title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Semiparametric point process modelling of blinking artifacts in PALM | Photoactivated localization microscopy (PALM) is a powerful imaging technique
for characterization of protein organization in biological cells. Due to the
stochastic blinking of fluorescent probes, and camera discretization effects,
each protein gives rise to a cluster of artificial observations. These blinking
artifacts are an obstacle for qualitative analysis of PALM data, and tools for
their correction are in high demand. We develop the Independent Blinking
Cluster point process (IBCpp) family of models, and present results on the mark
correlation function. We then construct the PALM-IBCpp - a semiparametric IBCpp
tailored for PALM data. We describe a procedure for estimation of parameters,
which can be used without parametric assumptions on the spatial organization of
proteins. The parameters include the kinetic rates that control blinking, and
as such can be used to correct subsequent data analysis. The method is
demonstrated on real data, and in a simulation study, where blinking artifacts
were precisely quantified in a range of realistic settings.
| stat |
Expanding tidy data principles to facilitate missing data exploration,
visualization and assessment of imputations | Despite the large body of research on missing value distributions and
imputation, there is comparatively little literature with a focus on how to
make it easy to handle, explore, and impute missing values in data. This paper
addresses this gap. The new methodology builds upon tidy data principles, with
the goal of integrating missing value handling as a key part of data analysis
workflows. We define a new data structure, and a suite of new operations.
Together, these provide a connected framework for handling, exploring, and
imputing missing values. These methods are available in the R package `naniar`.
| stat |
Estimating the sample mean and standard deviation from commonly reported
quantiles in meta-analysis | Researchers increasingly use meta-analysis to synthesize the results of
several studies in order to estimate a common effect. When the outcome variable
is continuous, standard meta-analytic approaches assume that the primary
studies report the sample mean and standard deviation of the outcome. However,
when the outcome is skewed, authors sometimes summarize the data by reporting
the sample median and one or both of (i) the minimum and maximum values and
(ii) the first and third quartiles, but do not report the mean or standard
deviation. To include these studies in meta-analysis, several methods have been
developed to estimate the sample mean and standard deviation from the reported
summary data. A major limitation of these widely used methods is that they
assume that the outcome distribution is normal, which is unlikely to be tenable
for studies reporting medians. We propose two novel approaches to estimate the
sample mean and standard deviation when data are suspected to be non-normal.
Our simulation results and empirical assessments show that the proposed methods
often perform better than the existing methods when applied to non-normal data.
| stat |
Kernel Two-Sample and Independence Tests for Non-Stationary Random
Processes | Two-sample and independence tests with the kernel-based MMD and HSIC have
shown remarkable results on i.i.d. data and stationary random processes.
However, these statistics are not directly applicable to non-stationary random
processes, a prevalent form of data in many scientific disciplines. In this
work, we extend the application of MMD and HSIC to non-stationary settings by
assuming access to independent realisations of the underlying random process.
These realisations - in the form of non-stationary time-series measured on the
same temporal grid - can then be viewed as i.i.d. samples from a multivariate
probability distribution, to which MMD and HSIC can be applied. We further show
how to choose suitable kernels over these high-dimensional spaces by maximising
the estimated test power with respect to the kernel hyper-parameters. In
experiments on synthetic data, we demonstrate superior performance of our
proposed approaches in terms of test power when compared to current
state-of-the-art functional or multivariate two-sample and independence tests.
Finally, we employ our methods on a real socio-economic dataset as an example
application.
| stat |
Supervised Learning for Multi-Block Incomplete Data | In the supervised high dimensional settings with a large number of variables
and a low number of individuals, one objective is to select the relevant
variables and thus to reduce the dimension. That subspace selection is often
managed with supervised tools. However, some data can be missing, compromising
the validity of the sub-space selection. We propose a Partial Least Square
(PLS) based method, called Multi-block Data-Driven sparse PLS mdd-sPLS,
allowing jointly variable selection and subspace estimation while training and
testing missing data imputation through a new algorithm called Koh-Lanta. This
method was challenged through simulations against existing methods such as mean
imputation, nipals, softImpute and imputeMFA. In the context of supervised
analysis of high dimensional data, the proposed method shows the lowest
prediction error of the response variables. So far this is the only method
combining data imputation and response variable prediction. The superiority of
the supervised multi-block mdd-sPLS method increases with the intra-block and
inter-block correlations. The application to a real data-set from a rVSV-ZEBOV
Ebola vaccine trial revealed interesting and biologically relevant results. The
method is implemented in a R-package available on the CRAN and a Python-package
available on pypi.
| stat |
Simultaneous prediction and community detection for networks with
application to neuroimaging | Community structure in networks is observed in many different domains, and
unsupervised community detection has received a lot of attention in the
literature. Increasingly the focus of network analysis is shifting towards
using network information in some other prediction or inference task rather
than just analyzing the network itself. In particular, in neuroimaging
applications brain networks are available for multiple subjects and the goal is
often to predict a phenotype of interest. Community structure is well known to
be a feature of brain networks, typically corresponding to different regions of
the brain responsible for different functions. There are standard parcellations
of the brain into such regions, usually obtained by applying clustering methods
to brain connectomes of healthy subjects. However, when the goal is predicting
a phenotype or distinguishing between different conditions, these static
communities from an unrelated set of healthy subjects may not be the most
useful for prediction. Here we present a method for supervised community
detection, aiming to find a partition of the network into communities that is
most useful for predicting a particular response. We use a block-structured
regularization penalty combined with a prediction loss function, and compute
the solution with a combination of a spectral method and an ADMM optimization
algorithm. We show that the spectral clustering method recovers the correct
communities under a weighted stochastic block model. The method performs well
on both simulated and real brain networks, providing support for the idea of
task-dependent brain regions.
| stat |
Learning high-dimensional probability distributions using tree tensor
networks | We consider the problem of the estimation of a high-dimensional probability
distribution using model classes of functions in tree-based tensor formats, a
particular case of tensor networks associated with a dimension partition tree.
The distribution is assumed to admit a density with respect to a product
measure, possibly discrete for handling the case of discrete random variables.
After discussing the representation of classical model classes in tree-based
tensor formats, we present learning algorithms based on empirical risk
minimization using a $L^2$ contrast.
These algorithms exploit the multilinear parametrization of the formats to
recast the nonlinear minimization problem into a sequence of empirical risk
minimization problems with linear models. A suitable parametrization of the
tensor in tree-based tensor format allows to obtain a linear model with
orthogonal bases, so that each problem admits an explicit expression of the
solution and cross-validation risk estimates. These estimations of the risk
enable the model selection, for instance when exploiting sparsity in the
coefficients of the representation.
A strategy for the adaptation of the tensor format (dimension tree and
tree-based ranks) is provided, which allows to discover and exploit some
specific structures of high-dimensional probability distributions such as
independence or conditional independence.
We illustrate the performances of the proposed algorithms for the
approximation of classical probabilistic models (such as Gaussian distribution,
graphical models, Markov chain).
| stat |
Bayesian Inference in Nonparanormal Graphical Models | Gaussian graphical models have been used to study intrinsic dependence among
several variables, but the Gaussianity assumption may be restrictive in many
applications. A nonparanormal graphical model is a semiparametric
generalization for continuous variables where it is assumed that the variables
follow a Gaussian graphical model only after some unknown smooth monotone
transformations on each of them. We consider a Bayesian approach in the
nonparanormal graphical model by putting priors on the unknown transformations
through a random series based on B-splines where the coefficients are ordered
to induce monotonicity. A truncated normal prior leads to partial conjugacy in
the model and is useful for posterior simulation using Gibbs sampling. On the
underlying precision matrix of the transformed variables, we consider a
spike-and-slab prior and use an efficient posterior Gibbs sampling scheme. We
use the Bayesian Information Criterion to choose the hyperparameters for the
spike-and-slab prior. We present a posterior consistency result on the
underlying transformation and the precision matrix. We study the numerical
performance of the proposed method through an extensive simulation study and
finally apply the proposed method on a real data set.
| stat |
Iterative Spectral Method for Alternative Clustering | Given a dataset and an existing clustering as input, alternative clustering
aims to find an alternative partition. One of the state-of-the-art approaches
is Kernel Dimension Alternative Clustering (KDAC). We propose a novel Iterative
Spectral Method (ISM) that greatly improves the scalability of KDAC. Our
algorithm is intuitive, relies on easily implementable spectral decompositions,
and comes with theoretical guarantees. Its computation time improves upon
existing implementations of KDAC by as much as 5 orders of magnitude.
| stat |
Inferring linear and nonlinear Interaction networks using neighborhood
support vector machines | In this paper, we consider modelling interaction between a set of variables
in the context of time series and high dimension. We suggest two approaches.
The first is similar to the neighborhood lasso when the lasso model is replaced
by a support vector machine (SVMs). The second is a restricted Bayesian network
adapted for time series. We show the efficiency of our approaches by
simulations using linear, nonlinear data set and a mixture of both.
| stat |
Gaussian Universal Likelihood Ratio Testing | The likelihood ratio test (LRT) based on the asymptotic chi-squared
distribution of the log likelihood is one of the fundamental tools of
statistical inference. A recent universal LRT approach based on sample
splitting provides valid hypothesis tests and confidence sets in any setting
for which we can compute the split likelihood ratio statistic (or, more
generally, an upper bound on the null maximum likelihood). The universal LRT is
valid in finite samples and without regularity conditions. This test empowers
statisticians to construct tests in settings for which no valid hypothesis test
previously existed. For the simple but fundamental case of testing the
population mean of d-dimensional Gaussian data, the usual LRT itself applies
and thus serves as a perfect test bed to compare against the universal LRT.
This work presents the first in-depth exploration of the size, power, and
relationships between several universal LRT variants. We show that a repeated
subsampling approach is the best choice in terms of size and power. We observe
reasonable performance even in a high-dimensional setting, where the expected
squared radius of the best universal LRT confidence set is approximately 3/2
times the squared radius of the standard LRT-based set. We illustrate the
benefits of the universal LRT through testing a non-convex doughnut-shaped null
hypothesis, where a universal inference procedure can have higher power than a
standard approach.
| stat |
Canonical spectral representation for exchangeable max-stable sequences | The set of infinite-dimensional, symmetric stable tail dependence functions
associated with exchangeable max-stable sequences of random variables with unit
Fr\'echet margins is shown to be a simplex. Except for a single element, the
extremal boundary is in one-to-one correspondence with the set of distribution
functions of non-negative random variables with unit mean. Consequently, each
element is uniquely represented by a pair of a constant and a probability
measure on the space of distribution functions of non-negative random variables
with unit mean. A canonical stochastic construction for arbitrary exchangeable
max-stable sequences and a stochastic representation for the Pickands
dependence measure of finite-dimensional margins are immediate corollaries. As
by-products, a canonical analytical description and an associated canonical Le
Page series representation for non-decreasing stochastic processes that are
strongly infinitely divisible with respect to time are obtained.
| stat |
Least Squares Approximation for a Distributed System | In this work, we develop a distributed least squares approximation (DLSA)
method that is able to solve a large family of regression problems (e.g.,
linear regression, logistic regression, and Cox's model) on a distributed
system. By approximating the local objective function using a local quadratic
form, we are able to obtain a combined estimator by taking a weighted average
of local estimators. The resulting estimator is proved to be statistically as
efficient as the global estimator. Moreover, it requires only one round of
communication. We further conduct a shrinkage estimation based on the DLSA
estimation using an adaptive Lasso approach. The solution can be easily
obtained by using the LARS algorithm on the master node. It is theoretically
shown that the resulting estimator possesses the oracle property and is
selection consistent by using a newly designed distributed Bayesian information
criterion (DBIC). The finite sample performance and computational efficiency
are further illustrated by an extensive numerical study and an airline dataset.
The airline dataset is 52 GB in size. The entire methodology has been
implemented in Python for a {\it de-facto} standard Spark system. The proposed
DLSA algorithm on the Spark system takes 26 minutes to obtain a logistic
regression estimator, which is more efficient and memory friendly than
conventional methods.
| stat |
Estimation of Poisson Autoregressive Model for Multiple Time Series | A Poisson autoregressive (PAR) model accounting for discreteness and
autocorrelation of count time series data is typically estimated in the
state-space modelling framework through extended Kalman filter. However,
because of the complex dependencies in count time series, estimation becomes
more challenging. PAR is viewed as an additive model and estimated using a
hybrid of cubic smoothing splines and maximum likelihood estimation (MLE) in
the backfitting framework. Simulation studies show that this estimation method
is comparable or better than PAR estimated in the state-space context,
especially with larger count values. However, as [2] formulated PAR for
stationary counts, both estimation procedures underestimate parameters in
nearly nonstationary models. The flexibility of the additive model has two
benefits though: robust estimation in the presence of temporary structural
change, and; viability to integrate PAR model into a more complex model
structure. We further generalized the PAR(p) model into multiple time series of
counts and illustrated with indicators in the financial markets.
| stat |
Learning Prediction Intervals for Regression: Generalization and
Calibration | We study the generation of prediction intervals in regression for uncertainty
quantification. This task can be formalized as an empirical constrained
optimization problem that minimizes the average interval width while
maintaining the coverage accuracy across data. We strengthen the existing
literature by studying two aspects of this empirical optimization. First is a
general learning theory to characterize the optimality-feasibility tradeoff
that encompasses Lipschitz continuity and VC-subgraph classes, which are
exemplified in regression trees and neural networks. Second is a calibration
machinery and the corresponding statistical theory to optimally select the
regularization parameter that manages this tradeoff, which bypasses the
overfitting issues in previous approaches in coverage attainment. We
empirically demonstrate the strengths of our interval generation and
calibration algorithms in terms of testing performances compared to existing
benchmarks.
| stat |
Sample selection from a given dataset to validate machine learning
models | The selection of a validation basis from a full dataset is often required in
industrial use of supervised machine learning algorithm. This validation basis
will serve to realize an independent evaluation of the machine learning model.
To select this basis, we propose to adopt a "design of experiments" point of
view, by using statistical criteria. We show that the "support points" concept,
based on Maximum Mean Discrepancy criteria, is particularly relevant. An
industrial test case from the company EDF illustrates the practical interest of
the methodology.
| stat |
Bayesian Inference for Stationary Points in Gaussian Process Regression
Models for Event-Related Potentials Analysis | Stationary points embedded in the derivatives are often critical for a model
to be interpretable and may be considered as key features of interest in many
applications. We propose a semiparametric Bayesian model to efficiently infer
the locations of stationary points of a nonparametric function, while treating
the function itself as a nuisance parameter. We use Gaussian processes as a
flexible prior for the underlying function and impose derivative constraints to
control the function's shape via conditioning. We develop an inferential
strategy that intentionally restricts estimation to the case of at least one
stationary point, bypassing possible mis-specifications in the number of
stationary points and avoiding the varying dimension problem that often brings
in computational complexity. We illustrate the proposed methods using
simulations and then apply the method to the estimation of event-related
potentials (ERP) derived from electroencephalography (EEG) signals. We show how
the proposed method automatically identifies characteristic components and
their latencies at the individual level, which avoids the excessive averaging
across subjects which is routinely done in the field to obtain smooth curves.
By applying this approach to EEG data collected from younger and older adults
during a speech perception task, we are able to demonstrate how the time course
of speech perception processes changes with age.
| stat |
A new approach to the gender pay gap decomposition by economic activity | The aim of this paper is to present an original approach to estimate the
gender pay gap. We propose a model-based decomposition, similar to the most
popular approaches, where the first component measures differences in group
characteristics and the second component measures the unexplained effect; the
latter being the real gap. The novel approach incorporates model selection and
bias correction. %, avoiding the main limitation of standard approaches, which
is the dependence on the choice of explanatory variables and the functional
form in regression. The pay gap problem in a small area context is considered
in this paper, although the approach is flexible to be applied to other
contexts.
Specifically, the methodology is validated for analysing wage differentials
by economic activities in the region of Galicia (Spain) and by analysing
simulated data from an experimental design that imitates the generation of real
data. The good performance of the proposed estimators is shown in both cases,
specifically when compared with those obtained from the widely used
Oaxaca-Blinder approach.
| stat |
Testing weak nulls in matched observational studies | We develop sensitivity analyses for weak nulls in matched observational
studies while allowing unit-level treatment effects to vary. The methods may be
applied to studies using any optimal without-replacement matching algorithm. In
contrast to randomized experiments and to paired observational studies, we show
for general matched designs that over a large class of test statistics, any
valid sensitivity analysis for the entirety of the weak null must be
unnecessarily conservative if Fisher's sharp null of no treatment effect for
any individual also holds. We present a sensitivity analysis valid for the weak
null, and illustrate why it is generally conservative if the sharp null holds
through new connections to inverse probability weighted estimators. An
alternative procedure is presented that is asymptotically sharp if treatment
effects are constant, and that is valid for the weak null under additional
restrictions which may be deemed benign by practitioners. Simulations
demonstrate that this alternative procedure results in a valid sensitivity
analysis for the weak null hypothesis under a host of reasonable
data-generating processes. The procedures allow practitioners to assess
robustness of estimated sample average treatment effects to hidden bias while
allowing for unspecified effect heterogeneity in matched observational studies.
| stat |
Global Fire Season Severity Analysis and Forecasting | In this paper, we divide the globe into a hexagonal grid and we extracted
time series of daily fire counts from each cell to estimate and analyze
worldwide fire season severity (FSS), here defined as the accumulated fire
detections in a season. The central question here is evaluating the accuracy of
time series forecasting methods to estimate short-term (months) and medium-term
(seasons) using only historical data of active fire detections. This approach
is simple, fast, and use globally available data, making it easier for large
scale prediction. Our results comprehend descriptive and predictive analyses of
the worldwide seasonal fire activity. We verified that in 99% of the cells, the
fire seasons have lengths shorter than seven months and that 57% have their
lengths decrease. We also observed a declining tendency in the number of active
fire counts during the seasons in 61% cells. However, some regions like the
Northeast Brazil and the West Coast of the USA present an increasing trend. We
verified that the forecasting error is lower than the mean FSS in 95% of the
cells, indicating clear predictability in the FSS.
| stat |
Assessing Biosimilarity using Functional Metrics | In recent years there have been a lot of interest to test for similarity
between biological drug products, commonly known as biologics. Biologics are
large and complex molecule drugs that are produced by living cells and hence
these are sensitive to the environmental changes. In addition, biologics
usually induce antibodies which raises the safety and efficacy issues. The
manufacturing process is also much more complicated and often costlier than the
small-molecule generic drugs. Because of these complexities and inherent
variability of the biologics, the testing paradigm of the traditional generic
drugs cannot be directly used to test for biosimilarity. Taking into account
some of these concerns we propose a functional distance based methodology that
takes into consideration the entire time course of the study and is based on a
class of flexible semi-parametric models. The empirical results show that the
proposed approach is more sensitive than the classical equivalence tests
approach which are usually based on arbitrarily chosen time point. Bootstrap
based methodologies are also presented for statistical inference.
| stat |
Gaussian process surrogate modeling with manipulating factors for carbon
nanotube growth experiments | This paper presents a new Gaussian process (GP) surrogate modeling for
predicting the outcome of a physical experiment where some experimental inputs
are controlled by other manipulating factors. Particularly, we are interested
in the case where the control precision is not very high, so the input factor
values vary significantly even under the same setting of the corresponding
manipulating factors. The case is observed in our main application to carbon
nanotube growth experiments, where one experimental input among many is
manipulated by another manipulating factors, and the relation between the input
and the manipulating factors significantly varies in the dates and times of
operations. Due to this variation, the standard GP surrogate that directly
relates the manipulating factors to the experimental outcome does not provide a
great predictive power on the outcome. At the same time, the GP model relating
the main factors to the outcome directly is not appropriate for the prediction
purpose because the main factors cannot be accurately set as planned for a
future experiment. Motivated by the carbon nanotube example, we propose a
two-tiered GP model, where the bottom tier relates the manipulating factors to
the corresponding main factors with potential biases and variation independent
of the manipulating factors, and the top tier relates the main factors to the
experimental outcome. Our two-tier model explicitly models the propagation of
the control uncertainty to the experimental outcome through the two GP modeling
tiers. We present the inference and hyper-parameter estimation of the proposed
model. The proposed approach is illustrated with the motivating example of a
closed-loop autonomous research system for carbon nanotube growth experiments,
and the test results are reported with the comparison to a benchmark method,
i.e. a standard GP model.
| stat |
Teaching Split Plot Experiments With a Boomerang Tin | This article presents an example on how to teach split-plot experimental
designs based on a hands-on exercise. This is a toy called boomerang tin which
utilizes a rubber band to store and release energy. The set up and mechanisms
of the hands-on example are explained followed by a description of how it can
be leveraged in teaching split plot DoE. An actual design is set up, analyzed
and discussed to provide reference for its usage in class.
| stat |
Geographically-dependent individual-level models for infectious diseases
transmission | Infectious disease models can be of great use for understanding the
underlying mechanisms that influence the spread of diseases and predicting
future disease progression. Modeling has been increasingly used to evaluate the
potential impact of different control measures and to guide public health
policy decisions. In recent years, there has been rapid progress in developing
spatio-temporal modeling of infectious diseases and an example of such recent
developments is the discrete time individual-level models (ILMs). These models
are well developed and provide a common framework for modeling many disease
systems, however, they assume the probability of disease transmission between
two individuals depends only on their spatial separation and not on their
spatial locations. In cases where spatial location itself is important for
understanding the spread of emerging infectious diseases and identifying their
causes, it would be beneficial to incorporate the effect of spatial location in
the model. In this study, we thus generalize the ILMs to a new class of
geographically-dependent ILMs (GD-ILMs), to allow for the evaluation of the
effect of spatially varying risk factors (e.g., education, environmental), as
well as unobserved spatial structure, upon the transmission of infectious
disease. Specifically, we consider a conditional autoregressive model to
capture the effects of unobserved spatially structured latent covariates or
measurement error. This results in flexible infectious disease models that can
be used for formulating etiological hypotheses and identifying geographical
regions of unusually high risk to formulate preventive action. The reliability
of these models are investigated on a combination of simulated epidemic data
and Alberta seasonal influenza outbreak data (2009). This new class of models
is fitted to data within a Bayesian statistical framework using MCMC methods.
| stat |
A Strongly Consistent Sparse $k$-means Clustering with Direct $l_1$
Penalization on Variable Weights | We propose the Lasso Weighted $k$-means ($LW$-$k$-means) algorithm as a
simple yet efficient sparse clustering procedure for high-dimensional data
where the number of features ($p$) can be much larger compared to the number of
observations ($n$). In the $LW$-$k$-means algorithm, we introduce a lasso-based
penalty term, directly on the feature weights to incorporate feature selection
in the framework of sparse clustering. $LW$-$k$-means does not make any
distributional assumption of the given dataset and thus, induces a
non-parametric method for feature selection. We also analytically investigate
the convergence of the underlying optimization procedure in $LW$-$k$-means and
establish the strong consistency of our algorithm. $LW$-$k$-means is tested on
several real-life and synthetic datasets and through detailed experimental
analysis, we find that the performance of the method is highly competitive
against some state-of-the-art procedures for clustering and feature selection,
not only in terms of clustering accuracy but also with respect to computational
time.
| stat |
General linear-time inference for Gaussian Processes on one dimension | Gaussian Processes (GPs) provide a powerful probabilistic framework for
interpolation, forecasting, and smoothing, but have been hampered by
computational scaling issues. Here we prove that for data sampled on one
dimension (e.g., a time series sampled at arbitrarily-spaced intervals),
approximate GP inference at any desired level of accuracy requires
computational effort that scales linearly with the number of observations; this
new theorem enables inference on much larger datasets than was previously
feasible. To achieve this improved scaling we propose a new family of
stationary covariance kernels: the Latent Exponentially Generated (LEG) family,
which admits a convenient stable state-space representation that allows
linear-time inference. We prove that any continuous integrable stationary
kernel can be approximated arbitrarily well by some member of the LEG family.
The proof draws connections to Spectral Mixture Kernels, providing new insight
about the flexibility of this popular family of kernels. We propose
parallelized algorithms for performing inference and learning in the LEG model,
test the algorithm on real and synthetic data, and demonstrate scaling to
datasets with billions of samples.
| stat |
Robust bootstrap prediction intervals for univariate and multivariate
autoregressive time series models | The bootstrap procedure has emerged as a general framework to construct
prediction intervals for future observations in autoregressive time series
models. Such models with outlying data points are standard in real data
applications, especially in the field of econometrics. These outlying data
points tend to produce high forecast errors, which reduce the forecasting
performances of the existing bootstrap prediction intervals calculated based on
non-robust estimators. In the univariate and multivariate autoregressive time
series, we propose a robust bootstrap algorithm for constructing prediction
intervals and forecast regions. The proposed procedure is based on the weighted
likelihood estimates and weighted residuals. Its finite sample properties are
examined via a series of Monte Carlo studies and two empirical data examples.
| stat |
Fast Exact Univariate Kernel Density Estimation | This paper presents new methodology for computationally efficient kernel
density estimation. It is shown that a large class of kernels allows for exact
evaluation of the density estimates using simple recursions. The same
methodology can be used to compute density derivative estimates exactly. Given
an ordered sample the computational complexity is linear in the sample size.
Combining the proposed methodology with existing approximation methods results
in extremely fast density estimation. Extensive experimentation documents the
effectiveness and efficiency of this approach compared with the existing
state-of-the-art.
| stat |
Nonparametric Estimation of Heterogeneous Treatment Effects: From Theory
to Learning Algorithms | The need to evaluate treatment effectiveness is ubiquitous in most of
empirical science, and interest in flexibly investigating effect heterogeneity
is growing rapidly. To do so, a multitude of model-agnostic, nonparametric
meta-learners have been proposed in recent years. Such learners decompose the
treatment effect estimation problem into separate sub-problems, each solvable
using standard supervised learning methods. Choosing between different
meta-learners in a data-driven manner is difficult, as it requires access to
counterfactual information. Therefore, with the ultimate goal of building
better understanding of the conditions under which some learners can be
expected to perform better than others a priori, we theoretically analyze four
broad meta-learning strategies which rely on plug-in estimation and
pseudo-outcome regression. We highlight how this theoretical reasoning can be
used to guide principled algorithm design and translate our analyses into
practice by considering a variety of neural network architectures as
base-learners for the discussed meta-learning strategies. In a simulation
study, we showcase the relative strengths of the learners under different
data-generating processes.
| stat |
Partial Recovery in the Graph Alignment Problem | In this paper, we consider the graph alignment problem, which is the problem
of recovering, given two graphs, a one-to-one mapping between nodes that
maximizes edge overlap. This problem can be viewed as a noisy version of the
well-known graph isomorphism problem and appears in many applications,
including social network deanonymization and cellular biology. Our focus here
is on \emph{partial recovery}, i.e., we look for a one-to-one mapping which is
correct on a fraction of the nodes of the graph rather than on all of them, and
we assume that the two input graphs to the problem are correlated
Erd\H{o}s-R\'enyi graphs of parameters $(n,q,s)$. Our main contribution is then
to give necessary and sufficient conditions on $(n,q,s)$ under which partial
recovery is possible with high probability as the number of nodes $n$ goes to
infinity. In particular, we show that it is possible to achieve partial
recovery in the $nqs=\Theta(1)$ regime under certain additional assumptions.
| stat |
Modeling the social media relationships of Irish politicians using a
generalized latent space stochastic blockmodel | D\'ail \'Eireann is the principal chamber of the Irish parliament. The 31st
D\'ail \'Eireann is the principal chamber of the Irish parliament. The 31st
D\'ail was in session from March 11th, 2011 to February 6th, 2016. Many of the
members of the D\'ail were active on social media and many were Twitter users
who followed other members of the D\'ail. The pattern of following amongst
these politicians provides insights into political alignment within the D\'ail.
We propose a new model, called the generalized latent space stochastic
blockmodel, which extends and generalizes both the latent space model and the
stochastic blockmodel to study social media connections between members of the
D\'ail. The probability of an edge between two nodes in a network depends on
their respective class labels as well as latent positions in an unobserved
latent space. The proposed model is capable of representing transitivity,
clustering, as well as disassortative mixing. A Bayesian method with Markov
chain Monte Carlo sampling is proposed for estimation of model parameters.
Model selection is performed using the WAIC criterion and models of different
number of classes or dimensions of latent space are compared. We use the model
to study Twitter following relationships of members of the D\'ail and interpret
structure found in these relationships. We find that the following
relationships amongst politicians is mainly driven by past and present
political party membership. We also find that the modeling outputs are
informative when studying voting within the D\'ail.
| stat |
Bayesian Factor Analysis for Inference on Interactions | This article is motivated by the problem of inference on interactions among
chemical exposures impacting human health outcomes. Chemicals often co-occur in
the environment or in synthetic mixtures and as a result exposure levels can be
highly correlated. We propose a latent factor joint model, which includes
shared factors in both the predictor and response components while assuming
conditional independence. By including a quadratic regression in the latent
variables in the response component, we induce flexible dimension reduction in
characterizing main effects and interactions. We propose a Bayesian approach to
inference under this Factor analysis for INteractions (FIN) framework. Through
appropriate modifications of the factor modeling structure, FIN can accommodate
higher order interactions and multivariate outcomes. We provide theory on
posterior consistency and the impact of misspecifying the number of factors. We
evaluate the performance using a simulation study and data from the National
Health and Nutrition Examination Survey (NHANES). Code is available on GitHub.
| stat |
An Efficient Sampling Algorithm for Non-smooth Composite Potentials | We consider the problem of sampling from a density of the form $p(x) \propto
\exp(-f(x)- g(x))$, where $f: \mathbb{R}^d \rightarrow \mathbb{R}$ is a smooth
and strongly convex function and $g: \mathbb{R}^d \rightarrow \mathbb{R}$ is a
convex and Lipschitz function. We propose a new algorithm based on the
Metropolis-Hastings framework, and prove that it mixes to within TV distance
$\varepsilon$ of the target density in at most $O(d \log (d/\varepsilon))$
iterations. This guarantee extends previous results on sampling from
distributions with smooth log densities ($g = 0$) to the more general composite
non-smooth case, with the same mixing time up to a multiple of the condition
number. Our method is based on a novel proximal-based proposal distribution
that can be efficiently computed for a large class of non-smooth functions $g$.
| stat |
Flexible models for nonstationary dependence: Methodology and examples | There are many situations when modelling environmental phenomena for which it
is not appropriate to assume a stationary dependence structure.
\cite{sampson1992} proposed an approach to allowing nonstationarity in
dependence based on a deformed space: coordinates from original geographic
"$G$" space are mapped to a new dispersion "$D$" space in which stationary
dependence is a reasonable assumption. \cite{sampson1992} achieve this with two
deformation functions, which are chosen as thin plate splines, each
representing how one of the two coordinates in $D$-space relates to the
original $G$-space coordinates. This works extends the deformation approach,
and the dimension expansion approach of \cite{bornn2012}, to a regression-based
framework in which all dimensions in $D$-space are treated as "smooths" as
found, for example, in generalized additive models. The framework offers an
intuitive and user-friendly approach to specifying $D$-space, allows different
levels of smoothing for dimensions in $D$-space, and allows objective inference
for all model parameters. Furthermore, a numerical approach is proposed to
avoid non-bijective deformations, should they occur, which applies to any
deformation. The proposed framework is demonstrated on the solar radiation data
studied in \cite{sampson1992}, and then on an example related to risk analysis,
which culminates in producing simulations of extreme rainfall for part of
Colorado, US.
| stat |
Bayesian Inference for Johnson's SB and Weibull distributions | The four-parameter Johnson's SB (JSB) and three-parameter Weibull
distributions have received much attention in the field of forestry for
characterizing diameters at breast height (DBH). In this work, we suggest the
Bayesian method for estimating parameters of the JBS distribution. The maximum
likelihood approach uses iterative methods such as Newton-Raphson (NR)
algorithm for maximizing the logarithm of the likelihood function. But there is
no guarantee that the NR method converges. This fact that the NR method for
estimating the parameters of the JSB distribution sometimes fails to converge
was verified through simulation in this study. Further, it was shown that the
Bayesian estimators presented in this work were robust with respect to the
initial values and estimate the parameters of the JSB distribution efficiently.
The performance of the JSB and three-parameter Weibull distributions was
compared in a Bayesian paradigm when these models were fitted to DBH data of
three plots that randomly selected from a study established in 107 plots of
mixed-age ponderosa pine (Pinus ponderosa Dougl. ex Laws.) with scattered
western junipers at the Malheur National Forest in south end of the Blue
Mountains near Burns, Oregon, USA. Bayesian paradigm demonstrated that JBS was
superior model than the three-parameter Weibull for characterizing the DBH
distribution when these models were fitted to the DBH data of the three plots.
| stat |
Anchored Causal Inference in the Presence of Measurement Error | We consider the problem of learning a causal graph in the presence of
measurement error. This setting is for example common in genomics, where gene
expression is corrupted through the measurement process. We develop a provably
consistent procedure for estimating the causal structure in a linear Gaussian
structural equation model from corrupted observations on its nodes, under a
variety of measurement error models. We provide an estimator based on the
method-of-moments, which can be used in conjunction with constraint-based
causal structure discovery algorithms. We prove asymptotic consistency of the
procedure and also discuss finite-sample considerations. We demonstrate our
method's performance through simulations and on real data, where we recover the
underlying gene regulatory network from zero-inflated single-cell RNA-seq data.
| stat |
Distributionally Robust Selection of the Best | Specifying a proper input distribution is often a challenging task in
simulation modeling. In practice, there may be multiple plausible distributions
that can fit the input data reasonably well, especially when the data volume is
not large. In this paper, we consider the problem of selecting the best from a
finite set of simulated alternatives, in the presence of such input
uncertainty. We model such uncertainty by an ambiguity set consisting of a
finite number of plausible input distributions, and aim to select the
alternative with the best worst-case mean performance over the ambiguity set.
We refer to this problem as robust selection of the best (RSB). To solve the
RSB problem, we develop a two-stage selection procedure and a sequential
selection procedure; we then prove that both procedures can achieve at least a
user-specified probability of correct selection under mild conditions.
Extensive numerical experiments are conducted to investigate the computational
efficiency of the two procedures. Finally, we apply the RSB approach to study a
queueing system's staffing problem using synthetic data and an
appointment-scheduling problem using real data from a large hospital in China.
We find that the RSB approach can generate decisions significantly better than
other widely used approaches.
| stat |
Time Series Classification via Topological Data Analysis | In this paper, we develop topological data analysis methods for
classification tasks on univariate time series. As an application we perform
binary and ternary classification tasks on two public datasets that consist of
physiological signals collected under stress and non-stress conditions. We
accomplish our goal by using persistent homology to engineer stable topological
features after we use a time delay embedding of the signals and perform a
subwindowing instead of using windows of fixed length. The combination of
methods we use can be applied to any univariate time series and in this
application allows us to reduce noise and use long window sizes without
incurring an extra computational cost. We then use machine learning models on
the features we algorithmically engineered to obtain higher accuracies with
fewer features.
| stat |
Statistical inference in two-sample summary-data Mendelian randomization
using robust adjusted profile score | Mendelian randomization (MR) is a method of exploiting genetic variation to
unbiasedly estimate a causal effect in presence of unmeasured confounding. MR
is being widely used in epidemiology and other related areas of population
science. In this paper, we study statistical inference in the increasingly
popular two-sample summary-data MR design. We show a linear model for the
observed associations approximately holds in a wide variety of settings when
all the genetic variants satisfy the exclusion restriction assumption, or in
genetic terms, when there is no pleiotropy. In this scenario, we derive a
maximum profile likelihood estimator with provable consistency and asymptotic
normality. However, through analyzing real datasets, we find strong evidence of
both systematic and idiosyncratic pleiotropy in MR, echoing the omnigenic model
of complex traits that is recently proposed in genetics. We model the
systematic pleiotropy by a random effects model, where no genetic variant
satisfies the exclusion restriction condition exactly. In this case we propose
a consistent and asymptotically normal estimator by adjusting the profile
score. We then tackle the idiosyncratic pleiotropy by robustifying the adjusted
profile score. We demonstrate the robustness and efficiency of the proposed
methods using several simulated and real datasets.
| stat |
Sparse, Low-bias, and Scalable Estimation of High Dimensional Vector
Autoregressive Models via Union of Intersections | Vector autoregressive (VAR) models are widely used for causal discovery and
forecasting in multivariate time series analyses in fields as diverse as
neuroscience, environmental science, and econometrics. In the high-dimensional
setting, model parameters are typically estimated by L1-regularized maximum
likelihood; yet, when applied to VAR models, this technique produces a sizable
trade-off between sparsity and bias with the choice of the regularization
hyperparameter, and thus between causal discovery and prediction. That is,
low-bias estimation entails dense parameter selection, and sparse selection
entails increased bias; the former is useful in forecasting but less likely to
yield scientific insight leading to discovery of causal influences, and
conversely for the latter. This paper presents a scalable algorithm for
simultaneous low-bias and low-variance estimation (hence good prediction) with
sparse selection for high-dimensional VAR models. The method leverages the
recently developed Union of Intersections (UoI) algorithmic framework for
flexible, modular, and scalable feature selection and estimation that allows
control of false discovery and false omission in feature selection while
maintaining low bias and low variance. This paper demonstrates the superior
performance of the UoI-VAR algorithm compared with other methods in simulation
studies, exhibits its application in data analysis, and illustrates its good
algorithmic scalability in multi-node distributed memory implementations.
| stat |
Reinforcement Learning for Dynamic Bidding in Truckload Markets: an
Application to Large-Scale Fleet Management with Advance Commitments | Truckload brokerages, a $100 billion/year industry in the U.S., plays the
critical role of matching shippers with carriers, often to move loads several
days into the future. Brokerages not only have to find companies that will
agree to move a load, the brokerage often has to find a price that both the
shipper and carrier will agree to. The price not only varies by shipper and
carrier, but also by the traffic lanes and other variables such as commodity
type. Brokerages have to learn about shipper and carrier response functions by
offering a price and observing whether each accepts the quote. We propose a
knowledge gradient policy with bootstrap aggregation for high-dimensional
contextual settings to guide price experimentation by maximizing the value of
information. The learning policy is tested using a carefully calibrated fleet
simulator that includes a stochastic lookahead policy that simulates fleet
movements, as well as the stochastic modeling of driver assignments and the
carrier's load commitment policies with advance booking.
| stat |
Convergence of adaptive algorithms for weakly convex constrained
optimization | We analyze the adaptive first order algorithm AMSGrad, for solving a
constrained stochastic optimization problem with a weakly convex objective. We
prove the $\mathcal{\tilde O}(t^{-1/4})$ rate of convergence for the norm of
the gradient of Moreau envelope, which is the standard stationarity measure for
this class of problems. It matches the known rates that adaptive algorithms
enjoy for the specific case of unconstrained smooth stochastic optimization.
Our analysis works with mini-batch size of $1$, constant first and second order
moment parameters, and possibly unbounded optimization domains. Finally, we
illustrate the applications and extensions of our results to specific problems
and algorithms.
| stat |
Adaptive Methods for Short-Term Electricity Load Forecasting During
COVID-19 Lockdown in France | The coronavirus disease 2019 (COVID-19) pandemic has urged many governments
in the world to enforce a strict lockdown where all nonessential businesses are
closed and citizens are ordered to stay at home. One of the consequences of
this policy is a significant change in electricity consumption patterns. Since
load forecasting models rely on calendar or meteorological information and are
trained on historical data, they fail to capture the significant break caused
by the lockdown and have exhibited poor performances since the beginning of the
pandemic. This makes the scheduling of the electricity production challenging,
and has a high cost for both electricity producers and grid operators. In this
paper we introduce adaptive generalized additive models using Kalman filters
and fine-tuning to adjust to new electricity consumption patterns.
Additionally, knowledge from the lockdown in Italy is transferred to anticipate
the change of behavior in France. The proposed methods are applied to forecast
the electricity demand during the French lockdown period, where they
demonstrate their ability to significantly reduce prediction errors compared to
traditional models. Finally expert aggregation is used to leverage the
specificities of each predictions and enhance results even further.
| stat |
Revisiting Adversarial Risk | Recent works on adversarial perturbations show that there is an inherent
trade-off between standard test accuracy and adversarial accuracy.
Specifically, they show that no classifier can simultaneously be robust to
adversarial perturbations and achieve high standard test accuracy. However,
this is contrary to the standard notion that on tasks such as image
classification, humans are robust classifiers with low error rate. In this
work, we show that the main reason behind this confusion is the inexact
definition of adversarial perturbation that is used in the literature. To fix
this issue, we propose a slight, yet important modification to the existing
definition of adversarial perturbation. Based on the modified definition, we
show that there is no trade-off between adversarial and standard accuracies;
there exist classifiers that are robust and achieve high standard accuracy. We
further study several properties of this new definition of adversarial risk and
its relation to the existing definition.
| stat |
Batch-Incremental Triplet Sampling for Training Triplet Networks Using
Bayesian Updating Theorem | Variants of Triplet networks are robust entities for learning a
discriminative embedding subspace. There exist different triplet mining
approaches for selecting the most suitable training triplets. Some of these
mining methods rely on the extreme distances between instances, and some others
make use of sampling. However, sampling from stochastic distributions of data
rather than sampling merely from the existing embedding instances can provide
more discriminative information. In this work, we sample triplets from
distributions of data rather than from existing instances. We consider a
multivariate normal distribution for the embedding of each class. Using
Bayesian updating and conjugate priors, we update the distributions of classes
dynamically by receiving the new mini-batches of training data. The proposed
triplet mining with Bayesian updating can be used with any triplet-based loss
function, e.g., triplet-loss or Neighborhood Component Analysis (NCA) loss.
Accordingly, Our triplet mining approaches are called Bayesian Updating Triplet
(BUT) and Bayesian Updating NCA (BUNCA), depending on which loss function is
being used. Experimental results on two public datasets, namely MNIST and
histopathology colorectal cancer (CRC), substantiate the effectiveness of the
proposed triplet mining method.
| stat |
High-dimensional modeling of spatial and spatio-temporal conditional
extremes using INLA and the SPDE approach | The conditional extremes framework allows for event-based stochastic modeling
of dependent extremes, and has recently been extended to spatial and
spatio-temporal settings. After standardizing the marginal distributions and
applying an appropriate linear normalization, certain non-stationary Gaussian
processes can be used as asymptotically-motivated models for the process
conditioned on threshold exceedances at a fixed reference location and time. In
this work, we adopt a Bayesian perspective by implementing estimation through
the integrated nested Laplace approximation (INLA), allowing for novel and
flexible semi-parametric specifications of the Gaussian mean function. By using
Gauss-Markov approximations of the Mat\'ern covariance function (known as the
Stochastic Partial Differential Equation approach) at a latent stage of the
model, likelihood-based inference becomes feasible even with several thousands
of observed locations. We explain how constraints on the spatial and
spatio-temporal Gaussian processes, arising from the conditioning mechanism,
can be implemented through the latent variable approach without losing the
computationally convenient Markov property. We discuss tools for the comparison
of posterior models, and illustrate the flexibility of the approach with
gridded Red Sea surface temperature data at over 6,000 observed locations.
Posterior sampling is exploited to study the probability distribution of
cluster functionals of spatial and spatio-temporal extreme episodes.
| stat |
Analysis of KNN Density Estimation | We analyze the $\ell_1$ and $\ell_\infty$ convergence rates of k nearest
neighbor density estimation method. Our analysis includes two different cases
depending on whether the support set is bounded or not. In the first case, the
probability density function has a bounded support and is bounded away from
zero. We show that kNN density estimation is minimax optimal under both
$\ell_1$ and $\ell_\infty$ criteria, if the support set is known. If the
support set is unknown, then the convergence rate of $\ell_1$ error is not
affected, while $\ell_\infty$ error does not converge. In the second case, the
probability density function can approach zero and is smooth everywhere.
Moreover, the Hessian is assumed to decay with the density values. For this
case, our result shows that the $\ell_\infty$ error of kNN density estimation
is nearly minimax optimal. The $\ell_1$ error does not reach the minimax lower
bound, but is better than kernel density estimation.
| stat |
Optimal Penalized Function-on-Function Regression under a Reproducing
Kernel Hilbert Space Framework | Many scientific studies collect data where the response and predictor
variables are both functions of time, location, or some other covariate.
Understanding the relationship between these functional variables is a common
goal in these studies. Motivated from two real-life examples, we present in
this paper a function-on-function regression model that can be used to analyze
such kind of functional data. Our estimator of the 2D coefficient function is
the optimizer of a form of penalized least squares where the penalty enforces a
certain level of smoothness on the estimator. Our first result is the
Representer Theorem which states that the exact optimizer of the penalized
least squares actually resides in a data-adaptive finite dimensional subspace
although the optimization problem is defined on a function space of infinite
dimensions. This theorem then allows us an easy incorporation of the Gaussian
quadrature into the optimization of the penalized least squares, which can be
carried out through standard numerical procedures. We also show that our
estimator achieves the minimax convergence rate in mean prediction under the
framework of function-on-function regression. Extensive simulation studies
demonstrate the numerical advantages of our method over the existing ones,
where a sparse functional data extension is also introduced. The proposed
method is then applied to our motivating examples of the benchmark Canadian
weather data and a histone regulation study.
| stat |
Hypothesis Testing in Nonlinear Function on Scalar Regression with
Application to Child Growth Study | We propose a kernel machine based hypothesis testing procedure in nonlinear
function-on-scalar regression model. Our research is motivated by the Newborn
Epigenetic Study (NEST) where the question of interest is whether a
pre-specified group of toxic metals or methylation at any of 9 differentially
methylated regions (DMRs) is associated with child growth. We take the child
growth trajectory as the functional response, and model the toxic metal
measurements jointly using a nonlinear function. We use a kernel machine
approach to model the unknown function and transform the hypothesis of no
effect to an appropriate variance component test. We demonstrate our proposed
methodology using a simulation study and by applying it to analyze the NEST
data.
| stat |
Efficient stochastic optimisation by unadjusted Langevin Monte Carlo.
Application to maximum marginal likelihood and empirical Bayesian estimation | Stochastic approximation methods play a central role in maximum likelihood
estimation problems involving intractable likelihood functions, such as
marginal likelihoods arising in problems with missing or incomplete data, and
in parametric empirical Bayesian estimation. Combined with Markov chain Monte
Carlo algorithms, these stochastic optimisation methods have been successfully
applied to a wide range of problems in science and industry. However, this
strategy scales poorly to large problems because of methodological and
theoretical difficulties related to using high-dimensional Markov chain Monte
Carlo algorithms within a stochastic approximation scheme. This paper proposes
to address these difficulties by using unadjusted Langevin algorithms to
construct the stochastic approximation. This leads to a highly efficient
stochastic optimisation methodology with favourable convergence properties that
can be quantified explicitly and easily checked. The proposed methodology is
demonstrated with three experiments, including a challenging application to
high-dimensional statistical audio analysis and a sparse Bayesian logistic
regression with random effects problem.
| stat |
Weighted Meta-Learning | Meta-learning leverages related source tasks to learn an initialization that
can be quickly fine-tuned to a target task with limited labeled examples.
However, many popular meta-learning algorithms, such as model-agnostic
meta-learning (MAML), only assume access to the target samples for fine-tuning.
In this work, we provide a general framework for meta-learning based on
weighting the loss of different source tasks, where the weights are allowed to
depend on the target samples. In this general setting, we provide upper bounds
on the distance of the weighted empirical risk of the source tasks and expected
target risk in terms of an integral probability metric (IPM) and Rademacher
complexity, which apply to a number of meta-learning settings including MAML
and a weighted MAML variant. We then develop a learning algorithm based on
minimizing the error bound with respect to an empirical IPM, including a
weighted MAML algorithm, $\alpha$-MAML. Finally, we demonstrate empirically on
several regression problems that our weighted meta-learning algorithm is able
to find better initializations than uniformly-weighted meta-learning
algorithms, such as MAML.
| stat |
Investigating linear relationships between non constant variances of
economic variables | In this paper we aim to assess linear relationships between the non constant
variances of economic variables. The proposed methodology is based on a
bootstrap cumulative sum (CUSUM) test. Simulations suggest a good behavior of
the test for sample sizes commonly encountered in practice. The tool we provide
is intended to highlight relations or draw common patterns between economic
variables through their non constant variances. The outputs of this paper is
illustrated considering U.S. regional data.
| stat |
Shared Subspace Models for Multi-Group Covariance Estimation | We develop a model-based method for evaluating heterogeneity among several p
x p covariance matrices in the large p, small n setting. This is done by
assuming a spiked covariance model for each group and sharing information about
the space spanned by the group-level eigenvectors. We use an empirical Bayes
method to identify a low-dimensional subspace which explains variation across
all groups and use an MCMC algorithm to estimate the posterior uncertainty of
eigenvectors and eigenvalues on this subspace. The implementation and utility
of our model is illustrated with analyses of high-dimensional multivariate gene
expression.
| stat |
MAXENT3D_PID: An Estimator for the Maximum-entropy Trivariate Partial
Information Decomposition | Chicharro (2017) introduced a procedure to determine multivariate partial
information measures within the maximum entropy framework, separating unique,
redundant, and synergistic components of information. Makkeh, Theis, and
Vicente (2018) formulated the latter trivariate partial information measure as
Cone Programming. In this paper, we present MAXENT3D_PID, a production-quality
software that computes the trivariate partial information measure based on the
Cone Programming model. We describe in detail our software, explain how to use
it, and perform some experiments reflecting its accuracy in estimating the
trivariate partial information decomposition.
| stat |
Feature Robust Optimal Transport for High-dimensional Data | Optimal transport is a machine learning problem with applications including
distribution comparison, feature selection, and generative adversarial
networks. In this paper, we propose feature-robust optimal transport (FROT) for
high-dimensional data, which solves high-dimensional OT problems using feature
selection to avoid the curse of dimensionality. Specifically, we find a
transport plan with discriminative features. To this end, we formulate the FROT
problem as a min--max optimization problem. We then propose a convex
formulation of the FROT problem and solve it using a Frank--Wolfe-based
optimization algorithm, whereby the subproblem can be efficiently solved using
the Sinkhorn algorithm. Since FROT finds the transport plan from selected
features, it is robust to noise features. To show the effectiveness of FROT, we
propose using the FROT algorithm for the layer selection problem in deep neural
networks for semantic correspondence. By conducting synthetic and benchmark
experiments, we demonstrate that the proposed method can find a strong
correspondence by determining important layers. We show that the FROT algorithm
achieves state-of-the-art performance in real-world semantic correspondence
datasets.
| stat |
Identification of Causal Effects Within Principal Strata Using Auxiliary
Variables | In causal inference, principal stratification is a framework for dealing with
a posttreatment intermediate variable between a treatment and an outcome, in
which the principal strata are defined by the joint potential values of the
intermediate variable. Because the principal strata are not fully observable,
the causal effects within them, also known as the principal causal effects, are
not identifiable without additional assumptions. Several previous empirical
studies leveraged auxiliary variables to improve the inference of principal
causal effects. We establish a general theory for identification and estimation
of the principal causal effects with auxiliary variables, which provides a
solid foundation for statistical inference and more insights for model building
in empirical research. In particular, we consider two commonly-used strategies
for principal stratification problems: principal ignorability, and the
conditional independence between the auxiliary variable and the outcome given
principal strata and covariates. For these two strategies, we give
non-parametric and semi-parametric identification results without modeling
assumptions on the outcome. When the assumptions for neither strategies are
plausible, we propose a large class of flexible parametric and semi-parametric
models for identifying principal causal effects. Our theory not only
establishes formal identification results of several models that have been used
in previous empirical studies but also generalizes them to allow for different
types of outcomes and intermediate variables.
| stat |
Variational Bayesian Monte Carlo with Noisy Likelihoods | Variational Bayesian Monte Carlo (VBMC) is a recently introduced framework
that uses Gaussian process surrogates to perform approximate Bayesian inference
in models with black-box, non-cheap likelihoods. In this work, we extend VBMC
to deal with noisy log-likelihood evaluations, such as those arising from
simulation-based models. We introduce new `global' acquisition functions, such
as expected information gain (EIG) and variational interquantile range (VIQR),
which are robust to noise and can be efficiently evaluated within the VBMC
setting. In a novel, challenging, noisy-inference benchmark comprising of a
variety of models with real datasets from computational and cognitive
neuroscience, VBMC+VIQR achieves state-of-the-art performance in recovering the
ground-truth posteriors and model evidence. In particular, our method vastly
outperforms `local' acquisition functions and other surrogate-based inference
methods while keeping a small algorithmic cost. Our benchmark corroborates VBMC
as a general-purpose technique for sample-efficient black-box Bayesian
inference also with noisy models.
| stat |
Ensemble Distribution Distillation | Ensembles of models often yield improvements in system performance. These
ensemble approaches have also been empirically shown to yield robust measures
of uncertainty, and are capable of distinguishing between different
\emph{forms} of uncertainty. However, ensembles come at a computational and
memory cost which may be prohibitive for many applications. There has been
significant work done on the distillation of an ensemble into a single model.
Such approaches decrease computational cost and allow a single model to achieve
an accuracy comparable to that of an ensemble. However, information about the
\emph{diversity} of the ensemble, which can yield estimates of different forms
of uncertainty, is lost. This work considers the novel task of \emph{Ensemble
Distribution Distillation} (EnD$^2$) --- distilling the distribution of the
predictions from an ensemble, rather than just the average prediction, into a
single model. EnD$^2$ enables a single model to retain both the improved
classification performance of ensemble distillation as well as information
about the diversity of the ensemble, which is useful for uncertainty
estimation. A solution for EnD$^2$ based on Prior Networks, a class of models
which allow a single neural network to explicitly model a distribution over
output distributions, is proposed in this work. The properties of EnD$^2$ are
investigated on both an artificial dataset, and on the CIFAR-10, CIFAR-100 and
TinyImageNet datasets, where it is shown that EnD$^2$ can approach the
classification performance of an ensemble, and outperforms both standard DNNs
and Ensemble Distillation on the tasks of misclassification and
out-of-distribution input detection.
| stat |
Variational Auto-Encoder for Recommender Systems with
Exploration-Exploitation | Recent years have witnessed rapid developments on collaborative filtering
techniques for improving the performance of recommender systems due to the
growing need of companies to help users discover new and relevant items.
However, the majority of existing literature focuses on delivering items which
match the user model learned from users' past preferences. A good
recommendation model is expected to recommend items that are known to enjoy and
items that are novel to try. In this work, we introduce an
exploitation-exploration motivated variational auto-encoder (XploVAE) to
collaborative filtering. To facilitate personalized recommendations, we
construct user-specific subgraphs, which contain first-order proximity
capturing observed user-item interactions for exploitation and higher-order
proximity for exploration. A hierarchical latent space model is utilized to
learn the personalized item embedding for a given user, along with the
population distribution of all user subgraphs. Finally, experimental results on
various real-world datasets clearly demonstrate the effectiveness of our
proposed model on leveraging the exploitation and exploration recommendation
tasks.
| stat |
Nonparametric false discovery rate control for identifying simultaneous
signals | It is frequently of interest to jointly analyze multiple sequences of
multiple tests in order to identify simultaneous signals, defined as features
tested in multiple studies whose test statistics are non-null in each. In many
problems, however, the null distributions of the test statistics may be
complicated or even unknown, and there do not currently exist any procedures
that can be employed in these cases. This paper proposes a new nonparametric
procedure that can identify simultaneous signals across multiple studies even
without knowing the null distributions of the test statistics. The method is
shown to asymptotically control the false discovery rate, and in simulations
had excellent power and error control. In an analysis of gene expression and
histone acetylation patterns in the brains of mice exposed to a conspecific
intruder, it identified genes that were both differentially expressed and next
to differentially accessible chromatin. The proposed method is available in the
R package github.com/sdzhao/ssa.
| stat |
Unsupervised Methods for Identifying Pass Coverage Among Defensive Backs
with NFL Player Tracking Data | Analysis of player tracking data for American football is in its infancy,
since the National Football League (NFL) released its Next Gen Stats tracking
data publicly for the first time in December 2018. While tracking datasets in
other sports often contain detailed annotations of on-field events, annotations
in the NFL's tracking data are limited. Methods for creating these annotations
typically require extensive human labeling, which is difficult and expensive.
We begin tackling this class of problems by creating annotations for pass
coverage types by defensive backs using unsupervised learning techniques, which
require no manual labeling or human oversight. We define a set of features from
the NFL's tracking data that help distinguish between "zone" and "man"
coverage. We use Gaussian mixture modeling and hierarchical clustering to
create clusters corresponding to each group, and we assign the appropriate type
of coverage to each cluster through qualitative analysis of the plays in each
cluster. We find that the mixture model's "soft" cluster assignments allow for
more flexibility when identifying coverage types. Our work makes possible
several potential avenues of future NFL research, and we provide a basic
exploration of these in this paper.
| stat |
Human experts vs. machines in taxa recognition | The step of expert taxa recognition currently slows down the response time of
many bioassessments. Shifting to quicker and cheaper state-of-the-art machine
learning approaches is still met with expert scepticism towards the ability and
logic of machines. In our study, we investigate both the differences in
accuracy and in the identification logic of taxonomic experts and machines. We
propose a systematic approach utilizing deep Convolutional Neural Nets with the
transfer learning paradigm and extensively evaluate it over a multi-pose
taxonomic dataset with hierarchical labels specifically created for this
comparison. We also study the prediction accuracy on different ranks of
taxonomic hierarchy in detail. Our results revealed that human experts using
actual specimens yield the lowest classification error ($\overline{CE}=6.1\%$).
However, a much faster, automated approach using deep Convolutional Neural Nets
comes close to human accuracy ($\overline{CE}=11.4\%$). Contrary to previous
findings in the literature, we find that for machines following a typical flat
classification approach commonly used in machine learning performs better than
forcing machines to adopt a hierarchical, local per parent node approach used
by human taxonomic experts. Finally, we publicly share our unique dataset to
serve as a public benchmark dataset in this field.
| stat |
A Degradation Performance Model With Mixed-type Covariates and Latent
Heterogeneity | Successful modeling of degradation performance data is essential for accurate
reliability assessment and failure predictions of highly reliable product
units. The degradation performance measurements over time are highly
heterogeneous. Such heterogeneity can be partially attributed to external
factors, such as accelerated/environmental conditions, and can also be
attributed to internal factors, such as material microstructure characteristics
of product units. The latent heterogeneity due to the unobserved/unknown
factors shared within each product unit may also exists and need to be
considered as well. Existing degradation models often fail to consider (i) the
influence of both external accelerated/environmental conditions and internal
material information, (ii) the influence of unobserved/unknown factors within
each unit. In this work, we propose a generic degradation performance modeling
framework with mixed-type covariates and latent heterogeneity to account for
both influences of observed internal and external factors as well as unobserved
factors. Effective estimation algorithm is also developed to jointly quantify
the influences of mixed-type covariates and individual latent heterogeneity,
and also to examine the potential interaction between mixed-type covariates.
Functional data analysis and data augmentation techniques are employed to
address a series of estimation issues. A real case study is further provided to
demonstrate the superior performance of the proposed approach over several
alternative modeling approaches. Besides, the proposed degradation performance
modeling framework also provides interpretable findings.
| stat |
Model-Based Domain Generalization | We consider the problem of domain generalization, in which a predictor is
trained on data drawn from a family of related training domains and tested on a
distinct and unseen test domain. While a variety of approaches have been
proposed for this setting, it was recently shown that no existing algorithm can
consistently outperform empirical risk minimization (ERM) over the training
domains. To this end, in this paper we propose a novel approach for the domain
generalization problem called Model-Based Domain Generalization. In our
approach, we first use unlabeled data from the training domains to learn
multi-modal domain transformation models that map data from one training domain
to any other domain. Next, we propose a constrained optimization-based
formulation for domain generalization which enforces that a trained predictor
be invariant to distributional shifts under the underlying domain
transformation model. Finally, we propose a novel algorithmic framework for
efficiently solving this constrained optimization problem. In our experiments,
we show that this approach outperforms both ERM and domain generalization
algorithms on numerous well-known, challenging datasets, including WILDS, PACS,
and ImageNet. In particular, our algorithms beat the current state-of-the-art
methods on the very-recently-proposed WILDS benchmark by up to 20 percentage
points.
| stat |
Reconstruction of Past Human land-use from Pollen Data and Anthropogenic
land-cover Changes Scenarios | Accurate maps of past land cover and human land-use are necessary when
studying the impact of anthropogenic land-cover changes on climate. Ideally the
maps of past land cover would be separated into naturally occurring vegetation
and human induced changes, allowing us to quantify the effect of human land-use
on past climate. Here we investigate the possibility of combining regional,
fossil pollen based, land-cover reconstructions with, population based,
estimates of past human land-use. By merging these two datasets and
interpolating the pollen based land-cover reconstructions we aim at obtaining
maps that provide both past natural land-cover and the anthropogenic land-cover
changes. We develop a Bayesian hierarchical model to handle the complex data,
using a latent Gaussian Markov random fields (GMRF) for the interpolation.
Estimation of the model is based on a block updated Markov chain Monte Carlo
(MCMC) algorithm. The sparse precision matrix of the GMRF together with an
adaptive Metropolis adjusted Langevin step allows for fast inference.
Uncertainties in the land-use predictions are computed from the MCMC posterior
samples. The model uses the pollen based observations to reconstruct three
composition of land cover; Coniferous forest, Broadleaved forest and
Unforested/Open land. The unforested land is then further decomposed into
natural and human induced openness by inclusion of the estimates of past human
land-use. The model is applied to five time periods - centred around 1900 CE,
1725 CE, 1425 CE, 1000 and, 4000 BCE over Europe. The results suggest pollen
based observations can be used to recover past human land-use by adjusting the
population based anthropogenic land-cover changes estimates.
| stat |
On Variable Screening in Multiple Nonparametric Regression Model | In this article, we study the problem of variable screening in multiple
nonparametric regression model. The proposed methodology is based on the fact
that the partial derivative of the regression function with respect to the
irrelevant variable should be negligible. The Statistical property of the
proposed methodology is investigated under both cases : (i) when the variance
of the error term is known, and (ii) when the variance of the error term is
unknown. Moreover, we establish the practicality of our proposed methodology
for various simulated and real data related to interdisciplinary sciences such
as Economics, Finance and other sciences.
| stat |
Bayesian Quadrature Optimization for Probability Threshold Robustness
Measure | In many product development problems, the performance of the product is
governed by two types of parameters called design parameter and environmental
parameter. While the former is fully controllable, the latter varies depending
on the environment in which the product is used. The challenge of such a
problem is to find the design parameter that maximizes the probability that the
performance of the product will meet the desired requisite level given the
variation of the environmental parameter. In this paper, we formulate this
practical problem as active learning (AL) problems and propose efficient
algorithms with theoretically guaranteed performance. Our basic idea is to use
Gaussian Process (GP) model as the surrogate model of the product development
process, and then to formulate our AL problems as Bayesian Quadrature
Optimization problems for probabilistic threshold robustness (PTR) measure. We
derive credible intervals for the PTR measure and propose AL algorithms for the
optimization and level set estimation of the PTR measure. We clarify the
theoretical properties of the proposed algorithms and demonstrate their
efficiency in both synthetic and real-world product development problems.
| stat |
Efficient, Doubly Robust Estimation of the Effect of Dose Switching for
Switchers in a Randomised Clinical Trial | Motivated by a clinical trial conducted by Janssen Pharmaceuticals in which a
flexible dosing regimen is compared to placebo, we evaluate how switchers in
the treatment arm (i.e., patients who were switched to the higher dose) would
have fared had they been kept on the low dose. This in order to understand
whether flexible dosing is potentially beneficial for them. Simply comparing
these patients' responses with those of patients who stayed on the low dose is
unsatisfactory because the latter patients are usually in a better health
condition. Because the available information in the considered trial is too
scarce to enable a reliable adjustment, we will instead transport data from a
fixed dosing trial that has been conducted concurrently on the same target,
albeit not in an identical patient population. In particular, we will propose
an estimator which relies on an outcome model and a propensity score model for
the association between study and patient characteristics. The proposed
estimator is asymptotically unbiased if at least one of both models is
correctly specified, and efficient (under the model defined by the restrictions
on the propensity score) when both models are correctly specified. We show that
the proposed method for using results from an external study is generically
applicable in studies where a classical confounding adjustment is not possible
due to positivity violation (e.g., studies where switching takes place in a
deterministic manner). Monte Carlo simulations and application to the
motivating study demonstrate adequate performance.
| stat |
Robust correlation for aggregated data with spatial characteristics | The objective of this paper is to study the robustness of computation of
correlations under spatial constraints. The motivation of our paper is the
specific case of functional magnetic resonance (fMRI) brain imaging data, where
voxels are aggregated to compute correlations. In this paper we show that the
way the spatial components are aggregating to compute correlation may have a
strong influence on the resulting estimations. We then propose various
estimators which take into account this spatial structure.
| stat |
Sensitivity Analysis of Deep Neural Networks | Deep neural networks (DNNs) have achieved superior performance in various
prediction tasks, but can be very vulnerable to adversarial examples or
perturbations. Therefore, it is crucial to measure the sensitivity of DNNs to
various forms of perturbations in real applications. We introduce a novel
perturbation manifold and its associated influence measure to quantify the
effects of various perturbations on DNN classifiers. Such perturbations include
various external and internal perturbations to input samples and network
parameters. The proposed measure is motivated by information geometry and
provides desirable invariance properties. We demonstrate that our influence
measure is useful for four model building tasks: detecting potential
'outliers', analyzing the sensitivity of model architectures, comparing network
sensitivity between training and test sets, and locating vulnerable areas.
Experiments show reasonably good performance of the proposed measure for the
popular DNN models ResNet50 and DenseNet121 on CIFAR10 and MNIST datasets.
| stat |
Combining multiple imputation with raking of weights: An efficient and
robust approach in the setting of nearly-true models | Multiple imputation provides us with efficient estimators in model-based
methods for handling missing data under the true model. It is also
well-understood that design-based estimators are robust methods that do not
require accurately modeling the missing data; however, they can be inefficient.
In any applied setting, it is difficult to know whether a missing data model
may be good enough to win the bias-efficiency trade-off. Raking of weights is
one approach that relies on constructing an auxiliary variable from data
observed on the full cohort, which is then used to adjust the weights for the
usual Horvitz-Thompson estimator. Computing the optimally efficient raking
estimator requires evaluating the expectation of the efficient score given the
full cohort data, which is generally infeasible. We demonstrate multiple
imputation (MI) as a practical method to compute a raking estimator that will
be optimal. We compare this estimator to common parametric and semi-parametric
estimators, including standard multiple imputation. We show that while
estimators, such as the semi-parametric maximum likelihood and MI estimator,
obtain optimal performance under the true model, the proposed raking estimator
utilizing MI maintains a better robustness-efficiency trade-off even under mild
model misspecification. We also show that the standard raking estimator,
without MI, is often competitive with the optimal raking estimator. We
demonstrate these properties through several numerical examples and provide a
theoretical discussion of conditions for asymptotically superior relative
efficiency of the proposed raking estimator.
| stat |
Using Small Domain Estimation to obtain better retrospective Age Period
Cohort insights | Recent changes in housing costs relative to income are likely to affect
people's propensity to Housing Affordability Stress (HAS), which is known to
have a detrimental effect on a range of health outcomes. The magnitude of these
effects may vary between subgroups of the population, in particular across age
groups. Estimating these effect sizes from longitudinal data requires Small
Domain Estimation (SDE) as available data is generally limited to small sample
sizes. In this paper we develop the rationale for smoothing-based SDE using two
case studies: (1) transitions into and out of HAS and (2) the mental health
effect associated with HAS. We apply cross-validation to assess the relative
performance of multiple SDE methods and discuss how SDE can be embedded into
g-computation for causal inference.
| stat |
Non-Linear Non-Stationary Heteroscedasticity Volatility for Tracking of
Jump Processes | In this paper, we introduce a new jump process modeling which involves a
particular kind of non-Gaussian stochastic processes with random jumps at
random time points. The main goal of this study is to provide an accurate
tracking technique based on non-linear non-stationary heteroscedasticity (NNH)
time series. It is, in fact, difficult to track jump processes regarding the
fact that non-Gaussianity is an inherent feature in these processes. The
proposed NNH model is conditionally Gaussian whose conditional variance is
time-varying. Therefore, we use Kalman filter for state tracking. We show
analytically that the proposed NNH model is superior to the traditional
methods. Furthermore, to validate the findings, simulations are performed.
Finally, the comparison between the proposed method and other alternatives
techniques has been made.
| stat |
Knockoffs with Side Information | We consider the problem of assessing the importance of multiple variables or
factors from a dataset when side information is available. In principle, using
side information can allow the statistician to pay attention to variables with
a greater potential, which in turn, may lead to more discoveries. We introduce
an adaptive knockoff filter, which generalizes the knockoff procedure (Barber
and Cand\`es, 2015; Cand\`es et al., 2018) in that it uses both the data at
hand and side information to adaptively order the variables under study and
focus on those that are most promising. Adaptive knockoffs controls the
finite-sample false discovery rate (FDR) and we demonstrate its power by
comparing it with other structured multiple testing methods. We also apply our
methodology to real genetic data in order to find associations between genetic
variants and various phenotypes such as Crohn's disease and lipid levels. Here,
adaptive knockoffs makes more discoveries than reported in previous studies on
the same datasets.
| stat |
Blang: Bayesian declarative modelling of arbitrary data structures | Consider a Bayesian inference problem where a variable of interest does not
take values in a Euclidean space. These "non-standard" data structures are in
reality fairly common. They are frequently used in problems involving latent
discrete factor models, networks, and domain specific problems such as sequence
alignments and reconstructions, pedigrees, and phylogenies. In principle,
Bayesian inference should be particularly well-suited in such scenarios, as the
Bayesian paradigm provides a principled way to obtain confidence assessment for
random variables of any type. However, much of the recent work on making
Bayesian analysis more accessible and computationally efficient has focused on
inference in Euclidean spaces. In this paper, we introduce Blang, a domain
specific language (DSL) and library aimed at bridging this gap. Blang allows
users to perform Bayesian analysis on arbitrary data types while using a
declarative syntax similar to BUGS. Blang is augmented with intuitive language
additions to invent data types of the user's choosing. To perform inference at
scale on such arbitrary state spaces, Blang leverages recent advances in
parallelizable, non-reversible Markov chain Monte Carlo methods.
| stat |
Stochastic Gradient Descent in Hilbert Scales: Smoothness,
Preconditioning and Earlier Stopping | Stochastic Gradient Descent (SGD) has become the method of choice for solving
a broad range of machine learning problems. However, some of its learning
properties are still not fully understood. We consider least squares learning
in reproducing kernel Hilbert spaces (RKHSs) and extend the classical SGD
analysis to a learning setting in Hilbert scales, including Sobolev spaces and
Diffusion spaces on compact Riemannian manifolds. We show that even for
well-specified models, violation of a traditional benchmark smoothness
assumption has a tremendous effect on the learning rate. In addition, we show
that for miss-specified models, preconditioning in an appropriate Hilbert scale
helps to reduce the number of iterations, i.e. allowing for "earlier stopping".
| stat |
High-dimensional clustering via Random Projections | In this work, we address the unsupervised classification issue by exploiting
the general idea of Random Projection Ensemble. Specifically, we propose to
generate a set of low dimensional independent random projections and to perform
model-based clustering on each of them. The top $B^*$ projections, i.e. the
projections which show the best grouping structure are then retained. The final
partition is obtained by aggregating the clusters found in the projections via
consensus. The performances of the method are assessed on both real and
simulated datasets. The obtained results suggest that the proposal represents a
promising tool for high-dimensional clustering.
| stat |
The Multivariate Hawkes Process in High Dimensions: Beyond Mutual
Excitation | The Hawkes process is a class of point processes whose future depends on
their own history. Previous theoretical work on the Hawkes process is limited
to a special case in which a past event can only increase the occurrence of
future events, and the link function is linear. However, in neuronal networks
and other real-world applications, inhibitory relationships may be present, and
the link function may be non-linear. In this paper, we develop a new approach
for investigating the properties of the Hawkes process without the restriction
to mutual excitation or linear link functions. To this end, we employ a
thinning process representation and a coupling construction to bound the
dependence coefficient of the Hawkes process. Using recent developments on
weakly dependent sequences, we establish a concentration inequality for
second-order statistics of the Hawkes process. We apply this concentration
inequality to cross-covariance analysis in the high-dimensional regime, and we
verify the theoretical claims with simulation studies.
| stat |
Use of Historical Individual Patient Data in Analysis of Clinical Trials | Historical data from previous clinical trials, observational studies and
health records may be utilized in analysis of clinical trials data to
strengthen inference. Under the Bayesian framework incorporation of information
obtained from any source other than the current data is facilitated through
construction of an informative prior. The existing methodology for defining an
informative prior based on historical data relies on measuring similarity to
the current data at the study level and does not take advantage of individual
patient data (IPD). This paper proposes a family of priors that utilize IPD to
strengthen statistical inference. It is demonstrated that the proposed prior
construction approach outperforms the existing methods where the historical
data are partially exchangeable with the present data. The proposed method is
applied to IPD from a set of trials in non-small cell lung cancer.
| stat |
Identifying the latent space geometry of network models through analysis
of curvature | Statistically modeling networks, across numerous disciplines and contexts, is
fundamentally challenging because of (often high-order) dependence between
connections. A common approach assigns each person in the graph to a position
on a low-dimensional manifold. Distance between individuals in this (latent)
space is inversely proportional to the likelihood of forming a connection. The
choice of the latent geometry (the manifold class, dimension, and curvature)
has consequential impacts on the substantive conclusions of the model. More
positive curvature in the manifold, for example, encourages more and tighter
communities; negative curvature induces repulsion among nodes. Currently,
however, the choice of the latent geometry is an a priori modeling assumption
and there is limited guidance about how to make these choices in a data-driven
way. In this work, we present a method to consistently estimate the manifold
type, dimension, and curvature from an empirically relevant class of latent
spaces: simply connected, complete Riemannian manifolds of constant curvature.
Our core insight comes by representing the graph as a noisy distance matrix
based on the ties between cliques. Leveraging results from statistical
geometry, we develop hypothesis tests to determine whether the observed
distances could plausibly be embedded isometrically in each of the candidate
geometries. We explore the accuracy of our approach with simulations and then
apply our approach to data-sets from economics and sociology as well as
neuroscience.
| stat |
Learning with Gradient Descent and Weakly Convex Losses | We study the learning performance of gradient descent when the empirical risk
is weakly convex, namely, the smallest negative eigenvalue of the empirical
risk's Hessian is bounded in magnitude. By showing that this eigenvalue can
control the stability of gradient descent, generalisation error bounds are
proven that hold under a wider range of step sizes compared to previous work.
Out of sample guarantees are then achieved by decomposing the test error into
generalisation, optimisation and approximation errors, each of which can be
bounded and traded off with respect to algorithmic parameters, sample size and
magnitude of this eigenvalue. In the case of a two layer neural network, we
demonstrate that the empirical risk can satisfy a notion of local weak
convexity, specifically, the Hessian's smallest eigenvalue during training can
be controlled by the normalisation of the layers, i.e., network scaling. This
allows test error guarantees to then be achieved when the population risk
minimiser satisfies a complexity assumption. By trading off the network
complexity and scaling, insights are gained into the implicit bias of neural
network scaling, which are further supported by experimental findings.
| stat |
Sequential model aggregation for production forecasting | Production forecasting is a key step to design the future development of a
reservoir. A classical way to generate such forecasts consists in simulating
future production for numerical models representative of the reservoir.
However, identifying such models can be very challenging as they need to be
constrained to all available data. In particular, they should reproduce past
production data, which requires to solve a complex non-linear inverse problem.
In this paper, we thus propose to investigate the potential of machine learning
algorithms to predict the future production of a reservoir based on past
production data without model calibration. We focus more specifically on robust
online aggregation, a deterministic approach that provides a robust framework
to make forecasts on a regular basis. This method does not rely on any specific
assumption or need for stochastic modeling. Forecasts are first simulated for a
set of base reservoir models representing the prior uncertainty, and then
combined to predict production at the next time step. The weight associated to
each forecast is related to its past performance. Three different algorithms
are considered for weight computations: the exponentially weighted average
algorithm, ridge regression and the Lasso regression. They are applied on a
synthetic reservoir case study, the Brugge case, for sequential predictions. To
estimate the potential of development scenarios, production forecasts are
needed on long periods of time without intermediary data acquisition. An
extension of the deterministic aggregation approach is thus proposed in this
paper to provide such multi-step-ahead forecasts.
| stat |
Estimation of COVID-19 spread curves integrating global data and
borrowing information | Currently, novel coronavirus disease 2019 (COVID-19) is a big threat to
global health. The rapid spread of the virus has created pandemic, and
countries all over the world are struggling with a surge in COVID-19 infected
cases. There are no drugs or other therapeutics approved by the US Food and
Drug Administration to prevent or treat COVID-19: information on the disease is
very limited and scattered even if it exists. This motivates the use of data
integration, combining data from diverse sources and eliciting useful
information with a unified view of them. In this paper, we propose a Bayesian
hierarchical model that integrates global data for real-time prediction of
infection trajectory for multiple countries. Because the proposed model takes
advantage of borrowing information across multiple countries, it outperforms an
existing individual country-based model. As fully Bayesian way has been
adopted, the model provides a powerful predictive tool endowed with uncertainty
quantification. Additionally, a joint variable selection technique has been
integrated into the proposed modeling scheme, which aimed to identify possible
country-level risk factors for severe disease due to COVID-19.
| stat |
Model-based fuzzy time series clustering of conditional higher moments | This paper develops a new time series clustering procedure allowing for
heteroskedasticity, non-normality and model's non-linearity. At this aim, we
follow a fuzzy approach. Specifically, considering a Dynamic Conditional Score
(DCS) model, we propose to cluster time series according to their estimated
conditional moments via the Autocorrelation-based fuzzy C-means (A-FCM)
algorithm. The DCS parametric modelling is appealing because of its generality
and computational feasibility. The usefulness of the proposed procedure is
illustrated using an experiment with simulated data and several empirical
applications with financial time series assuming both linear and nonlinear
models' specification and under several assumptions about time series density
function.
| stat |
Algorithmic Bayesian Group Gibbs Selection | Bayesian model selection, with precedents in George and McCulloch (1993) and
Abramovich et al. (1998), support credibility measures that relate model
uncertainty, but computation can be costly when sparse priors are approximate.
We design an exact selection engine suitable for Gauss noise, t-distributed
noise, and logistic learning, benefiting from data-structures derived from
coordinate descent lasso. Gibbs sampler chains are stored in a compressed
binary format compatible with Equi-Energy (Kou et al., 2006) tempering. We
achieve a grouped-effects selection model, similar to the setting for group
lasso, to determine co-entry of coefficients into the model. We derive a
functional integrand for group inclusion, and introduce a new MCMC switching
step to avoid numerical integration. Theorems show this step has exponential
convergence to target distribution. We demonstrate a role for group selection
to inform on genetic decomposition in a diallel experiment, and identify
potential quantitative trait loci in p > 40K Heterogenous Stock
haplotype/phenotype studies.
| stat |
Teacher-Student Domain Adaptation for Biosensor Models | We present an approach to domain adaptation, addressing the case where data
from the source domain is abundant, labelled data from the target domain is
limited or non-existent, and a small amount of paired source-target data is
available. The method is designed for developing deep learning models that
detect the presence of medical conditions based on data from consumer-grade
portable biosensors. It addresses some of the key problems in this area,
namely, the difficulty of acquiring large quantities of clinically labelled
data from the biosensor, and the noise and ambiguity that can affect the
clinical labels. The idea is to pre-train an expressive model on a large
dataset of labelled recordings from a sensor modality for which data is
abundant, and then to adapt the model's lower layers so that its predictions on
the target modality are similar to the original model's on paired examples from
the source modality. We show that the pre-trained model's predictions provide a
substantially better learning signal than the clinician-provided labels, and
that this teacher-student technique significantly outperforms both a naive
application of supervised deep learning and a label-supervised version of
domain adaptation on a synthetic dataset and in a real-world case study on
sleep apnea. By reducing the volume of data required and obviating the need for
labels, our approach should reduce the cost associated with developing
high-performance deep learning models for biosensors.
| stat |
Adversarially Robust Estimate and Risk Analysis in Linear Regression | Adversarially robust learning aims to design algorithms that are robust to
small adversarial perturbations on input variables. Beyond the existing studies
on the predictive performance to adversarial samples, our goal is to understand
statistical properties of adversarially robust estimates and analyze
adversarial risk in the setup of linear regression models. By discovering the
statistical minimax rate of convergence of adversarially robust estimators, we
emphasize the importance of incorporating model information, e.g., sparsity, in
adversarially robust learning. Further, we reveal an explicit connection of
adversarial and standard estimates, and propose a straightforward two-stage
adversarial learning framework, which facilitates to utilize model structure
information to improve adversarial robustness. In theory, the consistency of
the adversarially robust estimator is proven and its Bahadur representation is
also developed for the statistical inference purpose. The proposed estimator
converges in a sharp rate under either low-dimensional or sparse scenario.
Moreover, our theory confirms two phenomena in adversarially robust learning:
adversarial robustness hurts generalization, and unlabeled data help improve
the generalization. In the end, we conduct numerical simulations to verify our
theory.
| stat |
Subgroup identification in individual patient data meta-analysis using
model-based recursive partitioning | Model-based recursive partitioning (MOB) can be used to identify subgroups
with differing treatment effects. The detection rate of treatment-by-covariate
interactions and the accuracy of identified subgroups using MOB depend strongly
on the sample size. Using data from multiple randomized controlled clinical
trials can overcome the problem of too small samples. However, naively pooling
data from multiple trials may result in the identification of spurious
subgroups as differences in study design, subject selection and other sources
of between-trial heterogeneity are ignored. In order to account for
between-trial heterogeneity in individual participant data (IPD) meta-analysis
random-effect models are frequently used. Commonly, heterogeneity in the
treatment effect is modelled using random effects whereas heterogeneity in the
baseline risks is modelled by either fixed effects or random effects. In this
article, we propose metaMOB, a procedure using the generalized mixed-effects
model tree (GLMM tree) algorithm for subgroup identification in IPD
meta-analysis. Although the application of metaMOB is potentially wider, e.g.
randomized experiments with participants in social sciences or preclinical
experiments in life sciences, we focus on randomized controlled clinical
trials. In a simulation study, metaMOB outperformed GLMM trees assuming a
random intercept only and model-based recursive partitioning (MOB), whose
algorithm is the basis for GLMM trees, with respect to the false discovery
rates, accuracy of identified subgroups and accuracy of estimated treatment
effect. The most robust and therefore most promising method is metaMOB with
fixed effects for modelling the between-trial heterogeneity in the baseline
risks.
| stat |
Frank-Wolfe Methods in Probability Space | We introduce a new class of Frank-Wolfe algorithms for minimizing
differentiable functionals over probability measures. This framework can be
shown to encompass a diverse range of tasks in areas such as artificial
intelligence, reinforcement learning, and optimization. Concrete computational
complexities for these algorithms are established and demonstrate that these
methods enjoy convergence in regimes that go beyond convexity and require
minimal regularity of the underlying functional. Novel techniques used to
obtain these results also lead to the development of new complexity bounds and
duality theorems for a family of distributionally robust optimization problems.
The performance of our method is demonstrated on several nonparametric
estimation problems.
| stat |
Complementary Lipschitz continuity results for the distribution of
intersections or unions of independent random sets in finite discrete spaces | We prove that intersections and unions of independent random sets in finite
spaces achieve a form of Lipschitz continuity. More precisely, given the
distribution of a random set $\Xi$, the function mapping any random set
distribution to the distribution of its intersection (under independence
assumption) with $\Xi$ is Lipschitz continuous with unit Lipschitz constant if
the space of random set distributions is endowed with a metric defined as the
$L_k$ norm distance between inclusion functionals also known as commonalities.
Moreover, the function mapping any random set distribution to the distribution
of its union (under independence assumption) with $\Xi$ is Lipschitz continuous
with unit Lipschitz constant if the space of random set distributions is
endowed with a metric defined as the $L_k$ norm distance between hitting
functionals also known as plausibilities.
Using the epistemic random set interpretation of belief functions, we also
discuss the ability of these distances to yield conflict measures. All the
proofs in this paper are derived in the framework of Dempster-Shafer belief
functions. Let alone the discussion on conflict measures, it is straightforward
to transcribe the proofs into the general (non necessarily epistemic) random
set terminology.
| stat |
A Matrix--free Likelihood Method for Exploratory Factor Analysis of
High-dimensional Gaussian Data | This paper proposes a novel profile likelihood method for estimating the
covariance parameters in exploratory factor analysis of high-dimensional
Gaussian datasets with fewer observations than number of variables. An
implicitly restarted Lanczos algorithm and a limited-memory quasi-Newton method
are implemented to develop a matrix-free framework for likelihood maximization.
Simulation results show that our method is substantially faster than the
expectation-maximization solution without sacrificing accuracy. Our method is
applied to fit factor models on data from suicide attempters, suicide ideators
and a control group.
| stat |
Bayesian Structure Learning in Sparse Gaussian Graphical Models | Decoding complex relationships among large numbers of variables with
relatively few observations is one of the crucial issues in science. One
approach to this problem is Gaussian graphical modeling, which describes
conditional independence of variables through the presence or absence of edges
in the underlying graph. In this paper, we introduce a novel and efficient
Bayesian framework for Gaussian graphical model determination which is a
trans-dimensional Markov Chain Monte Carlo (MCMC) approach based on a
continuous-time birth-death process. We cover the theory and computational
details of the method. It is easy to implement and computationally feasible for
high-dimensional graphs. We show our method outperforms alternative Bayesian
approaches in terms of convergence, mixing in the graph space and computing
time. Unlike frequentist approaches, it gives a principled and, in practice,
sensible approach for structure learning. We illustrate the efficiency of the
method on a broad range of simulated data. We then apply the method on
large-scale real applications from human and mammary gland gene expression
studies to show its empirical usefulness. In addition, we implemented the
method in the R package BDgraph which is freely available at
http://CRAN.R-project.org/package=BDgraph
| stat |
Spline-Based Bayesian Emulators for Large Scale Spatial Inverse Problems | A Bayesian approach to nonlinear inverse problems is considered where the
unknown quantity (input) is a random spatial field. The forward model is
complex and non-linear, therefore computationally expensive. An emulator-based
methodology is developed, where the Bayesian multivariate adaptive regression
splines (BMARS) are used to model the function that maps the inputs to the
outputs. Discrete cosine transformation (DCT) is used for dimension reduction
of the input spatial field. The posterior sampling is carried out using
trans-dimensional Markov Chain Monte Carlo (MCMC) methods. Numerical results
are presented by analyzing simulated as well as real data on hydrocarbon
reservoir characterization.
| stat |
Interpolating Population Distributions using Public-use Data with
Application to the American Community Survey | Statistical agencies publish aggregate estimates of various features of the
distributions of several socio-demographic quantities of interest based on data
obtained from a survey. Often these area-level estimates are tabulated at small
geographies, but detailed distributional information is not necessarily
available at such a fine scale geography due to data quality and/or disclosure
limitations. We propose a model-based method to interpolate the disseminated
estimates for a given variable of interest that improves on previous approaches
by simultaneously allowing for the use of more types of estimates,
incorporating the standard error of the estimates into the estimation process,
and by providing uncertainty quantification so that, for example, interval
estimates can be obtained for quantities of interest. Our motivating example
uses the disseminated tabulations and PUMS from the American Community Survey
to estimate U.S. Census tract-level income distributions and statistics
associated with these distributions.
| stat |
Bayesian graphical compositional regression for microbiome data | An important task in microbiome studies is to test the existence of and give
characterization to differences in the microbiome composition across groups of
samples. Important challenges of this problem include the large within-group
heterogeneities among samples and the existence of potential confounding
variables that, when ignored, increase the chance of false discoveries and
reduce the power for identifying true differences. We propose a probabilistic
framework to overcome these issues by combining three ideas: (i) a phylogenetic
tree-based decomposition of the cross-group comparison problem into a series of
local tests, (ii) a graphical model that links the local tests to allow
information sharing across taxa, and (iii) a Bayesian testing strategy that
incorporates covariates and integrates out the within-group variation, avoiding
potentially unstable point estimates. We derive an efficient inference
algorithm based on numerical integration and junction-tree message passing,
conduct extensive simulation studies to investigate the performance of our
approach, and compare it to state-of-the-art methods in a number of
representative settings. We then apply our method to the American Gut data to
analyze the association of dietary habits and human's gut microbiome
composition in the presence of covariates, and illustrate the importance of
incorporating covariates in microbiome cross-group comparison.
| stat |
Monte Carlo Gradient Estimation in Machine Learning | This paper is a broad and accessible survey of the methods we have at our
disposal for Monte Carlo gradient estimation in machine learning and across the
statistical sciences: the problem of computing the gradient of an expectation
of a function with respect to parameters defining the distribution that is
integrated; the problem of sensitivity analysis. In machine learning research,
this gradient problem lies at the core of many learning problems, in
supervised, unsupervised and reinforcement learning. We will generally seek to
rewrite such gradients in a form that allows for Monte Carlo estimation,
allowing them to be easily and efficiently used and analysed. We explore three
strategies--the pathwise, score function, and measure-valued gradient
estimators--exploring their historical development, derivation, and underlying
assumptions. We describe their use in other fields, show how they are related
and can be combined, and expand on their possible generalisations. Wherever
Monte Carlo gradient estimators have been derived and deployed in the past,
important advances have followed. A deeper and more widely-held understanding
of this problem will lead to further advances, and it is these advances that we
wish to support.
| stat |
Conscientious Classification: A Data Scientist's Guide to
Discrimination-Aware Classification | Recent research has helped to cultivate growing awareness that machine
learning systems fueled by big data can create or exacerbate troubling
disparities in society. Much of this research comes from outside of the
practicing data science community, leaving its members with little concrete
guidance to proactively address these concerns. This article introduces issues
of discrimination to the data science community on its own terms. In it, we
tour the familiar data mining process while providing a taxonomy of common
practices that have the potential to produce unintended discrimination. We also
survey how discrimination is commonly measured, and suggest how familiar
development processes can be augmented to mitigate systems' discriminatory
potential. We advocate that data scientists should be intentional about
modeling and reducing discriminatory outcomes. Without doing so, their efforts
will result in perpetuating any systemic discrimination that may exist, but
under a misleading veil of data-driven objectivity.
| stat |
Rao-Blackwellized Stochastic Gradients for Discrete Distributions | We wish to compute the gradient of an expectation over a finite or countably
infinite sample space having $K \leq \infty$ categories. When $K$ is indeed
infinite, or finite but very large, the relevant summation is intractable.
Accordingly, various stochastic gradient estimators have been proposed. In this
paper, we describe a technique that can be applied to reduce the variance of
any such estimator, without changing its bias---in particular, unbiasedness is
retained. We show that our technique is an instance of Rao-Blackwellization,
and we demonstrate the improvement it yields on a semi-supervised
classification problem and a pixel attention task.
| stat |
Doubly Semi-Implicit Variational Inference | We extend the existing framework of semi-implicit variational inference
(SIVI) and introduce doubly semi-implicit variational inference (DSIVI), a way
to perform variational inference and learning when both the approximate
posterior and the prior distribution are semi-implicit. In other words, DSIVI
performs inference in models where the prior and the posterior can be expressed
as an intractable infinite mixture of some analytic density with a highly
flexible implicit mixing distribution. We provide a sandwich bound on the
evidence lower bound (ELBO) objective that can be made arbitrarily tight.
Unlike discriminator-based and kernel-based approaches to implicit variational
inference, DSIVI optimizes a proper lower bound on ELBO that is asymptotically
exact. We evaluate DSIVI on a set of problems that benefit from implicit
priors. In particular, we show that DSIVI gives rise to a simple modification
of VampPrior, the current state-of-the-art prior for variational autoencoders,
which improves its performance.
| stat |
Reliability of decisions based on tests: Fourier analysis of Boolean
decision functions | Items in a test are often used as a basis for making decisions and such tests
are therefore required to have good psychometric properties, like
unidimensionality. In many cases the sum score is used in combination with a
threshold to decide between pass or fail, for instance. Here we consider
whether such a decision function is appropriate, without a latent variable
model, and which properties of a decision function are desirable. We consider
reliability (stability) of the decision function, i.e., does the decision
change upon perturbations, or changes in a fraction of the outcomes of the
items (measurement error). We are concerned with questions of whether the sum
score is the best way to aggregate the items, and if so why. We use ideas from
test theory, social choice theory, graphical models, computer science and
probability theory to answer these questions. We conclude that a weighted sum
score has desirable properties that (i) fit with test theory and is observable
(similar to a condition like conditional association), (ii) has the property
that a decision is stable (reliable), and (iii) satisfies Rousseau's criterion
that the input should match the decision. We use Fourier analysis of Boolean
functions to investigate whether a decision function is stable and to figure
out which (set of) items has proportionally too large an influence on the
decision. To apply these techniques we invoke ideas from graphical models and
use a pseudo-likelihood factorisation of the probability distribution.
| stat |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.