation metrics.
| stat |
Estimation of mask effectiveness perception for small domains using
multiple data sources | All pandemics are local; so learning about the impacts of pandemics on public
health and related societal issues at granular levels is of great interest.
COVID-19 is affecting everyone in the globe and mask wearing is one of the few
precautions against it. To quantify people's perception of mask effectiveness
and to prevent the spread of COVID-19 for small areas, we use Understanding
America Study's (UAS) survey data on COVID-19 as our primary data source. Our
data analysis shows that direct survey-weighted estimates for small areas could
be highly unreliable. In this paper we develop a synthetic estimation method to
estimate proportions of mask effectiveness for small areas using a logistic
model that combines information from multiple data sources. We select our
working model using an extensive data analysis facilitated by a new variable
selection criterion for survey data and benchmarking ratios. We propose a
Jackknife method to estimate variance of our proposed estimator. From our data
analysis. it is evident that our proposed synthetic method outperforms direct
survey-weighted estimator with respect to commonly used evaluation measures.
| stat |
Transfer Learning in Large-scale Gaussian Graphical Models with False
Discovery Rate Control | Transfer learning for high-dimensional Gaussian graphical models (GGMs) is
studied with the goal of estimating the target GGM by utilizing the data from
similar and related auxiliary studies. The similarity between the target graph
and each auxiliary graph is characterized by the sparsity of a divergence
matrix. An estimation algorithm, Trans-CLIME, is proposed and shown to attain a
faster convergence rate than the minimax rate in the single study setting.
Furthermore, a debiased Trans-CLIME estimator is introduced and shown to be
element-wise asymptotically normal. It is used to construct a multiple testing
procedure for edge detection with false discovery rate control. The proposed
estimation and multiple testing procedures demonstrate superior numerical
performance in simulations and are applied to infer the gene networks in a
target brain tissue by leveraging the gene expressions from multiple other
brain tissues. A significant decrease in prediction errors and a significant
increase in power for link detection are observed.
| stat |
Tight Bounds on the Smallest Eigenvalue of the Neural Tangent Kernel for
Deep ReLU Networks | A recent line of work has analyzed the theoretical properties of deep neural
networks via the Neural Tangent Kernel (NTK). In particular, the smallest
eigenvalue of the NTK has been related to memorization capacity, convergence of
gradient descent algorithms and generalization of deep nets. However, existing
results either provide bounds in the two-layer setting or assume that the
spectrum of the NTK is bounded away from 0 for multi-layer networks. In this
paper, we provide tight bounds on the smallest eigenvalue of NTK matrices for
deep ReLU networks, both in the limiting case of infinite widths and for finite
widths. In the finite-width setting, the network architectures we consider are
quite general: we require the existence of a wide layer with roughly order of
$N$ neurons, $N$ being the number of data samples; and the scaling of the
remaining widths is arbitrary (up to logarithmic factors). To obtain our
results, we analyze various quantities of independent interest: we give lower
bounds on the smallest singular value of feature matrices, and upper bounds on
the Lipschitz constant of input-output feature maps.
| stat |
Corrected Bayesian information criterion for stochastic block models | Estimating the number of communities is one of the fundamental problems in
community detection. We re-examine the Bayesian paradigm for stochastic block
models and propose a "corrected Bayesian information criterion",to determine
the number of communities and show that the proposed estimator is consistent
under mild conditions. The proposed criterion improves those used in Wang and
Bickel (2016) and Saldana et al. (2017) which tend to underestimate and
overestimate the number of communities, respectively. Along the way, we
establish the Wilks theorem for stochastic block models. Moreover, we show
that, to obtain the consistency of model selection for stochastic block models,
we need a so-called "consistency condition". We also provide sufficient
conditions for both homogenous networks and non-homogenous networks. The
results are further extended to degree corrected stochastic block models.
Numerical studies demonstrate our theoretical results.
| stat |
ALPS: A Unified Framework for Modeling Time Series of Land Ice Changes | Modeling time series is a research focus in cryospheric sciences because of
the complexity and multiscale nature of events of interest. Highly non-uniform
sampling of measurements from different sensors with different levels of
accuracy, as is typical for measurements of ice sheet elevations, makes the
problem even more challenging. In this paper, we propose a spline-based
approximation framework (ALPS - Approximation by Localized Penalized Splines)
for modeling time series of land ice changes. The localized support of the
B-spline basis functions enable robustness to non-uniform sampling, a
considerable improvement over other global and piecewise local models. With
features like, discrete-coordinate-difference-based penalization and two-level
outlier detection, ALPS further guarantees the stability and quality of
approximations. ALPS incorporates rigorous model uncertainty estimates with all
approximations. As demonstrated by examples, ALPS performs well for a variety
of data sets, including time series of ice sheet thickness, elevation,
velocity, and terminus locations. The robust estimation of time series and
their derivatives facilitates new applications, such as the reconstruction of
high-resolution elevation change records by fusing sparsely sampled time series
of ice sheet thickness changes with modeled firn thickness changes, and the
analysis of the relationship between different outlet glacier observations to
gain new insight into processes and forcing.
| stat |
Robust Pruning at Initialization | Overparameterized Neural Networks (NN) display state-of-the-art performance.
However, there is a growing need for smaller, energy-efficient, neural networks
tobe able to use machine learning applications on devices with limited
computational resources. A popular approach consists of using pruning
techniques. While these techniques have traditionally focused on pruning
pre-trained NN (LeCun et al.,1990; Hassibi et al., 1993), recent work by Lee et
al. (2018) has shown promising results when pruning at initialization. However,
for Deep NNs, such procedures remain unsatisfactory as the resulting pruned
networks can be difficult to train and, for instance, they do not prevent one
layer from being fully pruned. In this paper, we provide a comprehensive
theoretical analysis of Magnitude and Gradient based pruning at initialization
and training of sparse architectures. This allows us to propose novel
principled approaches which we validate experimentally on a variety of NN
architectures.
| stat |
Fast Network Community Detection with Profile-Pseudo Likelihood Methods | The stochastic block model is one of the most studied network models for
community detection. It is well-known that most algorithms proposed for fitting
the stochastic block model likelihood function cannot scale to large-scale
networks. One prominent work that overcomes this computational challenge is
Amini et al.(2013), which proposed a fast pseudo-likelihood approach for
fitting stochastic block models to large sparse networks. However, this
approach does not have convergence guarantee, and is not well suited for small-
or medium- scale networks. In this article, we propose a novel likelihood based
approach that decouples row and column labels in the likelihood function, which
enables a fast alternating maximization; the new method is computationally
efficient, performs well for both small and large scale networks, and has
provable convergence guarantee. We show that our method provides strongly
consistent estimates of the communities in a stochastic block model. As
demonstrated in simulation studies, the proposed method outperforms the
pseudo-likelihood approach in terms of both estimation accuracy and computation
efficiency, especially for large sparse networks. We further consider
extensions of our proposed method to handle networks with degree heterogeneity
and bipartite properties.
| stat |
Variational Bayesian Optimal Experimental Design | Bayesian optimal experimental design (BOED) is a principled framework for
making efficient use of limited experimental resources. Unfortunately, its
applicability is hampered by the difficulty of obtaining accurate estimates of
the expected information gain (EIG) of an experiment. To address this, we
introduce several classes of fast EIG estimators by building on ideas from
amortized variational inference. We show theoretically and empirically that
these estimators can provide significant gains in speed and accuracy over
previous approaches. We further demonstrate the practicality of our approach on
a number of end-to-end experiments.
| stat |
Online Control of the False Coverage Rate and False Sign Rate | The false coverage rate (FCR) is the expected ratio of number of constructed
confidence intervals (CIs) that fail to cover their respective parameters to
the total number of constructed CIs. Procedures for FCR control exist in the
offline setting, but none so far have been designed with the online setting in
mind. In the online setting, there is an infinite sequence of fixed unknown
parameters $\theta_t$ ordered by time. At each step, we see independent data
that is informative about $\theta_t$, and must immediately make a decision
whether to report a CI for $\theta_t$ or not. If $\theta_t$ is selected for
coverage, the task is to determine how to construct a CI for $\theta_t$ such
that $\text{FCR} \leq \alpha$ for any $T\in \mathbb{N}$. A straightforward
solution is to construct at each step a $(1-\alpha)$ level conditional CI. In
this paper, we present a novel solution to the problem inspired by online false
discovery rate (FDR) algorithms, which only requires the statistician to be
able to construct a marginal CI at any given level. Apart from the fact that
marginal CIs are usually simpler to construct than conditional ones, the
marginal procedure has an important qualitative advantage over the conditional
solution, namely, it allows selection to be determined by the candidate CI
itself. We take advantage of this to offer solutions to some online problems
which have not been addressed before. For example, we show that our general CI
procedure can be used to devise online sign-classification procedures that
control the false sign rate (FSR). In terms of power and length of the
constructed CIs, we demonstrate that the two approaches have complementary
strengths and weaknesses using simulations. Last, all of our methodology
applies equally well to online FCR control for prediction intervals, having
particular implications for assumption-free selective conformal inference.
| stat |
Cyanure: An Open-Source Toolbox for Empirical Risk Minimization for
Python, C++, and soon more | Cyanure is an open-source C++ software package with a Python interface. The
goal of Cyanure is to provide state-of-the-art solvers for learning linear
models, based on stochastic variance-reduced stochastic optimization with
acceleration mechanisms. Cyanure can handle a large variety of loss functions
(logistic, square, squared hinge, multinomial logistic) and regularization
functions (l_2, l_1, elastic-net, fused Lasso, multi-task group Lasso). It
provides a simple Python API, which is very close to that of scikit-learn,
which should be extended to other languages such as R or Matlab in a near
future.
| stat |
MAZE: Data-Free Model Stealing Attack Using Zeroth-Order Gradient
Estimation | Model Stealing (MS) attacks allow an adversary with black-box access to a
Machine Learning model to replicate its functionality, compromising the
confidentiality of the model. Such attacks train a clone model by using the
predictions of the target model for different inputs. The effectiveness of such
attacks relies heavily on the availability of data necessary to query the
target model. Existing attacks either assume partial access to the dataset of
the target model or availability of an alternate dataset with semantic
similarities.
This paper proposes MAZE -- a data-free model stealing attack using
zeroth-order gradient estimation. In contrast to prior works, MAZE does not
require any data and instead creates synthetic data using a generative model.
Inspired by recent works in data-free Knowledge Distillation (KD), we train the
generative model using a disagreement objective to produce inputs that maximize
disagreement between the clone and the target model. However, unlike the
white-box setting of KD, where the gradient information is available, training
a generator for model stealing requires performing black-box optimization, as
it involves accessing the target model under attack. MAZE relies on
zeroth-order gradient estimation to perform this optimization and enables a
highly accurate MS attack.
Our evaluation with four datasets shows that MAZE provides a normalized clone
accuracy in the range of 0.91x to 0.99x, and outperforms even the recent
attacks that rely on partial data (JBDA, clone accuracy 0.13x to 0.69x) and
surrogate data (KnockoffNets, clone accuracy 0.52x to 0.97x). We also study an
extension of MAZE in the partial-data setting and develop MAZE-PD, which
generates synthetic data closer to the target distribution. MAZE-PD further
improves the clone accuracy (0.97x to 1.0x) and reduces the query required for
the attack by 2x-24x.
| stat |
A Survey of Wideband Spectrum Sensing Algorithms for Cognitive Radio
Networks and Sub-Nyquist Approaches | Cognitive Radio (CR) networks presents a paradigm shift aiming to alleviate
the spectrum scarcity problem exasperated by the increasing demand on this
limited resource. It promotes dynamic spectrum access, cooperation among
heterogeneous devices, and spectrum sharing. Spectrum sensing is a key
cognitive radio functionality, which entails scanning the RF spectrum to unveil
underutilised spectral bands for opportunistic use. To achieve higher data
rates while maintaining high quality of service QoS, effective wideband
spectrum sensing routines are crucial due to their capability of achieving
spectral awareness over wide frequency range(s)\ and efficiently harnessing the
available opportunities. However, implementing wideband sensing under stringent
size, weight, power and cost requirements (e.g., for portable devices) brings
formidable design challenges such as addressing potential prohibitively high
complexity and data acquisition rates. This article gives a survey of various
wideband spectrum sensing approaches outlining their advantages and
limitations; special attention is paid to approaches that utilise sub-Nyquist
sampling techniques. Other aspects of CR such as cooperative sensing and
performance requirements are briefly addressed. Comparison between sub-Nyquist
sensing approaches is also presented.
| stat |
Scalable nonparametric Bayesian learning for heterogeneous and dynamic
velocity fields | Analysis of heterogeneous patterns in complex spatio-temporal data finds
usage across various domains in applied science and engineering, including
training autonomous vehicles to navigate in complex traffic scenarios.
Motivated by applications arising in the transportation domain, in this paper
we develop a model for learning heterogeneous and dynamic patterns of velocity
field data. We draw from basic nonparameric Bayesian modeling elements such as
hierarchical Dirichlet process and infinite hidden Markov model, while the
smoothness of each homogeneous velocity field element is captured with a
Gaussian process prior. Of particular focus is a scalable approximate inference
method for the proposed model; this is achieved by employing sequential MAP
estimates from the infinite HMM model and an efficient sequential GP posterior
computation technique, which is shown to work effectively on simulated data
sets. Finally, we demonstrate the effectiveness of our techniques to the NGSIM
dataset of complex multi-vehicle interactions.
| stat |
Adapting Neural Networks for Uplift Models | Uplift is a particular case of individual treatment effect modeling. Such
models deal with cause-and-effect inference for a specific factor, such as a
marketing intervention. In practice, these models are built on customer data
who purchased products or services to improve product marketing. Uplift is
estimated using either i) conditional mean regression or ii) transformed
outcome regression. Most existing approaches are adaptations of classification
and regression trees for the uplift case. However, in practice, these
conventional approaches are prone to overfitting. Here we propose a new method
using neural networks. This representation allows to jointly optimize the
difference in conditional means and the transformed outcome losses. As a
consequence, the model not only estimates the uplift, but also ensures
consistency in predicting the outcome. We focus on fully randomized
experiments, which is the case of our data. We show our proposed method
improves the state-of-the-art on synthetic and real data.
| stat |
Regularized Sparse Gaussian Processes | Gaussian processes are a flexible Bayesian nonparametric modelling approach
that has been widely applied to learning tasks such as facial expression
recognition, image reconstruction, and human pose estimation. To address the
issues of poor scaling from exact inference methods, approximation methods
based on sparse Gaussian processes (SGP) and variational inference (VI) are
necessary for the inference on large datasets. However, one of the problems
involved in SGP, especially in latent variable models, is that the distribution
of the inducing inputs may fail to capture the distribution of training inputs,
which may lead to inefficient inference and poor model prediction. Hence, we
propose a regularization approach for sparse Gaussian processes. We also extend
this regularization approach into latent sparse Gaussian processes in a unified
view, considering the balance of the distribution of inducing inputs and
embedding inputs. Furthermore, we justify that performing VI on a sparse latent
Gaussian process with this regularization term is equivalent to performing VI
on a related empirical Bayes model with a prior on the inducing inputs. Also
stochastic variational inference is available for our regularization approach.
Finally, the feasibility of our proposed regularization method is demonstrated
on three real-world datasets.
| stat |
Analysis of time-to-event for observational studies: Guidance to the use
of intensity models | This paper provides guidance for researchers with some mathematical
background on the conduct of time-to-event analysis in observational studies
based on intensity (hazard) models. Discussions of basic concepts like time
axis, event definition and censoring are given. Hazard models are introduced,
with special emphasis on the Cox proportional hazards regression model. We
provide check lists that may be useful both when fitting the model and
assessing its goodness of fit and when interpreting the results. Special
attention is paid to how to avoid problems with immortal time bias by
introducing time-dependent covariates. We discuss prediction based on hazard
models and difficulties when attempting to draw proper causal conclusions from
such models. Finally, we present a series of examples where the methods and
check lists are exemplified. Computational details and implementation using the
freely available R software are documented in Supplementary Material. The paper
was prepared as part of the STRATOS initiative.
| stat |
Selective Inference via Marginal Screening for High Dimensional
Classification | Post-selection inference is a statistical technique for determining salient
variables after model or variable selection. Recently, selective inference, a
kind of post-selection inference framework, has garnered the attention in the
statistics and machine learning communities. By conditioning on a specific
variable selection procedure, selective inference can properly control for
so-called selective type I error, which is a type I error conditional on a
variable selection procedure, without imposing excessive additional
computational costs. While selective inference can provide a valid hypothesis
testing procedure, the main focus has hitherto been on Gaussian linear
regression models. In this paper, we develop a selective inference framework
for binary classification problem. We consider a logistic regression model
after variable selection based on marginal screening, and derive the high
dimensional statistical behavior of the post-selection estimator. This enables
us to asymptotically control for selective type I error for the purposes of
hypothesis testing after variable selection. We conduct several simulation
studies to confirm the statistical power of the test, and compare our proposed
method with data splitting and other methods.
| stat |
Fast Algorithms and Theory for High-Dimensional Bayesian Varying
Coefficient Models | Nonparametric varying coefficient (NVC) models are useful for modeling
time-varying effects on responses that are measured repeatedly. In this paper,
we introduce the nonparametric varying coefficient spike-and-slab lasso
(NVC-SSL) for Bayesian estimation and variable selection in NVC models. The
NVC-SSL simultaneously selects and estimates the significant varying
coefficients, while also accounting for temporal correlations. Our model can be
implemented using a computationally efficient expectation-maximization (EM)
algorithm. We also employ a simple method to make our model robust to
misspecification of the temporal correlation structure. In contrast to
frequentist approaches, little is known about the large-sample properties for
Bayesian NVC models when the dimension of the covariates $p$ grows much faster
than sample size $n$. In this paper, we derive posterior contraction rates for
the NVC-SSL model when $p \gg n$ under both correct specification and
misspecification of the temporal correlation structure. Thus, our results are
derived under weaker assumptions than those seen in other high-dimensional NVC
models which assume independent and identically distributed (iid) random
errors. Finally, we illustrate our methodology through simulation studies and
data analysis. Our method is implemented in the publicly available R package
NVCSSL.
| stat |
Statistical Inference for High-Dimensional Vector Autoregression with
Measurement Error | High-dimensional vector autoregression with measurement error is frequently
encountered in a large variety of scientific and business applications. In this
article, we study statistical inference of the transition matrix under this
model. While there has been a large body of literature studying sparse
estimation of the transition matrix, there is a paucity of inference solutions,
especially in the high-dimensional scenario. We develop inferential procedures
for both the global and simultaneous testing of the transition matrix. We first
develop a new sparse expectation-maximization algorithm to estimate the model
parameters, and carefully characterize their estimation precisions. We then
construct a Gaussian matrix, after proper bias and variance corrections, from
which we derive the test statistics. Finally, we develop the testing procedures
and establish their asymptotic guarantees. We study the finite-sample
performance of our tests through intensive simulations, and illustrate with a
brain connectivity analysis example.
| stat |
Backfitting for large scale crossed random effects regressions | Regression models with crossed random effect errors can be very expensive to
compute. The cost of both generalized least squares and Gibbs sampling can
easily grow as $N^{3/2}$ (or worse) for $N$ observations. Papaspiliopoulos et
al. (2020) present a collapsed Gibbs sampler that costs $O(N)$, but under an
extremely stringent sampling model. We propose a backfitting algorithm to
compute a generalized least squares estimate and prove that it costs $O(N)$. A
critical part of the proof is in ensuring that the number of iterations
required is $O(1)$ which follows from keeping a certain matrix norm below
$1-\delta$ for some $\delta>0$. Our conditions are greatly relaxed compared to
those for the collapsed Gibbs sampler, though still strict. Empirically, the
backfitting algorithm has a norm below $1-\delta$ under conditions that are
less strict than those in our assumptions. We illustrate the new algorithm on a
ratings data set from Stitch Fix.
| stat |
Integrating automated acoustic vocalization data and point count surveys
for estimation of bird abundance | Monitoring wildlife abundance across space and time is an essential task to
study their population dynamics and inform effective management. Acoustic
recording units are a promising technology for efficiently monitoring bird
populations and communities. We present an integrated modeling framework that
combines high-quality but temporally sparse bird point count survey data with
acoustic recordings. Using simulations, we compare the accuracy and precision
of abundance estimates using differing amounts of acoustic vocalizations
obtained from a clustering algorithm, point count data, and a subset of
manually validated acoustic vocalizations. We also use our modeling framework
in a case study to estimate abundance of the Eastern Wood-Pewee (Contopus
virens) in Vermont, U.S.A. The simulation study reveals that combining acoustic
and point count data via an integrated model improves accuracy and precision of
abundance estimates compared with models informed by either acoustic or point
count data alone. Combining acoustic data with only a small number of point
count surveys yields estimates of abundance without the need for validating any
of the identified vocalizations from the acoustic data. Within our case study,
the integrated models provided moderate support for a decline of the Eastern
Wood-Pewee in this region. Our integrated modeling approach combines dense
acoustic data with few point count surveys to deliver reliable estimates of
species abundance without the need for manual identification of acoustic
vocalizations or a prohibitively expensive large number of repeated point count
surveys. Our proposed approach offers an efficient monitoring alternative for
large spatio-temporal regions when point count data are difficult to obtain or
when monitoring is focused on rare species with low detection probability.
| stat |
Revisiting the probabilistic method of record linkage | In theory, the probabilistic linkage method provides two distinct advantages
over non-probabilistic methods, including minimal rates of linkage error and
accurate measures of these rates for data users. However, implementations can
fall short of these expectations either because the conditional independence
assumption is made, or because a model with interactions is used but lacks the
identification property. In official statistics, this is currently the main
challenge to the automated production and use of linked data. To address this
challenge, a new methodology is described for proper linkage problems, where
matched records may be identified with a probability that is bounded away from
zero, regardless of the population size. It models the number of neighbours of
a given record, i.e. the number of resembling records. To be specific, the
proposed model is a finite mixture where each component is the sum of a
Bernoulli variable with an independent Poisson variable. It has the
identification property and yields solutions for many longstanding problems,
including the evaluation of blocking criteria and the estimation of linkage
errors for probabilistic or non-probabilistic linkages, all without clerical
reviews or conditional independence assumptions. Thus it also enables
unsupervised machine learning solutions for record linkage problems.
| stat |
Central Limit Theorems for Classical Multidimensional Scaling | Classical multidimensional scaling is a widely used method in dimensionality
reduction and manifold learning. The method takes in a dissimilarity matrix and
outputs a low-dimensional configuration matrix based on a spectral
decomposition. In this paper, we present three noise models and analyze the
resulting configuration matrices, or embeddings. In particular, we show that
under each of the three noise models the resulting embedding gives rise to a
central limit theorem. We also provide compelling simulations and real data
illustrations of these central limit theorems. This perturbation analysis
represents a significant advancement over previous results regarding classical
multidimensional scaling behavior under randomness.
| stat |
Levels and trends in the sex ratio at birth in seven provinces of Nepal
between 1980 and 2016 with probabilistic projections to 2050: a Bayesian
modeling approach | The sex ratio at birth (SRB; ratio of male to female births) in Nepal has
been reported without imbalance on the national level. However, the national
SRB could mask the disparity within the country. Given the demographic and
cultural heterogeneities in Nepal, it is crucial to model Nepal SRB on the
subnational level. Prior studies on subnational SRB in Nepal are mostly based
on reporting observed values from surveys and census, and no study has provided
probabilistic projections. We aim to estimate and project SRB for the seven
provinces of Nepal from 1980 to 2050 using a Bayesian modeling approach. We
compiled an extensive database on provincial SRB of Nepal, consisting 2001,
2006, 2011, and 2016 Nepal Demographic and Health Surveys and 2011 Census. We
adopted a Bayesian hierarchical time series model to estimate and project the
provincial SRB, with a focus on modelling the potential SRB imbalance. In 2016,
the highest SRB is estimated in Province 5 at 1.102 with a 95% credible
interval (1.044, 1.127) and the lowest SRB is in Province 2 at 1.053 (1.035,
1.109). The SRB imbalance probabilities in all provinces are generally low and
vary from 16% in Province 2 to 81% in Province 5. SRB imbalances are estimated
to have begun at the earliest in 2001 in Province 5 with a 95% credible
interval (1992, 2022) and the latest in 2017 (1998, 2040) in Province 2. We
project SRB in all provinces to begin converging back to the national baseline
in the mid-2030s. Our findings imply that the majority of provinces in Nepal
have low risks of SRB imbalance for the period 1980-2016. However, we identify
a few provinces with higher probabilities of having SRB inflation. The
projected SRB is an important illustration of potential future prenatal sex
discrimination and shows the need to monitor SRB in provinces with higher
possibilities of SRB imbalance.
| stat |
Statistical modeling: the three cultures | Two decades ago, Leo Breiman identified two cultures for statistical
modeling. The data modeling culture (DMC) refers to practices aiming to conduct
statistical inference on one or several quantities of interest. The algorithmic
modeling culture (AMC) refers to practices defining a machine-learning (ML)
procedure that generates accurate predictions about an event of interest.
Breiman argued that statisticians should give more attention to AMC than to
DMC, because of the strengths of ML in adapting to data. While twenty years
later, DMC has lost some of its dominant role in statistics because of the
data-science revolution, we observe that this culture is still the leading
practice in the natural and social sciences. DMC is the modus operandi because
of the influence of the established scientific method, called the
hypothetico-deductive scientific method. Despite the incompatibilities of AMC
with this scientific method, among some research groups, AMC and DMC cultures
mix intensely. We argue that this mixing has formed a fertile spawning pool for
a mutated culture that we called the hybrid modeling culture (HMC) where
prediction and inference have fused into new procedures where they reinforce
one another. This article identifies key characteristics of HMC, thereby
facilitating the scientific endeavor and fueling the evolution of statistical
cultures towards better practices. By better, we mean increasingly reliable,
valid, and efficient statistical practices in analyzing causal relationships.
In combining inference and prediction, the result of HMC is that the
distinction between prediction and inference, taken to its limit, melts away.
We qualify our melting-away argument by describing three HMC practices, where
each practice captures an aspect of the scientific cycle, namely, ML for causal
inference, ML for data acquisition, and ML for theory prediction.
| stat |
Implementing Convex Optimization in R: Two Econometric Examples | Economists specify high-dimensional models to address heterogeneity in
empirical studies with complex big data. Estimation of these models calls for
optimization techniques to handle a large number of parameters. Convex problems
can be effectively executed in modern statistical programming languages. We
complement Koenker and Mizera (2014)'s work on numerical implementation of
convex optimization, with focus on high-dimensional econometric estimators.
Combining R and the convex solver MOSEK achieves faster speed and equivalent
accuracy, demonstrated by examples from Su, Shi, and Phillips (2016) and Shi
(2016). Robust performance of convex optimization is witnessed cross platforms.
The convenience and reliability of convex optimization in R make it easy to
turn new ideas into prototypes.
| stat |
Time-varying exposure history and subsequent health outcomes: a
two-stage approach to identify critical windows | Long-term behavioral and health risk factors constitute a primary focus of
research on the etiology of chronic diseases. Yet, identifying critical
time-windows during which risk factors have the strongest impact on disease
risk is challenging. To assess the trajectory of association of an exposure
history with an outcome, the weighted cumulative exposure index (WCIE) has been
proposed, with weights reflecting the relative importance of exposures at
different times. However, WCIE is restricted to a complete observed error-free
exposure whereas exposures are often measured with intermittent missingness and
error. Moreover, it rarely explores exposure history that is very distant from
the outcome as usually sought in life-course epidemiology. We extend the WCIE
methodology to (i) exposures that are intermittently measured with error, and
(ii) contexts where the exposure time-window precedes the outcome time-window
using a landmark approach. First, the individual exposure history up to the
landmark time is estimated using a mixed model that handles missing data and
error in exposure measurement, and the predicted complete error-free exposure
history is derived. Then the WCIE methodology is applied to assess the
trajectory of association between the predicted exposure history and the health
outcome collected after the landmark time. In our context, the health outcome
is a longitudinal marker analyzed using a mixed model. A simulation study first
demonstrates the correct inference obtained with this approach. Then, applied
to the Nurses' Health Study (19,415 women) to investigate the association
between BMI history (collected from midlife) and subsequent cognitive decline
after age 70. In conclusion, this approach, easy to implement, provides a
flexible tool for studying complex dynamic relationships and identifying
critical time windows while accounting for exposure measurement errors.
| stat |
Adaptive lasso and Dantzig selector for spatial point processes
intensity estimation | Lasso and Dantzig selector are standard procedures able to perform variable
selection and estimation simultaneously. This paper is concerned with extending
these procedures to spatial point process intensity estimation. We propose
adaptive versions of these procedures, develop efficient computational
methodologies and derive asymptotic results for a large class of spatial point
processes under the setting where the number of parameters, i.e. the number of
spatial covariates considered, increases with the volume of the observation
domain. Both procedures are compared theoretically and in a simulation study.
| stat |
Trustworthiness of statistical inference | We examine the role of trustworthiness and trust in statistical inference,
arguing that it is the extent of trustworthiness in inferential statistical
tools which enables trust in the conclusions. Certain tools, such as the
p-value and significance test, have recently come under renewed criticism, with
some arguing that they damage trust in statistics. We argue the contrary,
beginning from the position that the central role of these methods is to form
the basis for trusted conclusions in the face of uncertainty in the data, and
noting that it is the misuse and misunderstanding of these tools which damages
trustworthiness and hence trust. We go on to argue that recent calls to ban
these tools would tackle the symptom, not the cause, and themselves risk
damaging the capability of science to advance, and feeding into public
suspicion of the discipline of statistics. The consequence could be aggravated
mistrust of our discipline and of science more generally. In short, the very
proposals could work in quite the contrary direction from that intended. We
make some alternative proposals for tackling the misuse and misunderstanding of
these methods, and for how trust in our discipline might be promoted.
| stat |
Doubly Sparse Variational Gaussian Processes | The use of Gaussian process models is typically limited to datasets with a
few tens of thousands of observations due to their complexity and memory
footprint. The two most commonly used methods to overcome this limitation are
1) the variational sparse approximation which relies on inducing points and 2)
the state-space equivalent formulation of Gaussian processes which can be seen
as exploiting some sparsity in the precision matrix. We propose to take the
best of both worlds: we show that the inducing point framework is still valid
for state space models and that it can bring further computational and memory
savings. Furthermore, we provide the natural gradient formulation for the
proposed variational parameterisation. Finally, this work makes it possible to
use the state-space formulation inside deep Gaussian process models as
illustrated in one of the experiments.
| stat |
A causal framework for distribution generalization | We consider the problem of predicting a response from a set of covariates
when the test distribution may differ from the training distribution. Motivated
by the idea that such differences may have causal explanations, we consider a
class of test distributions that emerge from interventions in a structural
causal model, and focus on minimizing the worst-case risk over this class.
Causal regression models, which regress the response variable on all of its
direct parents, remain valid under arbitrary interventions on any subset of
covariates, but they are not always optimal in the above sense. For example, in
linear models, for a set of interventions with bounded strength, alternative
solutions have been shown to be minimax prediction optimal. We introduce the
formal framework of distribution generalization that allows us to analyze the
above problem in partially observed nonlinear models for both direct and
indirect interventions on the covariates. It takes into account that, in
practice, minimax solutions need to be identified from observational data. Our
framework allows us to characterize under which class of interventions the
causal function is minimax optimal. We prove several sufficient conditions for
distribution generalization and present corresponding impossibility results. We
propose a practical method, called NILE, that achieves distribution
generalization in a nonlinear instrumental variables setting with linear
extrapolation. We prove consistency and present empirical results.
| stat |
Martingale Posterior Distributions | The prior distribution on parameters of a likelihood is the usual starting
point for Bayesian uncertainty quantification. In this paper, we present a
different perspective. Given a finite data sample $Y_{1:n}$ of size $n$ from an
infinite population, we focus on the missing $Y_{n+1:\infty}$ as the source of
statistical uncertainty, with the parameter of interest being known precisely
given $Y_{1:\infty}$. We argue that the foundation of Bayesian inference is to
assign a predictive distribution on $Y_{n+1:\infty}$ conditional on $Y_{1:n}$,
which then induces a distribution on the parameter of interest. Demonstrating
an application of martingales, Doob shows that choosing the Bayesian predictive
distribution returns the conventional posterior as the distribution of the
parameter. Taking this as our cue, we relax the predictive machine, avoiding
the need for the predictive to be derived solely from the usual prior to
posterior to predictive density formula. We introduce the martingale posterior
distribution, which returns Bayesian uncertainty directly on any statistic of
interest without the need for the likelihood and prior, and this distribution
can be sampled through a computational scheme we name predictive resampling. To
that end, we introduce new predictive methodologies for multivariate density
estimation, regression and classification that build upon recent work on
bivariate copulas.
| stat |
(Almost) All of Entity Resolution | Whether the goal is to estimate the number of people that live in a
congressional district, to estimate the number of individuals that have died in
an armed conflict, or to disambiguate individual authors using bibliographic
data, all these applications have a common theme - integrating information from
multiple sources. Before such questions can be answered, databases must be
cleaned and integrated in a systematic and accurate way, commonly known as
record linkage, de-duplication, or entity resolution. In this article, we
review motivational applications and seminal papers that have led to the growth
of this area. Specifically, we review the foundational work that began in the
1940's and 50's that have led to modern probabilistic record linkage. We review
clustering approaches to entity resolution, semi- and fully supervised methods,
and canonicalization, which are being used throughout industry and academia in
applications such as human rights, official statistics, medicine, citation
networks, among others. Finally, we discuss current research topics of
practical importance.
| stat |
Essentially No Barriers in Neural Network Energy Landscape | Training neural networks involves finding minima of a high-dimensional
non-convex loss function. Knowledge of the structure of this energy landscape
is sparse. Relaxing from linear interpolations, we construct continuous paths
between minima of recent neural network architectures on CIFAR10 and CIFAR100.
Surprisingly, the paths are essentially flat in both the training and test
landscapes. This implies that neural networks have enough capacity for
structural changes, or that these changes are small between minima. Also, each
minimum has at least one vanishing Hessian eigenvalue in addition to those
resulting from trivial invariance.
| stat |
Particle-based, online estimation of tangent filters with application to
parameter estimation in nonlinear state-space models | This paper presents a novel algorithm for efficient online estimation of the
filter derivatives in general hidden Markov models. The algorithm, which has a
linear computational complexity and very limited memory requirements, is
furnished with a number of convergence results, including a central limit
theorem with an asymptotic variance that can be shown to be uniformly bounded
in time. Using the proposed filter derivative estimator we design a recursive
maximum likelihood algorithm updating the parameters according the gradient of
the one-step predictor log-likelihood. The efficiency of this online parameter
estimation scheme is illustrated in a simulation study.
| stat |
Scalable inference of topic evolution via models for latent geometric
structures | We develop new models and algorithms for learning the temporal dynamics of
the topic polytopes and related geometric objects that arise in topic model
based inference. Our model is nonparametric Bayesian and the corresponding
inference algorithm is able to discover new topics as the time progresses. By
exploiting the connection between the modeling of topic polytope evolution,
Beta-Bernoulli process and the Hungarian matching algorithm, our method is
shown to be several orders of magnitude faster than existing topic modeling
approaches, as demonstrated by experiments working with several million
documents in under two dozens of minutes.
| stat |
Seeded intervals and noise level estimation in change point detection: A
discussion of Fryzlewicz (2020) | In this discussion, we compare the choice of seeded intervals and that of
random intervals for change point segmentation from practical, statistical and
computational perspectives. Furthermore, we investigate a novel estimator of
the noise level, which improves many existing model selection procedures
(including the steepest drop to low levels), particularly for challenging
frequent change point scenarios with low signal-to-noise ratios.
| stat |
A General Bayesian Model for Heteroskedastic Data with Fully Conjugate
Full-Conditional Distributions | Models for heteroskedastic data are relevant in a wide variety of
applications ranging from financial time series to environmental statistics.
However, the topic of modeling the variance function conditionally has not seen
near as much attention as modeling the mean. Volatility models have been used
in specific applications, but these models can be difficult to fit in a
Bayesian setting due to posterior distributions that are challenging to sample
from efficiently. In this work, we introduce a general model for
heteroskedastic data. This approach models the conditional variance in a mixed
model approach as a function of any desired covariates or random effects. We
rely on new distribution theory in order to construct priors that yield fully
conjugate full conditional distributions. Thus, our approach can easily be fit
via Gibbs sampling. Furthermore, we extend the model to a deep learning
approach that can provide highly accurate estimates for time dependent data. We
also provide an extension for heavy-tailed data. We illustrate our methodology
via three applications. The first application utilizes a high dimensional soil
dataset with inherent spatial dependence. The second application involves
modeling of asset volatility. The third application focuses on clinical trial
data for creatinine.
| stat |
Generalized probabilistic principal component analysis of correlated
data | Principal component analysis (PCA) is a well-established tool in machine
learning and data processing. The principal axes in PCA were shown to be
equivalent to the maximum marginal likelihood estimator of the factor loading
matrix in a latent factor model for the observed data, assuming that the latent
factors are independently distributed as standard normal distributions.
However, the independence assumption may be unrealistic for many scenarios such
as modeling multiple time series, spatial processes, and functional data, where
the outcomes are correlated. In this paper, we introduce the generalized
probabilistic principal component analysis (GPPCA) to study the latent factor
model for multiple correlated outcomes, where each factor is modeled by a
Gaussian process. Our method generalizes the previous probabilistic formulation
of PCA (PPCA) by providing the closed-form maximum marginal likelihood
estimator of the factor loadings and other parameters. Based on the explicit
expression of the precision matrix in the marginal likelihood that we derived,
the number of the computational operations is linear to the number of output
variables. Furthermore, we also provide the closed-form expression of the
marginal likelihood when other covariates are included in the mean structure.
We highlight the advantage of GPPCA in terms of the practical relevance,
estimation accuracy and computational convenience. Numerical studies of
simulated and real data confirm the excellent finite-sample performance of the
proposed approach.
| stat |
Stratified sampling and bootstrapping for approximate Bayesian
computation | Approximate Bayesian computation (ABC) is computationally intensive for
complex model simulators. To exploit expensive simulations, data-resampling via
bootstrapping was used with success in Everitt (2017) to obtain many artificial
datasets at little cost and construct a synthetic likelihood. When using the
same approach within ABC to produce a pseudo-marginal ABC-MCMC algorithm, the
posterior variance is inflated, thus producing biased posterior inference. Here
we use stratified Monte Carlo to considerably reduce the bias induced by data
resampling. We also show that it is possible to obtain reliable inference using
a larger than usual ABC threshold. Finally, we show that with stratified Monte
Carlo we obtain a less variable ABC likelihood. We consider simulation studies
for static (Gaussian, g-and-k distribution, Ising model) and dynamic models
(Lotka-Volterra). For the Lotka-Volterra case study, we compare our results
against a standard pseudo-Marginal ABC and find that our approach is four times
more efficient and, given limited computational budget, it explores the
posterior surface more thoroughly. A comparison against state-of-art sequential
Monte Carlo ABC is also reported.
| stat |
Unbiased approximation of posteriors via coupled particle Markov chain
Monte Carlo | Markov chain Monte Carlo (MCMC) is a powerful methodology for the
approximation of posterior distributions. However, the iterative nature of MCMC
does not naturally facilitate its use with modern highly parallelisable
computation on HPC and cloud environments. Another concern is the
identification of the bias and Monte Carlo error of produced averages. The
above have prompted the recent development of fully (`embarrassingly')
parallelisable unbiased Monte Carlo methodology based on couplings of MCMC
algorithms. A caveat is that formulation of effective couplings is typically
not trivial and requires model-specific technical effort. We propose couplings
of sequential Monte Carlo (SMC) by considering adaptive SMC to approximate
complex, high-dimensional posteriors combined with recent advances in unbiased
estimation for state-space models. Coupling is then achieved at the SMC level
and is, in general, not problem-specific. The resulting methodology enjoys
desirable theoretical properties. We illustrate the effectiveness of the
algorithm via application to two statistical models in high dimensions: (i)
horseshoe regression; (ii) Gaussian graphical models.
| stat |
Community detection and percolation of information in a geometric
setting | We make the first steps towards generalizing the theory of stochastic block
models, in the sparse regime, towards a model where the discrete community
structure is replaced by an underlying geometry. We consider a geometric random
graph over a homogeneous metric space where the probability of two vertices to
be connected is an arbitrary function of the distance. We give sufficient
conditions under which the locations can be recovered (up to an isomorphism of
the space) in the sparse regime. Moreover, we define a geometric counterpart of
the model of flow of information on trees, due to Mossel and Peres, in which
one considers a branching random walk on a sphere and the goal is to recover
the location of the root based on the locations of leaves. We give some
sufficient conditions for percolation and for non-percolation of information in
this model.
| stat |
On the Optimization of Approximate Control Variates with Parametrically
Defined Estimators | Multi-model Monte Carlo methods, such as multi-level Monte Carlo (MLMC) and
multifidelity Monte Carlo (MFMC), allow for efficient estimation of the
expectation of a quantity of interest given a set of models of varying
fidelities. Recently, it was shown that the MLMC and MFMC estimators are both
instances of the approximate control variates (ACV) framework [Gorodetsky et
al. 2020]. In that same work, it was also shown that hand-tailored ACV
estimators could outperform MLMC and MFMC for a variety of model scenarios.
Because there is no reason to believe that these hand-tailored estimators are
the best among a myriad of possible ACV estimators, a more general approach to
estimator construction is pursued in this work. First, a general form of the
ACV estimator variance is formulated. Then, the formulation is utilized to
generate parametrically-defined estimators. These parametrically-defined
estimators allow for an optimization to be pursued over a larger domain of
possible ACV estimators. The parametrically-defined estimators are tested on a
large set of model scenarios, and it is found that the broader search domain
enabled by parametrically-defined estimators leads to greater variance
reduction.
| stat |
Simulations for a Q statistic with constant weights to assess
heterogeneity in meta-analysis of mean difference | A variety of problems in random-effects meta-analysis arise from the
conventional $Q$ statistic, which uses estimated inverse-variance (IV) weights.
In previous work on standardized mean difference and log-odds-ratio, we found
superior performance with an estimator of the overall effect whose weights use
only group-level sample sizes. The $Q$ statistic with those weights has the
form proposed by DerSimonian and Kacker. The distribution of this $Q$ and the
$Q$ with IV weights must generally be approximated. We investigate
approximations for those distributions, as a basis for testing and estimating
the between-study variance ($\tau^2$). Some approximations require the variance
and third moment of $Q$, which we derive. We describe the design and results of
a simulation study, with mean difference as the effect measure, which provides
a framework for assessing accuracy of the approximations, level and power of
the tests, and bias in estimating $\tau^2$. Use of $Q$ with sample-size-based
weights and its exact distribution (available for mean difference and evaluated
by Farebrother's algorithm) provides precise levels even for very small and
unbalanced sample sizes. The corresponding estimator of $\tau^2$ is almost
unbiased for 10 or more small studies. Under these circumstances this
performance compares favorably with the extremely liberal behavior of the
standard tests of heterogeneity and the largely biased estimators based on
inverse-variance weights.
| stat |
A Gaussian Process Model of Cross-Category Dynamics in Brand Choice | Understanding individual customers' sensitivities to prices, promotions,
brand, and other aspects of the marketing mix is fundamental to a wide swath of
marketing problems, including targeting and pricing. Companies that operate
across many product categories have a unique opportunity, insofar as they can
use purchasing data from one category to augment their insights in another.
Such cross-category insights are especially crucial in situations where
purchasing data may be rich in one category, and scarce in another. An
important aspect of how consumers behave across categories is dynamics:
preferences are not stable over time, and changes in individual-level
preference parameters in one category may be indicative of changes in other
categories, especially if those changes are driven by external factors. Yet,
despite the rich history of modeling cross-category preferences, the marketing
literature lacks a framework that flexibly accounts for \textit{correlated
dynamics}, or the cross-category interlinkages of individual-level sensitivity
dynamics. In this work, we propose such a framework, leveraging
individual-level, latent, multi-output Gaussian processes to build a
nonparametric Bayesian choice model that allows information sharing of
preference parameters across customers, time, and categories. We apply our
model to grocery purchase data, and show that our model detects interesting
dynamics of customers' price sensitivities across multiple categories.
Managerially, we show that capturing correlated dynamics yields substantial
predictive gains, relative to benchmarks. Moreover, we find that capturing
correlated dynamics can have implications for understanding changes in
consumers preferences over time, and developing targeted marketing strategies
based on those dynamics.
| stat |
An efficient ADMM algorithm for high dimensional precision matrix
estimation via penalized quadratic loss | The estimation of high dimensional precision matrices has been a central
topic in statistical learning. However, as the number of parameters scales
quadratically with the dimension $p$, many state-of-the-art methods do not
scale well to solve problems with a very large $p$. In this paper, we propose a
very efficient algorithm for precision matrix estimation via penalized
quadratic loss functions. Under the high dimension low sample size setting, the
computation complexity of our algorithm is linear in both the sample size and
the number of parameters. Such a computation complexity is in some sense
optimal, as it is the same as the complexity needed for computing the sample
covariance matrix. Numerical studies show that our algorithm is much more
efficient than other state-of-the-art methods when the dimension $p$ is very
large.
| stat |
The Price of Local Fairness in Multistage Selection | The rise of algorithmic decision making led to active researches on how to
define and guarantee fairness, mostly focusing on one-shot decision making. In
several important applications such as hiring, however, decisions are made in
multiple stage with additional information at each stage. In such cases,
fairness issues remain poorly understood.
In this paper we study fairness in $k$-stage selection problems where
additional features are observed at every stage. We first introduce two
fairness notions, local (per stage) and global (final stage) fairness, that
extend the classical fairness notions to the $k$-stage setting. We propose a
simple model based on a probabilistic formulation and show that the locally and
globally fair selections that maximize precision can be computed via a linear
program. We then define the price of local fairness to measure the loss of
precision induced by local constraints; and investigate theoretically and
empirically this quantity. In particular, our experiments show that the price
of local fairness is generally smaller when the sensitive attribute is observed
at the first stage; but globally fair selections are more locally fair when the
sensitive attribute is observed at the second stage---hence in both cases it is
often possible to have a selection that has a small price of local fairness and
is close to locally fair.
| stat |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.