title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Fair Bayesian Optimization | Given the increasing importance of machine learning (ML) in our lives,
several algorithmic fairness techniques have been proposed to mitigate biases
in the outcomes of the ML models. However, most of these techniques are
specialized to cater to a single family of ML models and a specific definition
of fairness, limiting their adaptibility in practice. We introduce a general
constrained Bayesian optimization (BO) framework to optimize the performance of
any ML model while enforcing one or multiple fairness constraints. BO is a
model-agnostic optimization method that has been successfully applied to
automatically tune the hyperparameters of ML models. We apply BO with fairness
constraints to a range of popular models, including random forests, gradient
boosting, and neural networks, showing that we can obtain accurate and fair
solutions by acting solely on the hyperparameters. We also show empirically
that our approach is competitive with specialized techniques that enforce
model-specific fairness constraints, and outperforms preprocessing methods that
learn fair representations of the input data. Moreover, our method can be used
in synergy with such specialized fairness techniques to tune their
hyperparameters. Finally, we study the relationship between fairness and the
hyperparameters selected by BO. We observe a correlation between regularization
and unbiased models, explaining why acting on the hyperparameters leads to ML
models that generalize well and are fair.
| stat |
Survival Cluster Analysis | Conventional survival analysis approaches estimate risk scores or
individualized time-to-event distributions conditioned on covariates. In
practice, there is often great population-level phenotypic heterogeneity,
resulting from (unknown) subpopulations with diverse risk profiles or survival
distributions. As a result, there is an unmet need in survival analysis for
identifying subpopulations with distinct risk profiles, while jointly
accounting for accurate individualized time-to-event predictions. An approach
that addresses this need is likely to improve characterization of individual
outcomes by leveraging regularities in subpopulations, thus accounting for
population-level heterogeneity. In this paper, we propose a Bayesian
nonparametrics approach that represents observations (subjects) in a clustered
latent space, and encourages accurate time-to-event predictions and clusters
(subpopulations) with distinct risk profiles. Experiments on real-world
datasets show consistent improvements in predictive performance and
interpretability relative to existing state-of-the-art survival analysis
models.
| stat |
Evaluating the Discrimination Ability of Proper Multivariate Scoring
Rules | Proper scoring rules are commonly applied to quantify the accuracy of
distribution forecasts. Given an observation they assign a scalar score to each
distribution forecast, with the the lowest expected score attributed to the
true distribution. The energy and variogram scores are two rules that have
recently gained some popularity in multivariate settings because their
computation does not require a forecast to have parametric density function and
so they are broadly applicable. Here we conduct a simulation study to compare
the discrimination ability between the energy score and three variogram scores.
Compared with other studies, our simulation design is more realistic because it
is supported by a historical data set containing commodity prices, currencies
and interest rates, and our data generating processes include a diverse
selection of models with different marginal distributions, dependence
structure, and calibration windows. This facilitates a comprehensive comparison
of the performance of proper scoring rules in different settings. To compare
the scores we use three metrics: the mean relative score, error rate and a
generalised discrimination heuristic. Overall, we find that the variogram score
with parameter p=0.5 outperforms the energy score and the other two variogram
scores.
| stat |
Transporting stochastic direct and indirect effects to new populations | Transported mediation effects may contribute to understanding how and why
interventions may work differently when applied to new populations. However, we
are not aware of any estimators for such effects. Thus, we propose several
different estimators of transported stochastic direct and indirect effects: an
inverse-probability of treatment stabilized weighted estimator, a doubly robust
estimator that solves the estimating equation, and a doubly robust substitution
estimator in the targeted minimum loss-based framework. We demonstrate their
finite sample properties in a simulation study.
| stat |
Smoothly Adaptively Centered Ridge Estimator | With a focus on linear models with smooth functional covariates, we propose a
penalization framework (SACR) based on the nonzero centered ridge, where the
center of the penalty is optimally reweighted in a supervised way, starting
from the ordinary ridge solution as the initial centerfunction. In particular,
we introduce a convex formulation that jointly estimates the model's
coefficients and the weight function, with a roughness penalty on the
centerfunction and constraints on the weights in order to recover a possibly
smooth and/or sparse solution. This allows for a non-iterative and continuous
variable selection mechanism, as the weight function can either inflate or
deflate the initial center, in order to target the penalty towards a suitable
center, with the objective to reduce the unwanted shrinkage on the nonzero
coefficients, instead of uniformly shrinking the whole coefficient function. As
empirical evidence of the interpretability and predictive power of our method,
we provide a simulation study and two real world spectroscopy applications with
both classification and regression.
| stat |
Split-BOLFI for for misspecification-robust likelihood free inference in
high dimensions | Likelihood-free inference for simulator-based statistical models has recently
grown rapidly from its infancy to a useful tool for practitioners. However,
models with more than a very small number of parameters as the target of
inference have remained an enigma, in particular for the approximate Bayesian
computation (ABC) community. To advance the possibilities for performing
likelihood-free inference in high-dimensional parameter spaces, here we
introduce an extension of the popular Bayesian optimisation based approach to
approximate discrepancy functions in a probabilistic manner which lends itself
to an efficient exploration of the parameter space. Our method achieves
computational scalability by using separate acquisition procedures for the
discrepancies defined for different parameters. These efficient
high-dimensional simulation acquisitions are combined with exponentiated
loss-likelihoods to provide a misspecification-robust characterisation of the
marginal posterior distribution for all model parameters. The method
successfully performs computationally efficient inference in a 100-dimensional
space on canonical examples and compares favourably to existing Copula-ABC
methods. We further illustrate the potential of this approach by fitting a
bacterial transmission dynamics model to daycare centre data, which provides
biologically coherent results on the strain competition in a 30-dimensional
parameter space.
| stat |
Convex Covariate Clustering for Classification | Clustering, like covariate selection for classification, is an important step
to compress and interpret the data. However, clustering of covariates is often
performed independently of the classification step, which can lead to
undesirable clustering results that harm interpretability and compression rate.
Therefore, we propose a method that can cluster covariates while taking into
account class label information of samples. We formulate the problem as a
convex optimization problem which uses both, a-priori similarity information
between covariates, and information from class-labeled samples. Like ordinary
convex clustering [Chi and Lange, 2015], the proposed method offers a unique
global minima making it insensitive to initialization. In order to solve the
convex problem, we propose a specialized alternating direction method of
multipliers (ADMM), which scales up to several thousands of variables.
Furthermore, in order to circumvent computationally expensive cross-validation,
we propose a model selection criterion based on approximating the marginal
likelihood. Experiments on synthetic and real data confirm the usefulness of
the proposed clustering method and the selection criterion.
| stat |
Nonparametric regression for multiple heterogeneous networks | We study nonparametric methods for the setting where multiple distinct
networks are observed on the same set of nodes. Such samples may arise in the
form of replicated networks drawn from a common distribution, or in the form of
heterogeneous networks, with the network generating process varying from one
network to another, e.g.~dynamic and cross-sectional networks. Nonparametric
methods for undirected networks have focused on estimation of the graphon
model. While the graphon model accounts for nodal heterogeneity, it does not
account for network heterogeneity, a feature specific to applications where
multiple networks are observed. To address this setting of multiple networks,
we propose a multi-graphon model which allows node-level as well as
network-level heterogeneity. We show how information from multiple networks can
be leveraged to enable estimation of the multi-graphon via standard
nonparametric regression techniques, e.g. kernel regression, orthogonal series
estimation. We study theoretical properties of the proposed estimator
establishing recovery of the latent nodal positions up to negligible error, and
convergence of the multi-graphon estimator to the normal distribution. Finite
sample performance are investigated in a simulation study and application to
two real-world networks---a dynamic contact network of ants and a collection of
structural brain networks from different subjects---illustrate the utility of
our approach.
| stat |
Adaptive Sampling for Estimating Multiple Probability Distributions | We consider the problem of allocating samples to a finite set of discrete
distributions in order to learn them uniformly well in terms of four common
distance measures: $\ell_2^2$, $\ell_1$, $f$-divergence, and separation
distance. To present a unified treatment of these distances, we first propose a
general optimistic tracking algorithm and analyze its sample allocation
performance w.r.t.~an oracle. We then instantiate this algorithm for the four
distance measures and derive bounds on the regret of their resulting allocation
schemes. We verify our theoretical findings through some experiments. Finally,
we show that the techniques developed in the paper can be easily extended to
the related setting of minimizing the average error (in terms of the four
distances) in learning a set of distributions.
| stat |
Robust Optimal Design when Missing Data Happen at Random | In this article, we investigate the robust optimal design problem for the
prediction of response when the fitted regression models are only approximately
specified, and observations might be missing completely at random. The
intuitive idea is as follows: We assume that data are missing at random, and
the complete case analysis is applied. To account for the occurrence of missing
data, the design criterion we choose is the mean, for the missing indicator, of
the averaged (over the design space) mean squared errors of the predictions. To
describe the uncertainty in the specification of the real underlying model, we
impose a neighborhood structure on the regression response and maximize,
analytically, the \textbf{M}ean of the averaged \textbf{M}ean squared
\textbf{P}rediction \textbf{E}rrors (MMPE), over the entire neighborhood. The
maximized MMPE is the ``worst'' loss in the neighborhood of the fitted
regression model. Minimizing the maximum MMPE over the class of designs, we
obtain robust ``minimax'' designs. The robust designs constructed afford
protection from increases in prediction errors resulting from model
misspecifications.
| stat |
Reconsidering Analytical Variational Bounds for Output Layers of Deep
Networks | The combination of the re-parameterization trick with the use of variational
auto-encoders has caused a sensation in Bayesian deep learning, allowing the
training of realistic generative models of images and has considerably
increased our ability to use scalable latent variable models. The
re-parameterization trick is necessary for models in which no analytical
variational bound is available and allows noisy gradients to be computed for
arbitrary models. However, for certain standard output layers of a neural
network, analytical bounds are available and the variational auto-encoder may
be used both without the re-parameterization trick or the need for any Monte
Carlo approximation. In this work, we show that using Jaakola and Jordan bound,
we can produce a binary classification layer that allows a Bayesian output
layer to be trained, using the standard stochastic gradient descent algorithm.
We further demonstrate that a latent variable model utilizing the Bouchard
bound for multi-class classification allows for fast training of a fully
probabilistic latent factor model, even when the number of classes is very
large.
| stat |
An iterative algorithm for joint covariate and random effect selection
in mixed effects models | We consider joint selection of fixed and random effects in general
mixed-effects models. The interpretation of estimated mixed-effects models is
challenging since changing the structure of one set of effects can lead to
different choices of important covariates in the model. We propose a stepwise
selection algorithm to perform simultaneous selection of the fixed and random
effects. It is based on BIC-type criteria whose penalties are adapted to
mixed-effects models. The proposed procedure performs model selection in both
linear and nonlinear models. It should be used in the low-dimension setting
where the number of covariates and the number of random effects are moderate
with respect to the total number of observations. The performance of the
algorithm is assessed via a simulation study, that includes also a comparative
study with alternatives when available in the literature. The use of the method
is illustrated in the clinical study of an antibiotic agent kinetics.
| stat |
Statistically Robust Neural Network Classification | Recently there has been much interest in quantifying the robustness of neural
network classifiers through adversarial risk metrics. However, for problems
where test-time corruptions occur in a probabilistic manner, rather than being
generated by an explicit adversary, adversarial metrics typically do not
provide an accurate or reliable indicator of robustness. To address this, we
introduce a statistically robust risk (SRR) framework which measures robustness
in expectation over both network inputs and a corruption distribution. Unlike
many adversarial risk metrics, which typically require separate applications on
a point-by-point basis, the SRR can easily be directly estimated for an entire
network and used as a training objective in a stochastic gradient scheme.
Furthermore, we show both theoretically and empirically that it can scale to
higher-dimensional networks by providing superior generalization performance
compared with comparable adversarial risks.
| stat |
Optimal Distributed Subsampling for Maximum Quasi-Likelihood Estimators
with Massive Data | Nonuniform subsampling methods are effective to reduce computational burden
and maintain estimation efficiency for massive data. Existing methods mostly
focus on subsampling with replacement due to its high computational efficiency.
If the data volume is so large that nonuniform subsampling probabilities cannot
be calculated all at once, then subsampling with replacement is infeasible to
implement. This paper solves this problem using Poisson subsampling. We first
derive optimal Poisson subsampling probabilities in the context of
quasi-likelihood estimation under the A- and L-optimality criteria. For a
practically implementable algorithm with approximated optimal subsampling
probabilities, we establish the consistency and asymptotic normality of the
resultant estimators. To deal with the situation that the full data are stored
in different blocks or at multiple locations, we develop a distributed
subsampling framework, in which statistics are computed simultaneously on
smaller partitions of the full data. Asymptotic properties of the resultant
aggregated estimator are investigated. We illustrate and evaluate the proposed
strategies through numerical experiments on simulated and real data sets.
| stat |
Achieving Fairness via Post-Processing in Web-Scale Recommender Systems | Building fair recommender systems is a challenging and extremely important
area of study due to its immense impact on society. We focus on two commonly
accepted notions of fairness for machine learning models powering such
recommender systems, namely equality of opportunity and equalized odds. These
measures of fairness make sure that equally "qualified" (or "unqualified")
candidates are treated equally regardless of their protected attribute status
(such as gender or race). In this paper, we propose scalable methods for
achieving equality of opportunity and equalized odds in rankings in the
presence of position bias, which commonly plagues data generated from
recommendation systems. Our algorithms are model agnostic in the sense that
they depend only on the final scores provided by a model, making them easily
applicable to virtually all web-scale recommender systems. We conduct extensive
simulations as well as real-world experiments to show the efficacy of our
approach.
| stat |
Neural Empirical Bayes: Source Distribution Estimation and its
Applications to Simulation-Based Inference | We revisit empirical Bayes in the absence of a tractable likelihood function,
as is typical in scientific domains relying on computer simulations. We
investigate how the empirical Bayesian can make use of neural density
estimators first to use all noise-corrupted observations to estimate a prior or
source distribution over uncorrupted samples, and then to perform
single-observation posterior inference using the fitted source distribution. We
propose an approach based on the direct maximization of the log-marginal
likelihood of the observations, examining both biased and de-biased estimators,
and comparing to variational approaches. We find that, up to symmetries, a
neural empirical Bayes approach recovers ground truth source distributions.
With the learned source distribution in hand, we show the applicability to
likelihood-free inference and examine the quality of the resulting posterior
estimates. Finally, we demonstrate the applicability of Neural Empirical Bayes
on an inverse problem from collider physics.
| stat |
Principal Boundary on Riemannian Manifolds | We consider the classification problem and focus on nonlinear methods for
classification on manifolds. For multivariate datasets lying on an embedded
nonlinear Riemannian manifold within the higher-dimensional ambient space, we
aim to acquire a classification boundary for the classes with labels, using the
intrinsic metric on the manifolds. Motivated by finding an optimal boundary
between the two classes, we invent a novel approach -- the principal boundary.
From the perspective of classification, the principal boundary is defined as an
optimal curve that moves in between the principal flows traced out from two
classes of data, and at any point on the boundary, it maximizes the margin
between the two classes. We estimate the boundary in quality with its
direction, supervised by the two principal flows. We show that the principal
boundary yields the usual decision boundary found by the support vector machine
in the sense that locally, the two boundaries coincide. Some optimality and
convergence properties of the random principal boundary and its population
counterpart are also shown. We illustrate how to find, use and interpret the
principal boundary with an application in real data.
| stat |
Controlling the EWMA S^2 control chart false alarm behavior when the
in-control variance level must be estimated | Investigating the problem of setting control limits in the case of parameter
uncertainty is more accessible when monitoring the variance because only one
parameter has to be estimated. Simply ignoring the induced uncertainty
frequently leads to control charts with poor false alarm performances.
Adjusting the unconditional in-control (IC) average run length (ARL) makes the
situation even worse. Guaranteeing a minimum conditional IC ARL with some given
probability is another very popular approach to solving these difficulties.
However, it is very conservative as well as more complex and more difficult to
communicate. We utilize the probability of a false alarm within the planned
number of points to be plotted on the control chart. It turns out that
adjusting this probability produces notably different limit adjustments
compared to controlling the unconditional IC ARL. We then develop numerical
algorithms to determine the respective modifications of the upper and two-sided
exponentially weighted moving average (EWMA) charts based on the sample
variance for normally distributed data. These algorithms are made available
within an R package. Finally, the impacts of the EWMA smoothing constant and
the size of the preliminary sample on the control chart design and its
performance are studied.
| stat |
Robust Boosting for Regression Problems | Gradient boosting algorithms construct a regression predictor using a linear
combination of ``base learners''. Boosting also offers an approach to obtaining
robust non-parametric regression estimators that are scalable to applications
with many explanatory variables. The robust boosting algorithm is based on a
two-stage approach, similar to what is done for robust linear regression: it
first minimizes a robust residual scale estimator, and then improves it by
optimizing a bounded loss function. Unlike previous robust boosting proposals
this approach does not require computing an ad-hoc residual scale estimator in
each boosting iteration. Since the loss functions involved in this robust
boosting algorithm are typically non-convex, a reliable initialization step is
required, such as an L1 regression tree, which is also fast to compute. A
robust variable importance measure can also be calculated via a permutation
procedure. Thorough simulation studies and several data analyses show that,
when no atypical observations are present, the robust boosting approach works
as well as the standard gradient boosting with a squared loss. Furthermore,
when the data contain outliers, the robust boosting estimator outperforms the
alternatives in terms of prediction error and variable selection accuracy.
| stat |
Compositional Abstraction Error and a Category of Causal Models | Interventional causal models describe joint distributions over some variables
used to describe a system, one for each intervention setting. They provide a
formal recipe for how to move between joint distributions and make predictions
about the variables upon intervening on the system. Yet, it is difficult to
formalise how we may change the underlying variables used to describe the
system, say from fine-grained to coarse-grained variables. Here, we argue that
compositionality is a desideratum for model transformations and the associated
errors. We develop a framework for model transformations and abstractions with
a notion of error that is compositional: when abstracting a reference model M
modularly, first obtaining M' and then further simplifying that to obtain M'',
then the composite transformation from M to M'' exists and its error can be
bounded by the errors incurred by each individual transformation step. Category
theory, the study of mathematical objects via the compositional transformations
between them, offers a natural language for developing our framework. We
introduce a category of finite interventional causal models and, leveraging
theory of enriched categories, prove that our framework enjoys the desired
compositionality properties.
| stat |
Testing One Hypothesis Multiple Times: The Multidimensional Case | The identification of new rare signals in data, the detection of a sudden
change in a trend, and the selection of competing models, are among the most
challenging problems in statistical practice. These challenges can be tackled
using a test of hypothesis where a nuisance parameter is present only under the
alternative, and a computationally efficient solution can be obtained by the
"Testing One Hypothesis Multiple times" (TOHM) method. In the one-dimensional
setting, a fine discretization of the space of the non-identifiable parameter
is specified, and a global p-value is obtained by approximating the
distribution of the supremum of the resulting stochastic process. In this
paper, we propose a computationally efficient inferential tool to perform TOHM
in the multidimensional setting. Here, the approximations of interest typically
involve the expected Euler Characteristics (EC) of the excursion set of the
underlying random field. We introduce a simple algorithm to compute the EC in
multiple dimensions and for arbitrary large significance levels. This leads to
an highly generalizable computational tool to perform inference under
non-standard regularity conditions.
| stat |
Comparing Covariate Prioritization via Matching to Machine Learning
Methods for Causal Inference using Five Empirical Applications | When investigators seek to estimate causal effects, they often assume that
selection into treatment is based only on observed covariates. Under this
identification strategy, analysts must adjust for observed confounders. While
basic regression models have long been the dominant method of statistical
adjustment, more robust methods based on matching or weighting have become more
common. Of late, even more flexible methods based on machine learning methods
have been developed for statistical adjustment. These machine learning methods
are designed to be black box methods with little input from the researcher.
Recent research used a data competition to evaluate various methods of
statistical adjustment and found that black box methods out performed all other
methods of statistical adjustment. Matching methods with covariate
prioritization are designed for direct input from substantive investigators in
direct contrast to black methods. In this article, we use a different research
design to compare matching with covariate prioritization to black box methods.
We use black box methods to replicate results from five studies where matching
with covariate prioritization was used to customize the statistical adjustment
in direct response to substantive expertise. We find little difference across
the methods. We conclude with advice for investigators.
| stat |
TCMI: a non-parametric mutual-dependence estimator for multivariate
continuous distributions | The identification of relevant features, i.e., the driving variables that
determine a process or the property of a system, is an essential part of the
analysis of data sets whose entries are described by a large number of
variables. The preferred measure for quantifying the relevance of nonlinear
statistical dependencies is mutual information, which requires as input
probability distributions. Probability distributions cannot be reliably sampled
and estimated from limited data, especially for real-valued data samples such
as lengths or energies. Here, we introduce total cumulative mutual information
(TCMI), a measure of the relevance of mutual dependencies based on cumulative
probability distributions. TCMI can be estimated directly from sample data and
is a non-parametric, robust and deterministic measure that facilitates
comparisons and rankings between feature sets with different cardinality. The
ranking induced by TCMI allows for feature selection, i.e., the identification
of the set of relevant features that are statistical related to the process or
the property of a system, while taking into account the number of data samples
as well as the cardinality of the feature subsets. We evaluate the performance
of our measure with simulated data, compare its performance with similar
multivariate dependence measures, and demonstrate the effectiveness of our
feature selection method on a set of standard data sets and a typical scenario
in materials science.
| stat |
A monotone data augmentation algorithm for longitudinal data analysis
via multivariate skew-t, skew-normal or t distributions | The mixed effects model for repeated measures (MMRM) has been widely used for
the analysis of longitudinal clinical data collected at a number of fixed time
points. We propose a robust extension of the MMRM for skewed and heavy-tailed
data on basis of the multivariate skew-t distribution, and it includes the
multivariate normal, t, and skew-normal distributions as special cases. An
efficient Markov chain Monte Carlo algorithm is developed using the monotone
data augmentation and parameter expansion techniques. We employ the algorithm
to perform controlled pattern imputations for sensitivity analyses of
longitudinal clinical trials with nonignorable dropouts. The proposed methods
are illustrated by real data analyses. Sample SAS programs for the analyses are
provided in the online supplementary material.
| stat |
A Robust Spectral Clustering Algorithm for Sub-Gaussian Mixture Models
with Outliers | We consider the problem of clustering datasets in the presence of arbitrary
outliers. Traditional clustering algorithms such as k-means and spectral
clustering are known to perform poorly for datasets contaminated with even a
small number of outliers. In this paper, we develop a provably robust spectral
clustering algorithm that applies a simple rounding scheme to denoise a
Gaussian kernel matrix built from the data points and uses vanilla spectral
clustering to recover the cluster labels of data points. We analyze the
performance of our algorithm under the assumption that the "good" data points
are generated from a mixture of sub-gaussians (we term these "inliers"), while
the outlier points can come from any arbitrary probability distribution. For
this general class of models, we show that the misclassification error decays
at an exponential rate in the signal-to-noise ratio, provided the number of
outliers is a small fraction of the inlier points. Surprisingly, this derived
error bound matches with the best-known bound for semidefinite programs (SDPs)
under the same setting without outliers. We conduct extensive experiments on a
variety of simulated and real-world datasets to demonstrate that our algorithm
is less sensitive to outliers compared to other state-of-the-art algorithms
proposed in the literature.
| stat |
Scalable and Efficient Comparison-based Search without Features | We consider the problem of finding a target object $t$ using pairwise
comparisons, by asking an oracle questions of the form \emph{"Which object from
the pair $(i,j)$ is more similar to $t$?"}. Objects live in a space of latent
features, from which the oracle generates noisy answers. First, we consider the
{\em non-blind} setting where these features are accessible. We propose a new
Bayesian comparison-based search algorithm with noisy answers; it has low
computational complexity yet is efficient in the number of queries. We provide
theoretical guarantees, deriving the form of the optimal query and proving
almost sure convergence to the target $t$. Second, we consider the \emph{blind}
setting, where the object features are hidden from the search algorithm. In
this setting, we combine our search method and a new distributional triplet
embedding algorithm into one scalable learning framework called
\textsc{Learn2Search}. We show that the query complexity of our approach on two
real-world datasets is on par with the non-blind setting, which is not
achievable using any of the current state-of-the-art embedding methods.
Finally, we demonstrate the efficacy of our framework by conducting an
experiment with users searching for movie actors.
| stat |
Bayesian State Estimation for Unobservable Distribution Systems via Deep
Learning | The problem of state estimation for unobservable distribution systems is
considered. A deep learning approach to Bayesian state estimation is proposed
for real-time applications. The proposed technique consists of distribution
learning of stochastic power injection, a Monte Carlo technique for the
training of a deep neural network for state estimation, and a Bayesian bad-data
detection and filtering algorithm. Structural characteristics of the deep
neural networks are investigated. Simulations illustrate the accuracy of
Bayesian state estimation for unobservable systems and demonstrate the benefit
of employing a deep neural network. Numerical results show the robustness of
Bayesian state estimation against modeling and estimation errors and the
presence of bad and missing data. Comparing with pseudo-measurement techniques,
direct Bayesian state estimation via deep learning neural network outperforms
existing benchmarks.
| stat |
Distributed Inference for Linear Support Vector Machine | The growing size of modern data brings many new challenges to existing
statistical inference methodologies and theories, and calls for the development
of distributed inferential approaches. This paper studies distributed inference
for linear support vector machine (SVM) for the binary classification task.
Despite a vast literature on SVM, much less is known about the inferential
properties of SVM, especially in a distributed setting. In this paper, we
propose a multi-round distributed linear-type (MDL) estimator for conducting
inference for linear SVM. The proposed estimator is computationally efficient.
In particular, it only requires an initial SVM estimator and then successively
refines the estimator by solving simple weighted least squares problem.
Theoretically, we establish the Bahadur representation of the estimator. Based
on the representation, the asymptotic normality is further derived, which shows
that the MDL estimator achieves the optimal statistical efficiency, i.e., the
same efficiency as the classical linear SVM applying to the entire data set in
a single machine setup. Moreover, our asymptotic result avoids the condition on
the number of machines or data batches, which is commonly assumed in
distributed estimation literature, and allows the case of diverging dimension.
We provide simulation studies to demonstrate the performance of the proposed
MDL estimator.
| stat |
Multifidelity Computer Model Emulation with High-Dimensional Output: An
Application to Storm Surge | Hurricane-driven storm surge is one of the most deadly and costly natural
disasters, making precise quantification of the surge hazard of great
importance. Inference of such systems is done through physics-based computer
models of the process. Such surge simulators can be implemented with a wide
range of fidelity levels, with computational burdens varying by several orders
of magnitude due to the nature of the system. The danger posed by surge makes
greater fidelity highly desirable, however such models and their high-volume
output tend to come at great computational cost, which can make detailed study
of coastal flood hazards prohibitive. These needs make the development of an
emulator combining high-dimensional output from multiple complex computer
models with different fidelity levels important. We propose a parallel partial
autoregressive cokriging model to predict highly-accurate storm surges in a
computationally efficient way over a large spatial domain. This emulator has
the capability of predicting storm surges as accurately as a high-fidelity
computer model given any storm characteristics and allows accurate assessment
of the hazards from storm surges over a large spatial domain.
| stat |
Large-Scale Shrinkage Estimation under Markovian Dependence | We consider the problem of simultaneous estimation of a sequence of dependent
parameters that are generated from a hidden Markov model. Based on observing a
noise contaminated vector of observations from such a sequence model, we
consider simultaneous estimation of all the parameters irrespective of their
hidden states under square error loss. We study the roles of statistical
shrinkage for improved estimation of these dependent parameters. Being
completely agnostic on the distributional properties of the unknown underlying
Hidden Markov model, we develop a novel non-parametric shrinkage algorithm. Our
proposed method elegantly combines \textit{Tweedie}-based non-parametric
shrinkage ideas with efficient estimation of the hidden states under Markovian
dependence. Based on extensive numerical experiments, we establish superior
performance our our proposed algorithm compared to non-shrinkage based
state-of-the-art parametric as well as non-parametric algorithms used in hidden
Markov models. We provide decision theoretic properties of our methodology and
exhibit its enhanced efficacy over popular shrinkage methods built under
independence. We demonstrate the application of our methodology on real-world
datasets for analyzing of temporally dependent social and economic indicators
such as search trends and unemployment rates as well as estimating spatially
dependent Copy Number Variations.
| stat |
Revising the Wilks Scoring System for pro RAW Powerlifting | Purpose: In powerlifting the total result is highly dependent on the athletes
bodyweight. Powerlifting is divided to equipped and RAW types. Pro RAW
powerlifting competitions use the Wilks scoring system to compare and rank
powerlifting results across bodyweights, to choose the winners. The Wilks
formulas are 5th order polynomials fitted to equipped powerlifting results from
the years 1987 - 1994. The Wilks formulas were meant to be updated every 2 to 5
years, but have never been revised. This study aims to update the Wilks scoring
system for use in pro RAW powerlifting.
Methods: The 10 highest RAW powerlifting totals per weight class for
bodyweights 60 - 175 kg for men, and 44 kg and up for women, were collected
from the openpowerlifting.org database. Polynomials were fitted to these data
separately for men and women, using polynomial regression. The scores for men
and women were calculated using these fits, and the results are assessed.
Results: 4th order polynomials were chosen for best describing the dependence
of powerlifting total result on bodyweight. The mens scores were normalized to
500 points, and womens to 455 points, on those fitted curves, to bring the top
scores to the same level across weight classes of each sex. These scores
appeared similar across the different weight classes for both men and women.
This even scoring across body weights is here considered the single most
important characteristic of a scoring system for pro competitions.
Key words: Wilks score, revised prediction curve, polynomial regression,
powerlifting
| stat |
A hypothesis testing framework for the ratio of means of two negative
binomial distributions: classifying the efficacy of anthelmintic treatment
against intestinal parasites | Over-dispersed count data typically pose a challenge to analysis using
standard statistical methods, particularly when evaluating the efficacy of an
intervention through the observed effect on the mean. We outline a novel
statistical method for analysing such data, along with a statistically coherent
framework within which the observed efficacy is assigned one of four easily
interpretable classifications relative to a target efficacy: "adequate",
"reduced", "borderline" or "inconclusive". We illustrate our approach by
analysing the anthelmintic efficacy of mebendazole using a dataset of egg
reduction rates relating to three intestinal parasites from a treatment arm of
a randomised controlled trial involving 91 children on Pemba Island, Tanzania.
Numerical validation of the type I error rates of the novel method indicate
that it performs as well as the best existing computationally-simple method,
but with the additional advantage of providing valid inference in the case of
an observed efficacy of 100%. The framework and statistical analysis method
presented also allow the required sample size of a prospective study to be
determined via simulation. Both the framework and method presented have high
potential utility within medical parasitology, as well as other fields where
over-dispersed count datasets are commonplace. In order to facilitate the use
of these methods within the wider medical community, user interfaces for both
study planning and analysis of existing datasets are freely provided along with
our open-source code via: http://www.fecrt.com/framework
| stat |
Circular Regression Trees and Forests with an Application to
Probabilistic Wind Direction Forecasting | While circular data occur in a wide range of scientific fields, the
methodology for distributional modeling and probabilistic forecasting of
circular response variables is rather limited. Most of the existing methods are
built on the framework of generalized linear and additive models, which are
often challenging to optimize and to interpret. Therefore, we suggest circular
regression trees and random forests as an intuitive alternative approach that
is relatively easy to fit. Building on previous ideas for trees modeling
circular means, we suggest a distributional approach for both trees and forests
yielding probabilistic forecasts based on the von Mises distribution. The
resulting tree-based models simplify the estimation process by using the
available covariates for partitioning the data into sufficiently homogeneous
subgroups so that a simple von Mises distribution without further covariates
can be fitted to the circular response in each subgroup. These circular
regression trees are straightforward to interpret, can capture nonlinear
effects and interactions, and automatically select the relevant covariates that
are associated with either location and/or scale changes in the von Mises
distribution. Combining an ensemble of circular regression trees to a circular
regression forest yields a local adaptive likelihood estimator for the von
Mises distribution that can regularize and smooth the covariate effects. The
new methods are evaluated in a case study on probabilistic wind direction
forecasting at two Austrian airports, considering other common approaches as a
benchmark.
| stat |
BeSS: An R Package for Best Subset Selection in Linear, Logistic and
CoxPH Models | We introduce a new R package, BeSS, for solving the best subset selection
problem in linear, logistic and Cox's proportional hazard (CoxPH) models. It
utilizes a highly efficient active set algorithm based on primal and dual
variables, and supports sequential and golden search strategies for best subset
selection. We provide a C++ implementation of the algorithm using Rcpp
interface. We demonstrate through numerical experiments based on enormous
simulation and real datasets that the new BeSS package has competitive
performance compared to other R packages for best subset selection purpose.
| stat |
Comment on "Statistical Modeling: The Two Cultures" by Leo Breiman | Motivated by Breiman's rousing 2001 paper on the "two cultures" in
statistics, we consider the role that different modeling approaches play in
causal inference. We discuss the relationship between model complexity and
causal (mis)interpretation, the relative merits of plug-in versus targeted
estimation, issues that arise in tuning flexible estimators of causal effects,
and some outstanding cultural divisions in causal inference.
| stat |
BART with Targeted Smoothing: An analysis of patient-specific stillbirth
risk | This article introduces BART with Targeted Smoothing, or tsBART, a new
Bayesian tree-based model for nonparametric regression. The goal of tsBART is
to introduce smoothness over a single target covariate t, while not necessarily
requiring smoothness over other covariates x. TsBART is based on the Bayesian
Additive Regression Trees (BART) model, an ensemble of regression trees. TsBART
extends BART by parameterizing each tree's terminal nodes with smooth functions
of t, rather than independent scalars. Like BART, tsBART captures complex
nonlinear relationships and interactions among the predictors. But unlike BART,
tsBART guarantees that the response surface will be smooth in the target
covariate. This improves interpretability and helps regularize the estimate.
After introducing and benchmarking the tsBART model, we apply it to our
motivating example: pregnancy outcomes data from the National Center for Health
Statistics. Our aim is to provide patient-specific estimates of stillbirth risk
across gestational age (t), based on maternal and fetal risk factors (x).
Obstetricians expect stillbirth risk to vary smoothly over gestational age, but
not necessarily over other covariates, and tsBART has been designed precisely
to reflect this structural knowledge. The results of our analysis show the
clear superiority of the tsBART model for quantifying stillbirth risk, thereby
providing patients and doctors with better information for managing the risk of
perinatal mortality. All methods described here are implemented in the R
package tsbart.
| stat |
A trial emulation approach for policy evaluations with group-level
longitudinal data | To limit the spread of the novel coronavirus, governments across the world
implemented extraordinary physical distancing policies, such as stay-at-home
orders, and numerous studies aim to estimate their effects. Many statistical
and econometric methods, such as difference-in-differences, leverage repeated
measurements and variation in timing to estimate policy effects, including in
the COVID-19 context. While these methods are less common in epidemiology,
epidemiologic researchers are well accustomed to handling similar complexities
in studies of individual-level interventions. "Target trial emulation"
emphasizes the need to carefully design a non-experimental study in terms of
inclusion and exclusion criteria, covariates, exposure definition, and outcome
measurement -- and the timing of those variables. We argue that policy
evaluations using group-level longitudinal ("panel") data need to take a
similar careful approach to study design, which we refer to as "policy trial
emulation." This is especially important when intervention timing varies across
jurisdictions; the main idea is to construct target trials separately for each
"treatment cohort" (states that implement the policy at the same time) and then
aggregate. We present a stylized analysis of the impact of state-level
stay-at-home orders on total coronavirus cases. We argue that estimates from
panel methods -- with the right data and careful modeling and diagnostics --
can help add to our understanding of many policies, though doing so is often
challenging.
| stat |
Robust Fitting for Generalized Additive Models for Location, Scale and
Shape | The validity of estimation and smoothing parameter selection for the wide
class of generalized additive models for location, scale and shape (GAMLSS)
relies on the correct specification of a likelihood function. Deviations from
such assumption are known to mislead any likelihood-based inference and can
hinder penalization schemes meant to ensure some degree of smoothness for
non-linear effects. We propose a general approach to achieve robustness in
fitting GAMLSSs by limiting the contribution of observations with low
log-likelihood values. Robust selection of the smoothing parameters can be
carried out either by minimizing information criteria that naturally arise from
the robustified likelihood or via an extended Fellner-Schall method. The latter
allows for automatic smoothing parameter selection and is particularly
advantageous in applications with multiple smoothing parameters. We also
address the challenge of tuning robust estimators for models with non-linear
effects by proposing a novel median downweighting proportion criterion. This
enables a fair comparison with existing robust estimators for the special case
of generalized additive models, where our estimator competes favorably. The
overall good performance of our proposal is illustrated by further simulations
in the GAMLSS setting and by an application to functional magnetic resonance
brain imaging using bivariate smoothing splines.
| stat |
stratamatch: Prognostic ScoreStratification using a Pilot Design | Optimal propensity score matching has emerged as one of the most ubiquitous
approaches for causal inference studies on observational data; However,
outstanding critiques of the statistical properties of propensity score
matching have cast doubt on the statistical efficiency of this technique, and
the poor scalability of optimal matching to large data sets makes this approach
inconvenient if not infeasible for sample sizes that are increasingly
commonplace in modern observational data. The stratamatch package provides
implementation support and diagnostics for `stratified matching designs,' an
approach which addresses both of these issues with optimal propensity score
matching for large-sample observational studies. First, stratifying the data
enables more computationally efficient matching of large data sets. Second,
stratamatch implements a `pilot design' approach in order to stratify by a
prognostic score, which may increase the precision of the effect estimate and
increase power in sensitivity analyses of unmeasured confounding.
| stat |
A recipe for accurate estimation of lifespan brain trajectories,
distinguishing longitudinal and cohort effects | We address the problem of estimating how different parts of the brain develop
and change throughout the lifespan, and how these trajectories are affected by
genetic and environmental factors. Estimation of these lifespan trajectories is
statistically challenging, since their shapes are typically highly nonlinear,
and although true change can only be quantified by longitudinal examinations,
as follow-up intervals in neuroimaging studies typically cover less than 10 \%
of the lifespan, use of cross-sectional information is necessary. Linear mixed
models (LMMs) and structural equation models (SEMs) commonly used in
longitudinal analysis rely on assumptions which are typically not met with
lifespan data, in particular when the data consist of observations combined
from multiple studies. While LMMs require a priori specification of a
polynomial functional form, SEMs do not easily handle data with unstructured
time intervals between measurements. Generalized additive mixed models (GAMMs)
offer an attractive alternative, and in this paper we propose various ways of
formulating GAMMs for estimation of lifespan trajectories of 12 brain regions,
using a large longitudinal dataset and realistic simulation experiments. We
show that GAMMs are able to more accurately fit lifespan trajectories,
distinguish longitudinal and cross-sectional effects, and estimate effects of
genetic and environmental exposures. Finally, we discuss and contrast questions
related to lifespan research which strictly require repeated measures data and
questions which can be answered with a single measurement per participant, and
in the latter case, which simplifying assumptions that need to be made. The
examples are accompanied with R code, providing a tutorial for researchers
interested in using GAMMs.
| stat |
Modelling rankings in R: the PlackettLuce package | This paper presents the R package PlackettLuce, which implements a
generalization of the Plackett-Luce model for rankings data. The generalization
accommodates both ties (of arbitrary order) and partial rankings (complete
rankings of subsets of items). By default, the implementation adds a set of
pseudo-comparisons with a hypothetical item, ensuring that the underlying
network of wins and losses between items is always strongly connected. In this
way, the worth of each item always has a finite maximum likelihood estimate,
with finite standard error. The use of pseudo-comparisons also has a
regularization effect, shrinking the estimated parameters towards equal item
worth. In addition to standard methods for model summary, PlackettLuce provides
a method to compute quasi standard errors for the item parameters. This
provides the basis for comparison intervals that do not change with the choice
of identifiability constraint placed on the item parameters. Finally, the
package provides a method for model-based partitioning using covariates whose
values vary between rankings, enabling the identification of subgroups of
judges or settings that have different item worths. The features of the package
are demonstrated through application to classic and novel data sets.
| stat |
FLAME: A Fast Large-scale Almost Matching Exactly Approach to Causal
Inference | A classical problem in causal inference is that of matching, where treatment
units need to be matched to control units based on covariate information. In
this work, we propose a method that computes high quality almost-exact matches
for high-dimensional categorical datasets. This method, called FLAME (Fast
Large-scale Almost Matching Exactly), learns a distance metric for matching
using a hold-out training data set. In order to perform matching efficiently
for large datasets, FLAME leverages techniques that are natural for query
processing in the area of database management, and two implementations of FLAME
are provided: the first uses SQL queries and the second uses bit-vector
techniques. The algorithm starts by constructing matches of the highest quality
(exact matches on all covariates), and successively eliminates variables in
order to match exactly on as many variables as possible, while still
maintaining interpretable high-quality matches and balance between treatment
and control groups. We leverage these high quality matches to estimate
conditional average treatment effects (CATEs). Our experiments show that FLAME
scales to huge datasets with millions of observations where existing
state-of-the-art methods fail, and that it achieves significantly better
performance than other matching methods.
| stat |
When the ends don't justify the means: Learning a treatment strategy to
prevent harmful indirect effects | There is a growing literature on finding so-called optimal treatment rules,
which are rules by which to assign treatment to individuals based on an
individual's characteristics, such that a desired outcome is maximized. A
related goal entails identifying individuals who are predicted to have a
harmful indirect effect (the effect of treatment on an outcome through
mediators) even in the presence of an overall beneficial effect of the
treatment on the outcome. In some cases, the likelihood of a harmful indirect
effect may outweigh a likely beneficial overall effect, and would be reason to
caution against treatment for indicated individuals. We build on both the
current mediation and optimal treatment rule literature to propose a method of
identifying a subgroup for which the treatment effect through the mediator is
harmful. Our approach is nonparametric, incorporates post-treatment variables
that may confound the mediator-outcome relationship, and does not make
restrictions on the distribution of baseline covariates, mediating variables
(considered jointly), or outcomes. We apply the proposed approach to identify a
subgroup of boys in the Moving to Opportunity housing voucher experiment who
are predicted to have harmful indirect effects, though the average total effect
is beneficial.
| stat |
Model Selection for Simulator-based Statistical Models: A Kernel
Approach | We propose a novel approach to model selection for simulator-based
statistical models. The proposed approach defines a mixture of candidate
models, and then iteratively updates the weight coefficients for those models
as well as the parameters in each model simultaneously; this is done by
recursively applying Bayes' rule, using the recently proposed kernel recursive
ABC algorithm. The practical advantage of the method is that it can be used
even when a modeler lacks appropriate prior knowledge about the parameters in
each model. We demonstrate the effectiveness of the proposed approach with a
number of experiments, including model selection for dynamical systems in
ecology and epidemiology.
| stat |
Survival Analysis Using a 5-Step Stratified Testing and Amalgamation
Routine in Randomized Clinical Trials | Randomized clinical trials are often designed to assess whether a test
treatment prolongs survival relative to a control treatment. Increased patient
heterogeneity, while desirable for generalizability of results, can weaken the
ability of common statistical approaches to detect treatment differences,
potentially hampering the regulatory approval of safe and efficacious
therapies. A novel solution to this problem is proposed. A list of baseline
covariates that have the potential to be prognostic for survival under either
treatment is pre-specified in the analysis plan. At the analysis stage, using
all observed survival times but blinded to patient-level treatment assignment,
'noise' covariates are removed with elastic net Cox regression. The shortened
covariate list is used by a conditional inference tree algorithm to segment the
heterogeneous trial population into subpopulations of prognostically
homogeneous patients (risk strata). After patient-level treatment unblinding, a
treatment comparison is done within each formed risk stratum and stratum-level
results are combined for overall statistical inference. The impressive
power-boosting performance of our proposed 5-step stratified testing and
amalgamation routine (5-STAR), relative to that of the logrank test and other
common approaches that do not leverage inherently structured patient
heterogeneity, is illustrated using a hypothetical and two real datasets along
with simulation results. Furthermore, the importance of reporting stratum-level
comparative treatment effects (time ratios from accelerated failure time model
fits in conjunction with model averaging and, as needed, hazard ratios from Cox
proportional hazard model fits) is highlighted as a potential enabler of
personalized medicine. A fiveSTAR R package is available at
https://github.com/rmarceauwest/fiveSTAR.
| stat |
Estimation of a function of low local dimensionality by deep neural
networks | Deep neural networks (DNNs) achieve impressive results for complicated tasks
like object detection on images and speech recognition. Motivated by this
practical success, there is now a strong interest in showing good theoretical
properties of DNNs. To describe for which tasks DNNs perform well and when they
fail, it is a key challenge to understand their performance. The aim of this
paper is to contribute to the current statistical theory of DNNs. We apply DNNs
on high dimensional data and we show that the least squares regression
estimates using DNNs are able to achieve dimensionality reduction in case that
the regression function has locally low dimensionality. Consequently, the rate
of convergence of the estimate does not depend on its input dimension $d$, but
on its local dimension $d^*$ and the DNNs are able to circumvent the curse of
dimensionality in case that $d^*$ is much smaller than $d$. In our simulation
study we provide numerical experiments to support our theoretical result and we
compare our estimate with other conventional nonparametric regression
estimates. The performance of our estimates is also validated in experiments
with real data.
| stat |
Linear pooling of sample covariance matrices | We consider covariance matrix estimation in a setting, where there are
multiple classes (populations). We propose to estimate each class covariance
matrix as a linear combination of all of the class sample covariance matrices.
This approach is shown to reduce the estimation error when the sample sizes are
limited and the true class covariance matrices share a similar structure. We
develop an effective method for estimating the minimum mean squared error
coefficients for the linear combination when the samples are drawn from
(unspecified) elliptically symmetric distributions with finite fourth-order
moments. To this end, we utilize the spatial sign covariance matrix, which we
show (under rather general conditions) to be a consistent estimator of the
trace normalized covariance matrix as both the sample size and the dimension
grow to infinity. We also show how the proposed method can be used in choosing
the regularization parameters for multiple target matrices in a single class
covariance matrix estimation problem. We assess the proposed method via
numerical simulation studies including an application in global minimum
variance portfolio optimization using real stock data, where it is shown to
outperform several state-of-the-art methods.
| stat |
An Empirical Characterization of Fair Machine Learning For Clinical Risk
Prediction | The use of machine learning to guide clinical decision making has the
potential to worsen existing health disparities. Several recent works frame the
problem as that of algorithmic fairness, a framework that has attracted
considerable attention and criticism. However, the appropriateness of this
framework is unclear due to both ethical as well as technical considerations,
the latter of which include trade-offs between measures of fairness and model
performance that are not well-understood for predictive models of clinical
outcomes. To inform the ongoing debate, we conduct an empirical study to
characterize the impact of penalizing group fairness violations on an array of
measures of model performance and group fairness. We repeat the analyses across
multiple observational healthcare databases, clinical outcomes, and sensitive
attributes. We find that procedures that penalize differences between the
distributions of predictions across groups induce nearly-universal degradation
of multiple performance metrics within groups. On examining the secondary
impact of these procedures, we observe heterogeneity of the effect of these
procedures on measures of fairness in calibration and ranking across
experimental conditions. Beyond the reported trade-offs, we emphasize that
analyses of algorithmic fairness in healthcare lack the contextual grounding
and causal awareness necessary to reason about the mechanisms that lead to
health disparities, as well as about the potential of algorithmic fairness
methods to counteract those mechanisms. In light of these limitations, we
encourage researchers building predictive models for clinical use to step
outside the algorithmic fairness frame and engage critically with the broader
sociotechnical context surrounding the use of machine learning in healthcare.
| stat |
On the Estimation of Network Complexity: Dimension of Graphons | Network complexity has been studied for over half a century and has found a
wide range of applications. Many methods have been developed to characterize
and estimate the complexity of networks. However, there has been little
research with statistical guarantees. In this paper, we develop a statistical
theory of graph complexity in a general model of random graphs, the so-called
graphon model.
Given a graphon, we endow the latent space of the nodes with the neighborhood
distance that measures the propensity of two nodes to be connected with similar
nodes. Our complexity index is then based on the covering number and the
Minkowski dimension of (a purified version of) this metric space. Although the
latent space is not identifiable, these indices turn out to be identifiable.
This notion of complexity has simple interpretations on popular examples of
random graphs: it matches the number of communities in stochastic block models;
the dimension of the Euclidean space in random geometric graphs; the regularity
of the link function in H\"older graphon models.
From a single observation of the graph, we construct an estimator of the
neighborhood-distance and show universal non-asymptotic bounds for its risk,
matching minimax lower bounds. Based on this estimated distance, we compute the
corresponding covering number and Minkowski dimension and we provide optimal
non-asymptotic error bounds for these two plug-in estimators.
| stat |
A bracketing relationship between difference-in-differences and
lagged-dependent-variable adjustment | Difference-in-differences is a widely-used evaluation strategy that draws
causal inference from observational panel data. Its causal identification
relies on the assumption of parallel trends, which is scale dependent and may
be questionable in some applications. A common alternative is a regression
model that adjusts for the lagged dependent variable, which rests on the
assumption of ignorability conditional on past outcomes. In the context of
linear models, \citet{APbook} show that the difference-in-differences and
lagged-dependent-variable regression estimates have a bracketing relationship.
Namely, for a true positive effect, if ignorability is correct, then mistakenly
assuming parallel trends will overestimate the effect; in contrast, if the
parallel trends assumption is correct, then mistakenly assuming ignorability
will underestimate the effect. We show that the same bracketing relationship
holds in general nonparametric (model-free) settings. We also extend the result
to semiparametric estimation based on inverse probability weighting. We provide
three examples to illustrate the theoretical results with replication files in
\citet{ding2019bracketingData}.
| stat |
Neural Clustering Processes | Probabilistic clustering models (or equivalently, mixture models) are basic
building blocks in countless statistical models and involve latent random
variables over discrete spaces. For these models, posterior inference methods
can be inaccurate and/or very slow. In this work we introduce deep network
architectures trained with labeled samples from any generative model of
clustered datasets. At test time, the networks generate approximate posterior
samples of cluster labels for any new dataset of arbitrary size. We develop two
complementary approaches to this task, requiring either O(N) or O(K) network
forward passes per dataset, where N is the dataset size and K the number of
clusters. Unlike previous approaches, our methods sample the labels of all the
data points from a well-defined posterior, and can learn nonparametric Bayesian
posteriors since they do not limit the number of mixture components. As a
scientific application, we present a novel approach to neural spike sorting for
high-density multielectrode arrays.
| stat |
Forecasting Corn Yield with Machine Learning Ensembles | The emerge of new technologies to synthesize and analyze big data with
high-performance computing, has increased our capacity to more accurately
predict crop yields. Recent research has shown that Machine learning (ML) can
provide reasonable predictions, faster, and with higher flexibility compared to
simulation crop modeling. The earlier the prediction during the growing season
the better, but this has not been thoroughly investigated as previous studies
considered all data available to predict yields. This paper provides a machine
learning based framework to forecast corn yields in three US Corn Belt states
(Illinois, Indiana, and Iowa) considering complete and partial in-season
weather knowledge. Several ensemble models are designed using blocked
sequential procedure to generate out-of-bag predictions. The forecasts are made
in county-level scale and aggregated for agricultural district, and state level
scales. Results show that ensemble models based on weighted average of the base
learners outperform individual models. Specifically, the proposed ensemble
model could achieve best prediction accuracy (RRMSE of 7.8%) and least mean
bias error (-6.06 bu/acre) compared to other developed models. Comparing our
proposed model forecasts with the literature demonstrates the superiority of
forecasts made by our proposed ensemble model. Results from the scenario of
having partial in-season weather knowledge reveal that decent yield forecasts
can be made as early as June 1st. To find the marginal effect of each input
feature on the forecasts made by the proposed ensemble model, a methodology is
suggested that is the basis for finding feature importance for the ensemble
model. The findings suggest that weather features corresponding to weather in
weeks 18-24 (May 1st to June 1st) are the most important input features.
| stat |
Transporting Experimental Results with Entropy Balancing | We show how entropy balancing can be used for transporting experimental
treatment effects from a trial population onto a target population. This method
is doubly-robust in the sense that if either the outcome model or the
probability of trial participation is correctly specified, then the estimate of
the target population average treatment effect is consistent. Furthermore, we
only require the sample moments of the effect modifiers drawn from the target
population to consistently estimate the target population average treatment
effect. We compared the finite-sample performance of entropy balancing with
several alternative methods for transporting treatment effects between
populations. Entropy balancing techniques are efficient and robust to
violations of model misspecification. We also examine the results of our
proposed method in an applied analysis of the Action to Control Cardiovascular
Risk in Diabetes Blood Pressure (ACCORD-BP) trial transported to a sample of US
adults with diabetes taken from the National Health and Nutrition Examination
Survey (NHANES) cohort.
| stat |
Semiparametric Tensor Factor Analysis by Iteratively Projected SVD | This paper introduces a general framework of Semiparametric TEnsor FActor
analysis (STEFA) that focuses on the methodology and theory of low-rank tensor
decomposition with auxiliary covariates. STEFA models extend tensor factor
models by incorporating instrumental covariates in the loading matrices. We
propose an algorithm of Iteratively Projected SVD (IP-SVD) for the
semiparametric estimations. It iteratively projects tensor data onto the linear
space spanned by covariates and applies SVD on matricized tensors over each
mode. We establish the convergence rates of the loading matrices and the core
tensor factor. Compared with the Tucker decomposition, IP-SVD yields more
accurate estimates with a faster convergence rate. Besides estimation, we show
several prediction methods with newly observed covariates based on the STEFA
model. On both real and synthetic tensor data, we demonstrate the efficacy of
the STEFA model and the IP-SVD algorithm on both the estimation and prediction
tasks.
| stat |
A Conditional Empirical Likelihood Based Method for Model Parameter
Estimation from Complex survey Datasets | We consider an empirical likelihood framework for inference for a statistical
model based on an informative sampling design. Covariate information is
incorporated both through the weights and the estimating equations. The
estimator is based on conditional weights. We show that under usual conditions,
with population size increasing unbounded, the estimates are strongly
consistent, asymptotically unbiased and normally distributed. Our framework
provides additional justification for inverse probability weighted score
estimators in terms of conditional empirical likelihood. In doing so, it
bridges the gap between design-based and model-based modes of inference in
survey sampling settings. We illustrate these ideas with an application to an
electoral survey.
| stat |
Does Terrorism Trigger Online Hate Speech? On the Association of Events
and Time Series | Hate speech is ubiquitous on the Web. Recently, the offline causes that
contribute to online hate speech have received increasing attention. A
recurring question is whether the occurrence of extreme events offline
systematically triggers bursts of hate speech online, indicated by peaks in the
volume of hateful social media posts. Formally, this question translates into
measuring the association between a sparse event series and a time series. We
propose a novel statistical methodology to measure, test and visualize the
systematic association between rare events and peaks in a time series. In
contrast to previous methods for causal inference or independence tests on time
series, our approach focuses only on the timing of events and peaks, and no
other distributional characteristics. We follow the framework of event
coincidence analysis (ECA) that was originally developed to correlate point
processes. We formulate a discrete-time variant of ECA and derive all required
distributions to enable analyses of peaks in time series, with a special focus
on serial dependencies and peaks over multiple thresholds. The analysis gives
rise to a novel visualization of the association via quantile-trigger rate
plots. We demonstrate the utility of our approach by analyzing whether Islamist
terrorist attacks in Western Europe and North America systematically trigger
bursts of hate speech and counter-hate speech on Twitter.
| stat |
Closed test procedures for the comparison of dose groups against a
negative control group or placebo | Dose groups are compared with a control assuming an order restriction usually
by the Williams trend test. Here, as an alternative, two variants of the closed
testing procedure are considered, one where global Williams tests are used in
the partition hypotheses, and another where pairwise contrast tests are used
for this purpose. Related R software is provided.
| stat |
Clustering of longitudinal curves via a penalized method and EM
algorithm | A new algorithm is proposed for clustering longitudinal curves. The mean
curves and the principal component functions are modeled using B-spline. The
clusters of the mean curves are identified through a concave pairwise fusion
method. The EM algorithm and the alternating direction method of multiplier
algorithm are combined to estimate the group structure, mean functions and the
principal components simultaneously. The proposed method also allows to
incorporate the prior neighborhood information to have more meaningful groups
by adding pairwise weights in the pairwise penalties. In the simulation study,
the performance of the proposed method is compared to two existing clustering
methods in terms of the accuracy for estimating the number of subgroups and
mean functions. The results suggest that ignoring covariance structure will
have a great effect on the performance of estimating the number of groups and
estimating accuracy. The effect of including pairwise weights is also explored
in a spatial lattice setting to take consideration of the spatial information.
The results show that incorporating spatial weights will improve the
performance. An example is used to illustrate the algorithm.
| stat |
Large-P Variable Selection in Two-Stage Models | Model selection in the large-P small-N scenario is discussed in the framework
of two-stage models. Two specific models are considered, namely, two-stage
least squares (TSLS) involving instrumental variables (IVs), and mediation
models. In both cases, the number of putative variables (e.g. instruments or
mediators) is large, but only a small subset should be included in the
two-stage model. We use two variable selection methods which are designed for
high-dimensional settings, and compare their performance in terms of their
ability to find the true IVs or mediators. Our approach is demonstrated via
simulations and case studies.
| stat |
Selected Methods for non-Gaussian Data Analysis | The basic goal of computer engineering is the analysis of data. Such data are
often large data sets distributed according to various distribution models. In
this manuscript we focus on the analysis of non-Gaussian distributed data. In
the case of univariate data analysis we discuss stochastic processes with
auto-correlated increments and univariate distributions derived from specific
stochastic processes, i.e. Levy and Tsallis distributions. Deep investigation
of multivariate non-Gaussian distributions requires the copula approach. A
copula is an component of multivariate distribution that models the mutual
interdependence between marginals. There are many copula families characterised
by various measures of the dependence between marginals. Importantly, one of
those are `tail' dependencies that model the simultaneous appearance of extreme
values in many marginals. Those extreme events may reflect a crisis given
financial data, outliers in machine learning, or a traffic congestion. Next we
discuss higher order multivariate cumulants that are non-zero if multivariate
distribution is non-Gaussian. However, the relation between cumulants and
copulas is not straight forward and rather complicated. We discuss the
application of those cumulants to extract information about non-Gaussian
multivariate distributions, such that information about non-Gaussian copulas.
The use of higher order multivariate cumulants in computer science is inspired
by financial data analysis, especially by the safe investment portfolio
evaluation. There are many other applications of higher order multivariate
cumulants in data engineering, especially in: signal processing, non-linear
system identification, blind sources separation, and direction finding
algorithms of multi-source signals.
| stat |
Nothing to see here? Non-inferiority approaches to parallel trends and
other model assumptions | Many causal models make assumptions of "no difference" or "no effect." For
example, difference-in-differences (DID) assumes that there is no trend
difference between treatment and comparison groups' untreated potential
outcomes ("parallel trends"). Tests of these assumptions typically assume a
null hypothesis that there is no violation. When researchers fail to reject the
null, they consider the assumption to hold. We argue this approach is incorrect
and frequently misleading. These tests reverse the roles of Type I and Type II
error and have a high probability of missing assumption violations. Even when
power is high, they may detect statistically significant violations too small
to be of practical importance. We present test reformulations in a
non-inferiority framework that rule out violations of model assumptions that
exceed some threshold. We then focus on the parallel trends assumption, for
which we propose a "one step up" method: 1) reporting treatment effect
estimates from a model with a more complex trend difference than is believed to
be the case and 2) testing that that the estimated treatment effect falls
within a specified distance of the treatment effect from the simpler model. We
show that this reduces bias while also considering power, controlling
mean-squared error. Our base model also aligns power to detect a treatment
effect with power to rule out meaningful violations of parallel trends. We
apply our approach to 4 data sets used to analyze the Affordable Care Act's
dependent coverage mandate and demonstrate that coverage gains may have been
smaller than previously estimated.
| stat |
Generalized Kernel Two-Sample Tests | Kernel two-sample tests have been widely used for multivariate data in
testing equal distribution. However, existing tests based on mapping
distributions into a reproducing kernel Hilbert space do not work well for some
common alternatives when the dimension of the data is moderate to high due to
the curse of dimensionality. We propose a new test statistic that makes use of
an informative pattern under moderate and high dimensions and achieves
substantial power improvements over existing kernel two-sample tests for a wide
range of alternatives. We also propose alternative testing procedures that
maintain high power with low computational cost, offering easy off-the-shelf
tools for large datasets. We illustrate these new approaches through an
analysis of the New York City taxi data.
| stat |
A proposal for informative default priors scaled by the standard error
of estimates | If we have an unbiased estimate of some parameter of interest, then its
absolute value is positively biased for the absolute value of the parameter.
This bias is large when the signal-to-noise ratio (SNR) is small, and it
becomes even larger when we condition on statistical significance; the winner's
curse. This is a frequentist motivation for regularization. To determine a
suitable amount of shrinkage, we propose to estimate the distribution of the
SNR from a large collection or corpus of similar studies and use this as a
prior distribution. The wider the scope of the corpus, the less informative the
prior, but a wider scope does not necessarily result in a more diffuse prior.
We show that the estimation of the prior simplifies if we require that
posterior inference is equivariant under linear transformations of the data. We
demonstrate our approach with corpora of 86 replication studies from psychology
and 178 phase 3 clinical trials. Our suggestion is not intended to be a
replacement for a prior based on full information about a particular problem;
rather, it represents a familywise choice that should yield better long-term
properties than the current default uniform prior, which has led to systematic
overestimates of effect sizes and a replication crisis when these inflated
estimates have not shown up in later studies.
| stat |
Robust Factor Analysis Parameter Estimation | This paper considers the problem of robustly estimating the parameters of a
heavy-tailed multivariate distribution when the covariance matrix is known to
have the structure of a low-rank matrix plus a diagonal matrix as considered in
factor analysis (FA). By assuming the observed data to follow the multivariate
Student's t distribution, we can robustly estimate the parameters via maximum
likelihood estimation (MLE). However, the MLE of parameters becomes an
intractable problem when the multivariate Student's t distribution and the FA
structure are both introduced. In this paper, we propose an algorithm based on
the generalized expectation maximization (GEM) method to obtain estimators. The
robustness of our proposed method is further enhanced to cope with missing
values. Finally, we show the performance of our proposed algorithm using both
synthetic data and real financial data.
| stat |
Cross Validation for Penalized Quantile Regression with a Case-Weight
Adjusted Solution Path | Cross validation is widely used for selecting tuning parameters in
regularization methods, but it is computationally intensive in general. To
lessen its computational burden, approximation schemes such as generalized
approximate cross validation (GACV) are often employed. However, such
approximations may not work well when non-smooth loss functions are involved.
As a case in point, approximate cross validation schemes for penalized quantile
regression do not work well for extreme quantiles. In this paper, we propose a
new algorithm to compute the leave-one-out cross validation scores exactly for
quantile regression with ridge penalty through a case-weight adjusted solution
path. Resorting to the homotopy technique in optimization, we introduce a case
weight for each individual data point as a continuous embedding parameter and
decrease the weight gradually from one to zero to link the estimators based on
the full data and those with a case deleted. This allows us to design a
solution path algorithm to compute all leave-one-out estimators very
efficiently from the full-data solution. We show that the case-weight adjusted
solution path is piecewise linear in the weight parameter, and using the
solution path, we examine case influences comprehensively and observe that
different modes of case influences emerge, depending on the specified
quantiles, data dimensions and penalty parameter.
| stat |
Lessons Learned from the Bayesian Design and Analysis for the BNT162b2
COVID-19 Vaccine Phase 3 Trial | The phase III BNT162b2 mRNA COVID-19 vaccine trial is based on a Bayesian
design and analysis, and the main evidence of vaccine efficacy is presented in
Bayesian statistics. Confusion and mistakes are produced in the presentation of
the Bayesian results. Some key statistics, such as Bayesian credible intervals,
are mislabeled and stated as confidence intervals. Posterior probabilities of
the vaccine efficacy are not reported as the main results. We illustrate the
main differences in the reporting of Bayesian analysis results for a clinical
trial and provide four recommendations. We argue that statistical evidence from
a Bayesian trial, when presented properly, is easier to interpret and directly
addresses the main clinical questions, thereby better supporting regulatory
decision making. We also recommend using abbreviation "BI" to represent
Bayesian credible intervals as a differentiation to "CI" which stands for
confidence interval.
| stat |
NesPrInDT: Nested undersampling in PrInDT | In this paper, we extend our PrInDT method (Weihs, Buschfeld 2021) towards
additional undersampling of one of the predictors. This helps us to handle
multiple unbalanced data sets, i.e. data sets that are not only unbalanced with
respect to the class variable but also in one of the predictor variables.
Beyond the advantages of such an approach, our study reveals that the balanced
accuracy in the full data set can be much lower than in the predictor
undersamples. We discuss potential reasons for this problem and draw
methodological conclusions for linguistic studies.
| stat |
Invariance and identifiability issues for word embeddings | Word embeddings are commonly obtained as optimizers of a criterion function
$f$ of a text corpus, but assessed on word-task performance using a different
evaluation function $g$ of the test data. We contend that a possible source of
disparity in performance on tasks is the incompatibility between classes of
transformations that leave $f$ and $g$ invariant. In particular, word
embeddings defined by $f$ are not unique; they are defined only up to a class
of transformations to which $f$ is invariant, and this class is larger than the
class to which $g$ is invariant. One implication of this is that the apparent
superiority of one word embedding over another, as measured by word task
performance, may largely be a consequence of the arbitrary elements selected
from the respective solution sets. We provide a formal treatment of the above
identifiability issue, present some numerical examples, and discuss possible
resolutions.
| stat |
Data Mining in Large Frequency Tables With Ontology, with an Application
to the Vaccine Adverse Event Reporting System | Vaccine safety is a concerning problem of the public, and many signal
detecting methods have been developed to identify relative risks between
vaccines and adverse events (AEs). Those methods usually focus on individual
AEs, where the randomness of data is high. The results often turn out to be
inaccurate and lack of clinical meaning. The AE ontology contains information
about biological similarity of AEs. Based on this, we extend the concept of
relative risks (RRs) to AE group level, which allows the possibility of more
accurate and meaningful estimation by utilizing data from the whole group. In
this paper, we propose the method zGPS.AO (Zero Inflated Gamma Poisson Shrinker
with AE ontology) based on the zero inflated negative binomial distribution.
This model has two purples: a regression model estimating group level RRs, and
a empirical bayes framework to evaluate AE level RRs. The regression part can
handle both excess zeros and over dispersion in the data, and the empirical
method borrows information from both group level and AE level to reduce data
noise and stabilize the AE level result. We have demonstrate the unbiaseness
and low variance features of our model with simulated data, and obtained
meaningful results coherent with previous studies on the VAERS (Vaccine Adverse
Event Reporting System) database. The proposed methods are implemented in the R
package zGPS.AO, which can be installed from the Comprehensive R Archive
Network, CRAN. The results on VAERS data are visualized using the interactive
web app Rshiny.
| stat |
Distillation $\approx$ Early Stopping? Harvesting Dark Knowledge
Utilizing Anisotropic Information Retrieval For Overparameterized Neural
Network | Distillation is a method to transfer knowledge from one model to another and
often achieves higher accuracy with the same capacity. In this paper, we aim to
provide a theoretical understanding on what mainly helps with the distillation.
Our answer is "early stopping". Assuming that the teacher network is
overparameterized, we argue that the teacher network is essentially harvesting
dark knowledge from the data via early stopping. This can be justified by a new
concept, {Anisotropic Information Retrieval (AIR)}, which means that the neural
network tends to fit the informative information first and the non-informative
information (including noise) later. Motivated by the recent development on
theoretically analyzing overparameterized neural networks, we can characterize
AIR by the eigenspace of the Neural Tangent Kernel(NTK). AIR facilities a new
understanding of distillation. With that, we further utilize distillation to
refine noisy labels. We propose a self-distillation algorithm to sequentially
distill knowledge from the network in the previous training epoch to avoid
memorizing the wrong labels. We also demonstrate, both theoretically and
empirically, that self-distillation can benefit from more than just early
stopping. Theoretically, we prove convergence of the proposed algorithm to the
ground truth labels for randomly initialized overparameterized neural networks
in terms of $\ell_2$ distance, while the previous result was on convergence in
$0$-$1$ loss. The theoretical result ensures the learned neural network enjoy a
margin on the training data which leads to better generalization. Empirically,
we achieve better testing accuracy and entirely avoid early stopping which
makes the algorithm more user-friendly.
| stat |
Testing Conditional Independence on Discrete Data using Stochastic
Complexity | Testing for conditional independence is a core aspect of constraint-based
causal discovery. Although commonly used tests are perfect in theory, they
often fail to reject independence in practice, especially when conditioning on
multiple variables.
We focus on discrete data and propose a new test based on the notion of
algorithmic independence that we instantiate using stochastic complexity.
Amongst others, we show that our proposed test, SCI, is an asymptotically
unbiased as well as $L_2$ consistent estimator for conditional mutual
information (CMI). Further, we show that SCI can be reformulated to find a
sensible threshold for CMI that works well on limited samples. Empirical
evaluation shows that SCI has a lower type II error than commonly used tests.
As a result, we obtain a higher recall when we use SCI in causal discovery
algorithms, without compromising the precision.
| stat |
Spatial regression modeling using the spmoran package: Boston housing
price data examples | This study demonstrates how to use "spmoran", an R package estimating spatial
additive mixed models and other spatial regression models for Gaussian and
non-Gaussian data. Moran eigenvectors are used to an approximate Gaussian
process modeling which is interpretable in terms of the Moran coefficient. The
GP is used for modeling the spatial processes in residuals and regression
coefficients. All these models are estimated computationally efficiently. For
the sample code used in this paper, see https://github.com/dmuraka/spmoran.
| stat |
Automatic Registration and Clustering of Time Series | Clustering of time series data exhibits a number of challenges not present in
other settings, notably the problem of registration (alignment) of observed
signals. Typical approaches include pre-registration to a user-specified
template or time warping approaches which attempt to optimally align series
with a minimum of distortion. For many signals obtained from recording or
sensing devices, these methods may be unsuitable as a template signal is not
available for pre-registration, while the distortion of warping approaches may
obscure meaningful temporal information. We propose a new method for automatic
time series alignment within a clustering problem. Our approach, Temporal
Registration using Optimal Unitary Transformations (TROUT), is based on a novel
dissimilarity measure between time series that is easy to compute and
automatically identifies optimal alignment between pairs of time series. By
embedding our new measure in a optimization formulation, we retain well-known
advantages of computational and statistical performance. We provide an
efficient algorithm for TROUT-based clustering and demonstrate its superior
performance over a range of competitors.
| stat |
Connecting Bayes factor and the Region of Practical Equivalence (ROPE)
Procedure for testing interval null hypothesis | There has been strong recent interest in testing interval null hypothesis for
improved scientific inference. For example, Lakens et al (2018) and Lakens and
Harms (2017) use this approach to study if there is a pre-specified meaningful
treatment effect in gerontology and clinical trials, which is different from
the more traditional point null hypothesis that tests for any treatment effect.
Two popular Bayesian approaches are available for interval null hypothesis
testing. One is the standard Bayes factor and the other is the Region of
Practical Equivalence (ROPE) procedure championed by Kruschke and others over
many years. This paper establishes a formal connection between these two
approaches with two benefits. First, it helps to better understand and improve
the ROPE procedure. Second, it leads to a simple and effective algorithm for
computing Bayes factor in a wide range of problems using draws from posterior
distributions generated by standard Bayesian programs such as BUGS, JAGS and
Stan. The tedious and error-prone task of coding custom-made software specific
for Bayes factor is then avoided.
| stat |
Fixed-Effect Regressions on Network Data | This paper considers inference on fixed effects in a linear regression model
estimated from network data. An important special case of our setup is the
two-way regression model. This is a workhorse technique in the analysis of
matched data sets, such as employer-employee or student-teacher panel data. We
formalize how the structure of the network affects the accuracy with which the
fixed effects can be estimated. This allows us to derive sufficient conditions
on the network for consistent estimation and asymptotically-valid inference to
be possible. Estimation of moments is also considered. We allow for general
networks and our setup covers both the dense and sparse case. We provide
numerical results for the estimation of teacher value-added models and
regressions with occupational dummies.
| stat |
Selective Inference for Hierarchical Clustering | Testing for a difference in means between two groups is fundamental to
answering research questions across virtually every scientific area. Classical
tests control the Type I error rate when the groups are defined a priori.
However, when the groups are instead defined via a clustering algorithm, then
applying a classical test for a difference in means between the groups yields
an extremely inflated Type I error rate. Notably, this problem persists even if
two separate and independent data sets are used to define the groups and to
test for a difference in their means. To address this problem, in this paper,
we propose a selective inference approach to test for a difference in means
between two clusters obtained from any clustering method. Our procedure
controls the selective Type I error rate by accounting for the fact that the
null hypothesis was generated from the data. We describe how to efficiently
compute exact p-values for clusters obtained using agglomerative hierarchical
clustering with many commonly used linkages. We apply our method to simulated
data and to single-cell RNA-seq data.
| stat |
On unified framework for continuous-time grey models: an integral
matching perspective | Since most of the research about grey forecasting models is focused on
developing novel models and improving accuracy, relatively limited attention
has been paid to the modelling mechanism and relationships among diverse kinds
of models. This paper aims to unify and reconstruct continuous-time grey
models, highlighting the differences and similarities among different models.
First, the unified form of grey models is proposed and simplified into a
reduced-order ordinary differential equation. Then, the integral matching that
consists of integral transformation and least squares, is proposed to estimate
the structural parameter and initial value simultaneously. The cumulative sum
operator, an essential element in grey modelling, proves to be the discrete
approximation of the integral transformation formula. Next, grey models are
reconstructed by the integral matching-based ordinary differential equations.
Finally, the existing grey models are compared with the reconstructed models
through extensive simulation studies, and a real-world example shows how to
apply and further verify the reconstructed model.
| stat |
On overfitting and asymptotic bias in batch reinforcement learning with
partial observability | This paper provides an analysis of the tradeoff between asymptotic bias
(suboptimality with unlimited data) and overfitting (additional suboptimality
due to limited data) in the context of reinforcement learning with partial
observability. Our theoretical analysis formally characterizes that while
potentially increasing the asymptotic bias, a smaller state representation
decreases the risk of overfitting. This analysis relies on expressing the
quality of a state representation by bounding L1 error terms of the associated
belief states. Theoretical results are empirically illustrated when the state
representation is a truncated history of observations, both on synthetic POMDPs
and on a large-scale POMDP in the context of smartgrids, with real-world data.
Finally, similarly to known results in the fully observable setting, we also
briefly discuss and empirically illustrate how using function approximators and
adapting the discount factor may enhance the tradeoff between asymptotic bias
and overfitting in the partially observable context.
| stat |
Properties of Laplacian Pyramids for Extension and Denoising | We analyze the Laplacian pyramids algorithm of Rabin and Coifman for
extending and denoising a function sampled on a discrete set of points. We
provide mild conditions under which the algorithm converges, and prove
stability bounds on the extended function. We also consider the iterative
application of truncated Laplacian pyramids kernels for denoising signals by
non-local means.
| stat |
Referenced Thermodynamic Integration for Bayesian Model Selection:
Application to COVID-19 Model Selection | Model selection is a fundamental part of the applied Bayesian statistical
methodology. Metrics such as the Akaike Information Criterion are commonly used
in practice to select models but do not incorporate the uncertainty of the
models' parameters and can give misleading choices. One approach that uses the
full posterior distribution is to compute the ratio of two models' normalising
constants, known as the Bayes factor. Often in realistic problems, this
involves the integration of analytically intractable, high-dimensional
distributions, and therefore requires the use of stochastic methods such as
thermodynamic integration (TI). In this paper we apply a variation of the TI
method, referred to as referenced TI, which computes a single model's
normalising constant in an efficient way by using a judiciously chosen
reference density. The advantages of the approach and theoretical
considerations are set out, along with explicit pedagogical 1 and 2D examples.
Benchmarking is presented with comparable methods and we find favourable
convergence performance. The approach is shown to be useful in practice when
applied to a real problem - to perform model selection for a semi-mechanistic
hierarchical Bayesian model of COVID-19 transmission in South Korea involving
the integration of a 200D density.
| stat |
Rank Dynamics for Functional Data | The study of the dynamic behavior of cross-sectional ranks over time for
functional data and the ranks of the observed curves at each time point and
their temporal evolution can yield valuable insights into the time dynamics of
functional data. This approach is of interest in various application areas. For
the analysis of the dynamics of ranks, estimation of the cross-sectional ranks
of functional data is a first step. Several statistics of interest for ranked
functional data are proposed. To quantify the evolution of ranks over time, a
model for rank derivatives is introduced, where rank dynamics are decomposed
into two components. One component corresponds to population changes and the
other to individual changes that both affect the rank trajectories of
individuals. The joint asymptotic normality for suitable estimates of these two
components is established. The proposed approaches are illustrated with
simulations and three longitudinal data sets: Growth curves obtained from the
Z\"urich Longitudinal Growth Study, monthly house price data in the US from
1996 to 2015 and Major League Baseball offensive data for the 2017 season.
| stat |
Enhanced Cube Implementation For Highly Stratified Population | A balanced sampling design should always be the adopted strategies if
auxiliary information is available. Besides, integrating a stratified structure
of the population in the sampling process can considerably reduce the variance
of the estimators. We propose here a new method to handle the selection of a
balanced sample in a highly stratified population. The method improves
substantially the commonly used sampling design and reduces the time-consuming
problem that could arise if inclusion probabilities within strata do not sum to
an integer.
| stat |
Adversarial Monte Carlo Meta-Learning of Optimal Prediction Procedures | We frame the meta-learning of prediction procedures as a search for an
optimal strategy in a two-player game. In this game, Nature selects a prior
over distributions that generate labeled data consisting of features and an
associated outcome, and the Predictor observes data sampled from a distribution
drawn from this prior. The Predictor's objective is to learn a function that
maps from a new feature to an estimate of the associated outcome. We establish
that, under reasonable conditions, the Predictor has an optimal strategy that
is equivariant to shifts and rescalings of the outcome and is invariant to
permutations of the observations and to shifts, rescalings, and permutations of
the features. We introduce a neural network architecture that satisfies these
properties. The proposed strategy performs favorably compared to standard
practice in both parametric and nonparametric experiments.
| stat |
Deterministic tensor completion with hypergraph expanders | We provide a novel analysis of low-rank tensor completion based on hypergraph
expanders. As a proxy for rank, we minimize the max-quasinorm of the tensor,
which generalizes the max-norm for matrices. Our analysis is deterministic and
shows that the number of samples required to approximately recover an order-$t$
tensor with at most $n$ entries per dimension is linear in $n$, under the
assumption that the rank and order of the tensor are $O(1)$. As steps in our
proof, we find an improved expander mixing lemma for a $t$-partite, $t$-uniform
regular hypergraph model, and prove several new properties about tensor
max-quasinorm. To the best of our knowledge, this is the first deterministic
analysis of tensor completion. We develop a practical algorithm that solves a
relaxed version of the max-quasinorm minimization problem, and we demonstrate
its efficacy with numerical experiments.
| stat |
Functional Outlier Detection and Taxonomy by Sequential Transformations | Functional data analysis can be seriously impaired by abnormal observations,
which can be classified as either magnitude or shape outliers based on their
way of deviating from the bulk of data. Identifying magnitude outliers is
relatively easy, while detecting shape outliers is much more challenging. We
propose turning the shape outliers into magnitude outliers through data
transformation and detecting them using the functional boxplot. Besides easing
the detection procedure, applying several transformations sequentially provides
a reasonable taxonomy for the flagged outliers. A joint functional ranking,
which consists of several transformations, is also defined here. Simulation
studies are carried out to evaluate the performance of the proposed method
using different functional depth notions. Interesting results are obtained in
several practical applications.
| stat |
Bayesian Optimization with Binary Auxiliary Information | This paper presents novel mixed-type Bayesian optimization (BO) algorithms to
accelerate the optimization of a target objective function by exploiting
correlated auxiliary information of binary type that can be more cheaply
obtained, such as in policy search for reinforcement learning and
hyperparameter tuning of machine learning models with early stopping. To
achieve this, we first propose a mixed-type multi-output Gaussian process
(MOGP) to jointly model the continuous target function and binary auxiliary
functions. Then, we propose information-based acquisition functions such as
mixed-type entropy search (MT-ES) and mixed-type predictive ES (MT-PES) for
mixed-type BO based on the MOGP predictive belief of the target and auxiliary
functions. The exact acquisition functions of MT-ES and MT-PES cannot be
computed in closed form and need to be approximated. We derive an efficient
approximation of MT-PES via a novel mixed-type random features approximation of
the MOGP model whose cross-correlation structure between the target and
auxiliary functions can be exploited for improving the belief of the global
target maximizer using observations from evaluating these functions. We propose
new practical constraints to relate the global target maximizer to the binary
auxiliary functions. We empirically evaluate the performance of MT-ES and
MT-PES with synthetic and real-world experiments.
| stat |
Unsupervised Machine Learning for the Discovery of Latent Disease
Clusters and Patient Subgroups Using Electronic Health Records | Machine learning has become ubiquitous and a key technology on mining
electronic health records (EHRs) for facilitating clinical research and
practice. Unsupervised machine learning, as opposed to supervised learning, has
shown promise in identifying novel patterns and relations from EHRs without
using human created labels. In this paper, we investigate the application of
unsupervised machine learning models in discovering latent disease clusters and
patient subgroups based on EHRs. We utilized Latent Dirichlet Allocation (LDA),
a generative probabilistic model, and proposed a novel model named Poisson
Dirichlet Model (PDM), which extends the LDA approach using a Poisson
distribution to model patients' disease diagnoses and to alleviate age and sex
factors by considering both observed and expected observations. In the
empirical experiments, we evaluated LDA and PDM on three patient cohorts with
EHR data retrieved from the Rochester Epidemiology Project (REP), for the
discovery of latent disease clusters and patient subgroups. We compared the
effectiveness of LDA and PDM in identifying latent disease clusters through the
visualization of disease representations learned by two approaches. We also
tested the performance of LDA and PDM in differentiating patient subgroups
through survival analysis, as well as statistical analysis. The experimental
results show that the proposed PDM could effectively identify distinguished
disease clusters by alleviating the impact of age and sex, and that LDA could
stratify patients into more differentiable subgroups than PDM in terms of
p-values. However, the subgroups discovered by PDM might imply the underlying
patterns of diseases of greater interest in epidemiology research due to the
alleviation of age and sex. Both unsupervised machine learning approaches could
be leveraged to discover patient subgroups using EHRs but with different foci.
| stat |
Learning Fair and Transferable Representations | Developing learning methods which do not discriminate subgroups in the
population is a central goal of algorithmic fairness. One way to reach this
goal is by modifying the data representation in order to meet certain fairness
constraints. In this work we measure fairness according to demographic parity.
This requires the probability of the possible model decisions to be independent
of the sensitive information. We argue that the goal of imposing demographic
parity can be substantially facilitated within a multitask learning setting. We
leverage task similarities by encouraging a shared fair representation across
the tasks via low rank matrix factorization. We derive learning bounds
establishing that the learned representation transfers well to novel tasks both
in terms of prediction performance and fairness metrics. We present experiments
on three real world datasets, showing that the proposed method outperforms
state-of-the-art approaches by a significant margin.
| stat |
CD-split and HPD-split: efficient conformal regions in high dimensions | Conformal methods create prediction bands that control average coverage
assuming solely i.i.d. data. Although the literature has mostly focused on
prediction intervals, more general regions can often better represent
uncertainty. For instance, a bimodal target is better represented by the union
of two intervals. Such prediction regions are obtained by CD-split , which
combines the split method and a data-driven partition of the feature space
which scales to high dimensions. CD-split however contains many tuning
parameters, and their role is not clear. In this paper, we provide new insights
on CD-split by exploring its theoretical properties. In particular, we show
that CD-split converges asymptotically to the oracle highest predictive density
set and satisfies local and asymptotic conditional validity. We also present
simulations which show how to tune CD-split. Finally, we introduce HPD-split, a
variation of CD-split that requires less tuning, and show that it shares the
same theoretical guarantees as CD-split. In a wide variety of our simulations,
CD-split and HPD-split have a better conditional coverage and yield smaller
prediction regions than other methods.
| stat |
Meta-Analysis of Generalized Additive Models in Neuroimaging Studies | Analyzing data from multiple neuroimaging studies has great potential in
terms of increasing statistical power, enabling detection of effects of smaller
magnitude than would be possible when analyzing each study separately and also
allowing to systematically investigate between-study differences. Restrictions
due to privacy or proprietary data as well as more practical concerns can make
it hard to share neuroimaging datasets, such that analyzing all data in a
common location might be impractical or impossible. Meta-analytic methods
provide a way to overcome this issue, by combining aggregated quantities like
model parameters or risk ratios. Most meta-analytic tools focus on parametric
statistical models, and methods for meta-analyzing semi-parametric models like
generalized additive models have not been well developed. Parametric models are
often not appropriate in neuroimaging, where for instance age-brain
relationships may take forms that are difficult to accurately describe using
such models. In this paper we introduce meta-GAM, a method for meta-analysis of
generalized additive models which does not require individual participant data,
and hence is suitable for increasing statistical power while upholding privacy
and other regulatory concerns. We extend previous works by enabling the
analysis of multiple model terms as well as multivariate smooth functions. In
addition, we show how meta-analytic $p$-values can be computed for smooth
terms. The proposed methods are shown to perform well in simulation
experiments, and are demonstrated in a real data analysis on hippocampal volume
and self-reported sleep quality data from the Lifebrain consortium. We argue
that application of meta-GAM is especially beneficial in lifespan neuroscience
and imaging genetics. The methods are implemented in an accompanying R package
\verb!metagam!, which is also demonstrated.
| stat |
Group Linear non-Gaussian Component Analysis with Applications to
Neuroimaging | Independent component analysis (ICA) is an unsupervised learning method
popular in functional magnetic resonance imaging (fMRI). Group ICA has been
used to search for biomarkers in neurological disorders including autism
spectrum disorder and dementia. However, current methods use a principal
component analysis (PCA) step that may remove low-variance features. Linear
non-Gaussian component analysis (LNGCA) enables simultaneous dimension
reduction and feature estimation including low-variance features in
single-subject fMRI. We present a group LNGCA model to extract group components
shared by more than one subject and subject-specific components. To determine
the total number of components in each subject, we propose a parametric
resampling test that samples spatially correlated Gaussian noise to match the
spatial dependence observed in data. In simulations, our estimated group
components achieve higher accuracy compared to group ICA. We apply our method
to a resting-state fMRI study on autism spectrum disorder in 342 children (252
typically developing, 90 with autism), where the group signals include
resting-state networks. We find examples of group components that appear to
exhibit different levels of temporal engagement in autism versus typically
developing children, as revealed using group LNGCA. This novel approach to
matrix decomposition is a promising direction for feature detection in
neuroimaging.
| stat |
Corrected score methods for estimating Bayesian networks with
error-prone nodes | Motivated by inferring cellular signaling networks using noisy flow cytometry
data, we develop procedures to draw inference for Bayesian networks based on
error-prone data. Two methods for inferring causal relationships between nodes
in a network are proposed based on penalized estimation methods that account
for measurement error and encourage sparsity. We discuss consistency of the
proposed network estimators and develop an approach for selecting the tuning
parameter in the penalized estimation methods. Empirical studies are carried
out to compare the proposed methods and a naive method that ignores measurement
error with applications to synthetic data and to single cell flow cytometry
data.
| stat |
Tensor Graphical Model: Non-convex Optimization and Statistical
Inference | We consider the estimation and inference of graphical models that
characterize the dependency structure of high-dimensional tensor-valued data.
To facilitate the estimation of the precision matrix corresponding to each way
of the tensor, we assume the data follow a tensor normal distribution whose
covariance has a Kronecker product structure. A critical challenge in the
estimation and inference of this model is the fact that its penalized maximum
likelihood estimation involves minimizing a non-convex objective function. To
address it, this paper makes two contributions: (i) In spite of the
non-convexity of this estimation problem, we prove that an alternating
minimization algorithm, which iteratively estimates each sparse precision
matrix while fixing the others, attains an estimator with an optimal
statistical rate of convergence. (ii) We propose a de-biased statistical
inference procedure for testing hypotheses on the true support of the sparse
precision matrices, and employ it for testing a growing number of hypothesis
with false discovery rate (FDR) control. The asymptotic normality of our test
statistic and the consistency of FDR control procedure are established. Our
theoretical results are backed up by thorough numerical studies and our real
applications on neuroimaging studies of Autism spectrum disorder and users'
advertising click analysis bring new scientific findings and business insights.
The proposed methods are encoded into a publicly available R package Tlasso.
| stat |
A Structured Approach to the Analysis of Remote Sensing Images | The number of studies for the analysis of remote sensing images has been
growing exponentially in the last decades. Many studies, however, only report
results---in the form of certain performance metrics---by a few selected
algorithms on a training and testing sample. While this often yields valuable
insights, it tells little about some important aspects. For example, one might
be interested in understanding the nature of a study by the interaction of
algorithm, features, and the sample as these collectively contribute to the
outcome; among these three, which would be a more productive direction in
improving a study; how to assess the sample quality or the value of a set of
features etc. With a focus on land-use classification, we advocate the use of a
structured analysis. The output of a study is viewed as the result of the
interplay among three input dimensions: feature, sample, and algorithm.
Similarly, another dimension, the error, can be decomposed into error along
each input dimension. Such a structural decomposition of the inputs or error
could help better understand the nature of the problem and potentially suggest
directions for improvement. We use the analysis of a remote sensing image at a
study site in Guangzhou, China, to demonstrate how such a structured analysis
could be carried out and what insights it generates. The structured analysis
could be applied to a new study, or as a diagnosis to an existing one. We
expect this will inform practice in the analysis of remote sensing images, and
help advance the state-of-the-art of land-use classification.
| stat |
Hawkes processes for credit indices time series analysis: How random are
trades arrival times? | Targeting a better understanding of credit market dynamics, the authors have
studied a stochastic model named the Hawkes process. Describing trades arrival
times, this kind of model allows for the capture of self-excitement and mutual
interactions phenomena. The authors propose here a simple yet conclusive method
for fitting multidimensional Hawkes processes with exponential kernels, based
on a maximum likelihood non-convex optimization. The method was successfully
tested on simulated data, then used on new publicly available real trading data
for three European credit indices, thus enabling quantification of
self-excitement as well as volume impacts or cross indices influences.
| stat |
Model-based Personalized Synthetic MR Imaging | Synthetic Magnetic Resonance (MR) imaging predicts images at new design
parameter settings from a few observed MR scans. Model-based methods, that use
both the physical and statistical properties underlying the MR signal and its
acquisition, can predict images at any setting from as few as three scans,
allowing it to be used in individualized patient- and anatomy-specific
contexts. However, the estimation problem in model-based synthetic MR imaging
is ill-posed and so regularization, in the form of correlated Gaussian Markov
Random Fields, is imposed on the voxel-wise spin-lattice relaxation time,
spin-spin relaxation time and the proton density underlying the MR image. We
develop theoretically sound but computationally practical matrix-free
estimation methods for synthetic MR imaging. Our evaluations demonstrate
excellent ability of our methods to synthetize MR images in a clinical
framework and also estimation and prediction accuracy and consistency. An added
strength of our model-based approach, also developed and illustrated here, is
the accurate estimation of standard errors of regional means in the synthesized
images.
| stat |
Principled selection of effect modifiers: Comments on 'Matching-adjusted
indirect comparisons: Application to time-to-event data' | In this commentary, we raise our concerns about a recent simulation study
conducted by Aouni, Gaudel-Dedieu and Sebastien, evaluating the performance of
different versions of matching-adjusted indirect comparison (MAIC). The
following points are highlighted: (1) making a clear distinction between
prognostic and effect-modifying covariates is important; (2) in the anchored
setting, MAIC is necessary where there are cross-trial imbalances in effect
modifiers; (3) the standard indirect comparison provides greater precision and
accuracy than MAIC if there are no effect modifiers in imbalance; (4) while the
target estimand of the simulation study is a conditional treatment effect, MAIC
targets a marginal or population-average treatment effect; (5) in MAIC,
variable selection is a problem of low dimensionality and sparsity-inducing
methods like the LASSO may induce bias; and (6) individual studies are
underpowered to detect interactions and data-driven approaches do not obviate
the necessity for subject matter knowledge when selecting effect modifiers.
| stat |
Deriving information from missing data: implications for mood prediction | The availability of mobile technologies has enabled the efficient collection
prospective longitudinal, ecologically valid self-reported mood data from
psychiatric patients. These data streams have potential for improving the
efficiency and accuracy of psychiatric diagnosis as well predicting future mood
states enabling earlier intervention. However, missing responses are common in
such datasets and there is little consensus as to how this should be dealt with
in practice. A signature-based method was used to capture different elements of
self-reported mood alongside missing data to both classify diagnostic group and
predict future mood in patients with bipolar disorder, borderline personality
disorder and healthy controls. The missing-response-incorporated
signature-based method achieves roughly 66\% correct diagnosis, with f1 scores
for three different clinic groups 59\% (bipolar disorder), 75\% (healthy
control) and 61\% (borderline personality disorder) respectively. This was
significantly more efficient than the naive model which excluded missing data.
Accuracies of predicting subsequent mood states and scores were also improved
by inclusion of missing responses. The signature method provided an effective
approach to the analysis of prospectively collected mood data where missing
data was common and should be considered as an approach in other similar
datasets.
| stat |
From unbiased MDI Feature Importance to Explainable AI for Trees | We attempt to give a unifying view of the various recent attempts to (i)
improve the interpretability of tree-based models and (ii) debias the the
default variable-importance measure in random Forests, Gini importance. In
particular, we demonstrate a common thread among the out-of-bag based bias
correction methods and their connection to local explanation for trees. In
addition, we point out a bias caused by the inclusion of inbag data in the
newly developed explainable AI for trees algorithms.
| stat |
Reducing Interference Bias in Online Marketplace Pricing Experiments | Online marketplace designers frequently run A/B tests to measure the impact
of proposed product changes. However, given that marketplaces are inherently
connected, total average treatment effect estimates obtained through Bernoulli
randomized experiments are often biased due to violations of the stable unit
treatment value assumption. This can be particularly problematic for
experiments that impact sellers' strategic choices, affect buyers' preferences
over items in their consideration set, or change buyers' consideration sets
altogether. In this work, we measure and reduce bias due to interference in
online marketplace experiments by using observational data to create clusters
of similar listings, and then using those clusters to conduct
cluster-randomized field experiments. We provide a lower bound on the magnitude
of bias due to interference by conducting a meta-experiment that randomizes
over two experiment designs: one Bernoulli randomized, one cluster randomized.
In both meta-experiment arms, treatment sellers are subject to a different
platform fee policy than control sellers, resulting in different prices for
buyers. By conducting a joint analysis of the two meta-experiment arms, we find
a large and statistically significant difference between the total average
treatment effect estimates obtained with the two designs, and estimate that
32.60% of the Bernoulli-randomized treatment effect estimate is due to
interference bias. We also find weak evidence that the magnitude and/or
direction of interference bias depends on extent to which a marketplace is
supply- or demand-constrained, and analyze a second meta-experiment to
highlight the difficulty of detecting interference bias when treatment
interventions require intention-to-treat analysis.
| 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.