title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Distance approximation using Isolation Forests | This work briefly explores the possibility of approximating spatial distance
(alternatively, similarity) between data points using the Isolation Forest
method envisioned for outlier detection. The logic is similar to that of
isolation: the more similar or closer two points are, the more random splits it
will take to separate them. The separation depth between two points can be
standardized in the same way as the isolation depth, transforming it into a
distance metric that is limited in range, centered, and in compliance with the
axioms of distance. This metric presents some desirable properties such as
being invariant to the scales of variables or being able to account for
non-linear relationships between variables, which other metrics such as
Euclidean or Mahalanobis distance do not. Extensions to the Isolation Forest
method are also proposed for handling categorical variables and missing values,
resulting in a more generalizable and robust metric.
| stat |
Glucose values prediction five years ahead with a new framework of
missing responses in reproducing kernel Hilbert spaces, and the use of
continuous glucose monitoring technology | AEGIS study possesses unique information on longitudinal changes in
circulating glucose through continuous glucose monitoring technology (CGM).
However, as usual in longitudinal medical studies, there is a significant
amount of missing data in the outcome variables. For example, 40 percent of
glycosylated hemoglobin (A1C) biomarker data are missing five years ahead. With
the purpose to reduce the impact of this issue, this article proposes a new
data analysis framework based on learning in reproducing kernel Hilbert spaces
(RKHS) with missing responses that allows to capture non-linear relations
between variable studies in different supervised modeling tasks. First, we
extend the Hilbert-Schmidt dependence measure to test statistical independence
in this context introducing a new bootstrap procedure, for which we prove
consistency. Next, we adapt or use existing models of variable selection,
regression, and conformal inference to obtain new clinical findings about
glucose changes five years ahead with the AEGIS data. The most relevant
findings are summarized below: i) We identify new factors associated with
long-term glucose evolution; ii) We show the clinical sensibility of CGM data
to detect changes in glucose metabolism; iii) We can improve clinical
interventions based on our algorithms' expected glucose changes according to
patients' baseline characteristics.
| stat |
Heavy-tailed Representations, Text Polarity Classification & Data
Augmentation | The dominant approaches to text representation in natural language rely on
learning embeddings on massive corpora which have convenient properties such as
compositionality and distance preservation. In this paper, we develop a novel
method to learn a heavy-tailed embedding with desirable regularity properties
regarding the distributional tails, which allows to analyze the points far away
from the distribution bulk using the framework of multivariate extreme value
theory. In particular, a classifier dedicated to the tails of the proposed
embedding is obtained which performance outperforms the baseline. This
classifier exhibits a scale invariance property which we leverage by
introducing a novel text generation method for label preserving dataset
augmentation. Numerical experiments on synthetic and real text data demonstrate
the relevance of the proposed framework and confirm that this method generates
meaningful sentences with controllable attribute, e.g. positive or negative
sentiment.
| stat |
Why bigger is not always better: on finite and infinite neural networks | Recent work has argued that neural networks can be understood theoretically
by taking the number of channels to infinity, at which point the outputs become
Gaussian process (GP) distributed. However, we note that infinite Bayesian
neural networks lack a key facet of the behaviour of real neural networks: the
fixed kernel, determined only by network hyperparameters, implies that they
cannot do any form of representation learning. The lack of representation or
equivalently kernel learning leads to less flexibility and hence worse
performance, giving a potential explanation for the inferior performance of
infinite networks observed in the literature (e.g. Novak et al. 2019). We give
analytic results characterising the prior over representations and
representation learning in finite deep linear networks. We show empirically
that the representations in SOTA architectures such as ResNets trained with SGD
are much closer to those suggested by our deep linear results than by the
corresponding infinite network. This motivates the introduction of a new class
of network: infinite networks with bottlenecks, which inherit the theoretical
tractability of infinite networks while at the same time allowing
representation learning.
| stat |
Generative Learning of Counterfactual for Synthetic Control Applications
in Econometrics | A common statistical problem in econometrics is to estimate the impact of a
treatment on a treated unit given a control sample with untreated outcomes.
Here we develop a generative learning approach to this problem, learning the
probability distribution of the data, which can be used for downstream tasks
such as post-treatment counterfactual prediction and hypothesis testing. We use
control samples to transform the data to a Gaussian and homoschedastic form and
then perform Gaussian process analysis in Fourier space, evaluating the optimal
Gaussian kernel via non-parametric power spectrum estimation. We combine this
Gaussian prior with the data likelihood given by the pre-treatment data of the
single unit, to obtain the synthetic prediction of the unit post-treatment,
which minimizes the error variance of synthetic prediction. Given the
generative model the minimum variance counterfactual is unique, and comes with
an associated error covariance matrix. We extend this basic formalism to
include correlations of primary variable with other covariates of interest.
Given the probabilistic description of generative model we can compare
synthetic data prediction with real data to address the question of whether the
treatment had a statistically significant impact. For this purpose we develop a
hypothesis testing approach and evaluate the Bayes factor. We apply the method
to the well studied example of California (CA) tobacco sales tax of 1988. We
also perform a placebo analysis using control states to validate our
methodology. Our hypothesis testing method suggests 5.8:1 odds in favor of CA
tobacco sales tax having an impact on the tobacco sales, a value that is at
least three times higher than any of the 38 control states.
| stat |
How Does GAN-based Semi-supervised Learning Work? | Generative adversarial networks (GANs) have been widely used and have
achieved competitive results in semi-supervised learning. This paper
theoretically analyzes how GAN-based semi-supervised learning (GAN-SSL) works.
We first prove that, given a fixed generator, optimizing the discriminator of
GAN-SSL is equivalent to optimizing that of supervised learning. Thus, the
optimal discriminator in GAN-SSL is expected to be perfect on labeled data.
Then, if the perfect discriminator can further cause the optimization objective
to reach its theoretical maximum, the optimal generator will match the true
data distribution. Since it is impossible to reach the theoretical maximum in
practice, one cannot expect to obtain a perfect generator for generating data,
which is apparently different from the objective of GANs. Furthermore, if the
labeled data can traverse all connected subdomains of the data manifold, which
is reasonable in semi-supervised classification, we additionally expect the
optimal discriminator in GAN-SSL to also be perfect on unlabeled data. In
conclusion, the minimax optimization in GAN-SSL will theoretically output a
perfect discriminator on both labeled and unlabeled data by unexpectedly
learning an imperfect generator, i.e., GAN-SSL can effectively improve the
generalization ability of the discriminator by leveraging unlabeled
information.
| stat |
An Embedded Model Estimator for Non-Stationary Random Functions using
Multiple Secondary Variables | An algorithm for non-stationary spatial modelling using multiple secondary
variables is developed. It combines Geostatistics with Quantile Random Forests
to give a new interpolation and stochastic simulation algorithm. This paper
introduces the method and shows that it has consistency results that are
similar in nature to those applying to geostatistical modelling and to Quantile
Random Forests. The method allows for embedding of simpler interpolation
techniques, such as Kriging, to further condition the model. The algorithm
works by estimating a conditional distribution for the target variable at each
target location. The family of such distributions is called the envelope of the
target variable. From this, it is possible to obtain spatial estimates,
quantiles and uncertainty. An algorithm to produce conditional simulations from
the envelope is also developed. As they sample from the envelope, realizations
are therefore locally influenced by relative changes of importance of secondary
variables, trends and variability.
| stat |
Obesity Prediction with EHR Data: A deep learning approach with
interpretable elements | Childhood obesity is a major public health challenge. Obesity in early
childhood and adolescence can lead to obesity and other health problems in
adulthood. Early prediction and identification of the children at a high risk
of developing childhood obesity may help in engaging earlier and more effective
interventions to prevent and manage this and other related health conditions.
Existing predictive tools designed for childhood obesity primarily rely on
traditional regression-type methods without exploiting longitudinal patterns of
children's data (ignoring data temporality). In this paper, we present a
machine learning model specifically designed for predicting future obesity
patterns from generally available items on children's medical history. To do
this, we have used a large unaugmented EHR (Electronic Health Record) dataset
from a major pediatric health system in the US. We adopt a general LSTM (long
short-term memory) network architecture for our model for training over dynamic
(sequential) and static (demographic) EHR data. We have additionally included a
set embedding and attention layers to compute the feature ranking of each
timestamp and attention scores of each hidden layer corresponding to each input
timestamp. These feature ranking and attention scores added interpretability at
both the features and the timestamp-level.
| stat |
High-dimensional structure learning of binary pairwise Markov networks:
A comparative numerical study | Learning the undirected graph structure of a Markov network from data is a
problem that has received a lot of attention during the last few decades. As a
result of the general applicability of the model class, a myriad of methods
have been developed in parallel in several research fields. Recently, as the
size of the considered systems has increased, the focus of new methods has been
shifted towards the high-dimensional domain. In particular, introduction of the
pseudo-likelihood function has pushed the limits of score-based methods which
were originally based on the likelihood function. At the same time, methods
based on simple pairwise tests have been developed to meet the challenges
arising from increasingly large data sets in computational biology. Apart from
being applicable to high-dimensional problems, methods based on the
pseudo-likelihood and pairwise tests are fundamentally very different. To
compare the accuracy of the different types of methods, an extensive numerical
study is performed on data generated by binary pairwise Markov networks. A
parallelizable Gibbs sampler, based on restricted Boltzmann machines, is
proposed as a tool to efficiently sample from sparse high-dimensional networks.
The results of the study show that pairwise methods can be more accurate than
pseudo-likelihood methods in settings often encountered in high-dimensional
structure learning applications.
| stat |
A new tidy data structure to support exploration and modeling of
temporal data | Mining temporal data for information is often inhibited by a multitude of
formats: irregular or multiple time intervals, point events that need
aggregating, multiple observational units or repeated measurements on multiple
individuals, and heterogeneous data types. On the other hand, the software
supporting time series modeling and forecasting, makes strict assumptions on
the data to be provided, typically requiring a matrix of numeric data with
implicit time indexes. Going from raw data to model-ready data is painful. This
work presents a cohesive and conceptual framework for organizing and
manipulating temporal data, which in turn flows into visualization, modeling
and forecasting routines. Tidy data principles are extended to temporal data
by: (1) mapping the semantics of a dataset into its physical layout; (2)
including an explicitly declared index variable representing time; (3)
incorporating a "key" comprising single or multiple variables to uniquely
identify units over time. This tidy data representation most naturally supports
thinking of operations on the data as building blocks, forming part of a "data
pipeline" in time-based contexts. A sound data pipeline facilitates a fluent
workflow for analyzing temporal data. The infrastructure of tidy temporal data
has been implemented in the R package "tsibble".
| stat |
A central limit theorem for an omnibus embedding of multiple random
graphs and implications for multiscale network inference | Performing statistical analyses on collections of graphs is of import to many
disciplines, but principled, scalable methods for multi-sample graph inference
are few. Here we describe an "omnibus" embedding in which multiple graphs on
the same vertex set are jointly embedded into a single space with a distinct
representation for each graph. We prove a central limit theorem for this
embedding and demonstrate how it streamlines graph comparison, obviating the
need for pairwise subspace alignments. The omnibus embedding achieves
near-optimal inference accuracy when graphs arise from a common distribution
and yet retains discriminatory power as a test procedure for the comparison of
different graphs. Moreover, this joint embedding and the accompanying central
limit theorem are important for answering multiscale graph inference questions,
such as the identification of specific subgraphs or vertices responsible for
similarity or difference across networks. We illustrate this with a pair of
analyses of connectome data derived from dMRI and fMRI scans of human subjects.
In particular, we show that this embedding allows the identification of
specific brain regions associated with population-level differences. Finally,
we sketch how the omnibus embedding can be used to address pressing open
problems, both theoretical and practical, in multisample graph inference.
| stat |
On the variability of regression shrinkage methods for clinical
prediction models: simulation study on predictive performance | When developing risk prediction models, shrinkage methods are recommended,
especially when the sample size is limited. Several earlier studies have shown
that the shrinkage of model coefficients can reduce overfitting of the
prediction model and subsequently result in better predictive performance on
average. In this simulation study, we aimed to investigate the variability of
regression shrinkage on predictive performance for a binary outcome, with focus
on the calibration slope. The slope indicates whether risk predictions are too
extreme (slope < 1) or not extreme enough (slope > 1). We investigated the
following shrinkage methods in comparison to standard maximum likelihood
estimation: uniform shrinkage (likelihood-based and bootstrap-based), ridge
regression, penalized maximum likelihood, LASSO regression, adaptive LASSO,
non-negative garrote, and Firth's correction. There were three main findings.
First, shrinkage improved calibration slopes on average. Second, the
between-sample variability of calibration slopes was often increased relative
to maximum likelihood. Among the shrinkage methods, the bootstrap-based uniform
shrinkage worked well overall. In contrast to other shrinkage approaches,
Firth's correction had only a small shrinkage effect but did so with low
variability. Third, the correlation between the estimated shrinkage and the
optimal shrinkage to remove overfitting was typically negative. Hence, although
shrinkage improved predictions on average, it often worked poorly in individual
datasets, in particular when shrinkage was most needed. The observed
variability of shrinkage methods implies that these methods do not solve
problems associated with small sample size or low number of events per
variable.
| stat |
Modeling excess hazard with time--to--cure as a parameter | Cure models have been widely developed to estimate the cure fraction when
some subjects never experience the event of interest. However these models were
rarely focused on the estimation of the time-to-cure i.e. the delay elapsed
between the diagnosis and "the time from which cure is reached", an important
indicator, for instance to address the question of access to insurance or loans
for subjects with personal history of cancer. We propose a new excess hazard
regression model that includes the time-to-cure as a covariate dependent
parameter to be estimated. The model is written similarly to a Beta probability
distribution function and is shown to be a particular case of the non-mixture
cure models. Parameters are estimated through a maximum likelihood approach and
simulation studies demonstrate good performance of the model. Illustrative
applications to two cancer data sets are provided and some limitations as well
as possible extensions of the model are discussed. The proposed model offers a
simple and comprehensive way to estimate more accurately the time-to-cure. Key
words: Cancer; Cure model; Cure time; Net survival; Right to be forgotten.
| stat |
Enhancing Certifiable Robustness via a Deep Model Ensemble | We propose an algorithm to enhance certified robustness of a deep model
ensemble by optimally weighting each base model. Unlike previous works on using
ensembles to empirically improve robustness, our algorithm is based on
optimizing a guaranteed robustness certificate of neural networks. Our proposed
ensemble framework with certified robustness, RobBoost, formulates the optimal
model selection and weighting task as an optimization problem on a lower bound
of classification margin, which can be efficiently solved using coordinate
descent. Experiments show that our algorithm can form a more robust ensemble
than naively averaging all available models using robustly trained MNIST or
CIFAR base models. Additionally, our ensemble typically has better accuracy on
clean (unperturbed) data. RobBoost allows us to further improve certified
robustness and clean accuracy by creating an ensemble of already certified
models.
| stat |
Why did the distribution change? | We describe a formal approach based on graphical causal models to identify
the "root causes" of the change in the probability distribution of variables.
After factorizing the joint distribution into conditional distributions of each
variable, given its parents (the "causal mechanisms"), we attribute the change
to changes of these causal mechanisms. This attribution analysis accounts for
the fact that mechanisms often change independently and sometimes only some of
them change. Through simulations, we study the performance of our distribution
change attribution method. We then present a real-world case study identifying
the drivers of the difference in the income distribution between men and women.
| stat |
Modeling proportion of success in high school leaving examination- A
comparative study of Inflated Unit Lindley and Inflated Beta distribution | In this article, we first introduced the inflated unit Lindley distribution
considering zero or/and one inflation scenario and studied its basic
distributional and structural properties. Both the distributions are shown to
be members of exponential family with full rank. Different parameter estimation
methods are discussed and supporting simulation studies to check their efficacy
are also presented. Proportion of students passing the high school leaving
examination for the schools across the state of Manipur in India for the year
2020 are then modeled using the proposed distributions and compared with the
inflated beta distribution to justify its benefits.
| stat |
Statistical Methods for Replicability Assessment | Large-scale replication studies like the Reproducibility Project: Psychology
(RP:P) provide invaluable systematic data on scientific replicability, but most
analyses and interpretations of the data fail to agree on the definition of
"replicability" and disentangle the inexorable consequences of known selection
bias from competing explanations. We discuss three concrete definitions of
replicability based on (1) whether published findings about the signs of
effects are mostly correct, (2) how effective replication studies are in
reproducing whatever true effect size was present in the original experiment,
and (3) whether true effect sizes tend to diminish in replication. We apply
techniques from multiple testing and post-selection inference to develop new
methods that answer these questions while explicitly accounting for selection
bias. Our analyses suggest that the RP:P dataset is largely consistent with
publication bias due to selection of significant effects. The methods in this
paper make no distributional assumptions about the true effect sizes.
| stat |
Bayesian recurrent state space model for rs-fMRI | We propose a hierarchical Bayesian recurrent state space model for modeling
switching network connectivity in resting state fMRI data. Our model allows us
to uncover shared network patterns across disease conditions. We evaluate our
method on the ADNI2 dataset by inferring latent state patterns corresponding to
altered neural circuits in individuals with Mild Cognitive Impairment (MCI). In
addition to states shared across healthy and individuals with MCI, we discover
latent states that are predominantly observed in individuals with MCI. Our
model outperforms current state of the art deep learning method on ADNI2
dataset.
| stat |
Homogeneity Test for Functional Data based on Data-Depth Plots | One of the classic concerns in statistics is determining if two samples come
from thesame population, i.e. homogeneity testing. In this paper, we propose a
homogeneitytest in the context of Functional Data Analysis, adopting an idea
from multivariatedata analysis: the data depth plot (DD-plot). This DD-plot is
a generalization of theunivariate Q-Q plot (quantile-quantile plot). We propose
some statistics based onthese DD-plots, and we use bootstrapping techniques to
estimate their distributions.We estimate the finite-sample size and power of
our test via simulation, obtainingbetter results than other homogeneity test
proposed in the literature. Finally, weillustrate the procedure in samples of
real heterogeneous data and get consistent results.
| stat |
A Closed Form Solution to Best Rank-1 Tensor Approximation via KL
divergence Minimization | Tensor decomposition is a fundamentally challenging problem. Even the
simplest case of tensor decomposition, the rank-1 approximation in terms of the
Least Squares (LS) error, is known to be NP-hard. Here, we show that, if we
consider the KL divergence instead of the LS error, we can analytically derive
a closed form solution for the rank-1 tensor that minimizes the KL divergence
from a given positive tensor. Our key insight is to treat a positive tensor as
a probability distribution and formulate the process of rank-1 approximation as
a projection onto the set of rank-1 tensors. This enables us to solve rank-1
approximation by convex optimization. We empirically demonstrate that our
algorithm is an order of magnitude faster than the existing rank-1
approximation methods and gives better approximation of given tensors, which
supports our theoretical finding.
| stat |
Semi-Mechanistic Bayesian Modeling of COVID-19 with Renewal Processes | We propose a general Bayesian approach to modeling epidemics such as
COVID-19. The approach grew out of specific analyses conducted during the
pandemic, in particular an analysis concerning the effects of
non-pharmaceutical interventions (NPIs) in reducing COVID-19 transmission in 11
European countries. The model parameterizes the time varying reproduction
number $R_t$ through a regression framework in which covariates can e.g be
governmental interventions or changes in mobility patterns. This allows a joint
fit across regions and partial pooling to share strength. This innovation was
critical to our timely estimates of the impact of lockdown and other NPIs in
the European epidemics, whose validity was borne out by the subsequent course
of the epidemic. Our framework provides a fully generative model for latent
infections and observations deriving from them, including deaths, cases,
hospitalizations, ICU admissions and seroprevalence surveys. One issue
surrounding our model's use during the COVID-19 pandemic is the confounded
nature of NPIs and mobility. We use our framework to explore this issue. We
have open sourced an R package epidemia implementing our approach in Stan.
Versions of the model are used by New York State, Tennessee and Scotland to
estimate the current situation and make policy decisions.
| stat |
Robust Optimisation Monte Carlo | This paper is on Bayesian inference for parametric statistical models that
are defined by a stochastic simulator which specifies how data is generated.
Exact sampling is then possible but evaluating the likelihood function is
typically prohibitively expensive. Approximate Bayesian Computation (ABC) is a
framework to perform approximate inference in such situations. While basic ABC
algorithms are widely applicable, they are notoriously slow and much research
has focused on increasing their efficiency. Optimisation Monte Carlo (OMC) has
recently been proposed as an efficient and embarrassingly parallel method that
leverages optimisation to accelerate the inference. In this paper, we
demonstrate an important previously unrecognised failure mode of OMC: It
generates strongly overconfident approximations by collapsing regions of
similar or near-constant likelihood into a single point. We propose an
efficient, robust generalisation of OMC that corrects this. It makes fewer
assumptions, retains the main benefits of OMC, and can be performed either as
post-processing to OMC or as a stand-alone computation. We demonstrate the
effectiveness of the proposed Robust OMC on toy examples and tasks in
inverse-graphics where we perform Bayesian inference with a complex image
renderer.
| stat |
The Variational Bandwidth Bottleneck: Stochastic Evaluation on an
Information Budget | In many applications, it is desirable to extract only the relevant
information from complex input data, which involves making a decision about
which input features are relevant. The information bottleneck method formalizes
this as an information-theoretic optimization problem by maintaining an optimal
tradeoff between compression (throwing away irrelevant input information), and
predicting the target. In many problem settings, including the reinforcement
learning problems we consider in this work, we might prefer to compress only
part of the input. This is typically the case when we have a standard
conditioning input, such as a state observation, and a "privileged" input,
which might correspond to the goal of a task, the output of a costly planning
algorithm, or communication with another agent. In such cases, we might prefer
to compress the privileged input, either to achieve better generalization
(e.g., with respect to goals) or to minimize access to costly information
(e.g., in the case of communication). Practical implementations of the
information bottleneck based on variational inference require access to the
privileged input in order to compute the bottleneck variable, so although they
perform compression, this compression operation itself needs unrestricted,
lossless access. In this work, we propose the variational bandwidth bottleneck,
which decides for each example on the estimated value of the privileged
information before seeing it, i.e., only based on the standard input, and then
accordingly chooses stochastically, whether to access the privileged input or
not. We formulate a tractable approximation to this framework and demonstrate
in a series of reinforcement learning experiments that it can improve
generalization and reduce access to computationally costly information.
| stat |
Spatial multiresolution analysis approach to identify crash hotspots and
estimate crash risk | In this paper, the authors evaluate the performance of a spatial
multiresolution analysis (SMA) method that behaves like a variable bandwidth
kernel density estimation (KDE) method, for hazardous road segments
identification (HRSI) and crash risk (expected number of crashes) estimation.
The proposed SMA, is similar to the KDE method with the additional benefit of
allowing for the bandwidth to be different at different road segments depending
on how homogenous the segments are. Furthermore, the optimal bandwidth at each
road segment is determined solely based on the data by minimizing an unbiased
estimate of the mean square error. The authors compare the SMA method with the
state of the practice crash analysis method, the empirical Bayes (EB) method,
in terms of their HRSI ability and their ability to predict future crashes. The
results indicate that the SMA may outperform the EB method, at least with the
crash data of the entire Virginia interstate network used in this paper. The
SMA is implemented in an Excel spreadsheet that is freely available for
download.
| stat |
Finding Stable Groups of Cross-Correlated Features in Multi-View data | Multi-view data, in which data of different types are obtained from a common
set of samples, is now common in many scientific problems. An important problem
in the analysis of multi-view data is identifying interactions between groups
of features from different data types. A bimodule is a pair $(A,B)$ of feature
sets from two different data types such that the aggregate cross-correlation
between the features in $A$ and those in $B$ is large. A bimodule $(A,B)$ is
stable if $A$ coincides with the set of features having significant aggregate
correlation with the features in $B$, and vice-versa. At the population level,
stable bimodules correspond to connected components of the cross-correlation
network, which is the bipartite graph whose edges are pairs of features with
non-zero cross-correlations.
We develop an iterative, testing-based procedure, called BSP, to identify
stable bimodules in two moderate- to high-dimensional data sets. BSP relies on
permutation-based p-values for sums of squared cross-correlations. We
efficiently approximate the p-values using tail probabilities of gamma
distributions that are fit using analytical estimates of the permutation
moments of the test statistic. Our moment estimates depend on the eigenvalues
of the intra-correlation matrices of $A$ and $B$ and as a result, the
significance of observed cross-correlations accounts for the correlations
within each data type.
We carry out a thorough simulation study to assess the performance of BSP,
and present an extended application of BSP to the problem of expression
quantitative trait loci (eQTL) analysis using recent data from the GTEx
project. In addition, we apply BSP to climatology data in order to identify
regions in North America where annual temperature variation affects
precipitation.
| stat |
Models for Genetic Diversity Generated by Negative Binomial Point
Processes | We develop a model based on a generalised Poisson-Dirichlet distribution for
the analysis of genetic diversity, and illustrate its use on microsatellite
data for the genus Dasyurus (the quoll, a marsupial carnivore listed as
near-threatened in Australia). Our class of distributions, termed
$PD_\alpha^{(r)}$, is constructed from a negative binomial point process,
generalizing the usual one-parameter $PD_\alpha$ model, which is constructed
from a Poisson point process. Both models have at their heart a
Stable$(\alpha)$ process, but in $PD_\alpha^{(r)}$, an extra parameter $r>0$
adds flexibility, analogous to the way the negative binomial distribution
allows for "overdispersion" in the analysis of count data. A key result
obtained is a generalised version of Ewens' sampling formula for
$PD_\alpha^{(r)}$. We outline the theoretical basis for the model, and, for the
quolls data, estimate the parameters $\alpha$ and r by least squares, showing
how the extra parameter r aids in the interpretability of the data by
comparison with the standard $PD_\alpha$ model. The methods potentially have
implications for the management and conservation of threatened populations.
| stat |
Transparent Privacy is Principled Privacy | Differential privacy revolutionizes the way we think about statistical
disclosure limitation. Among the benefits it brings to the table, one is
particularly profound and impactful. Under this formal approach to privacy, the
mechanism with which data is privatized can be spelled out in full
transparency, without sacrificing the privacy guarantee. Curators of
open-source demographic and scientific data are at a position to offer privacy
without obscurity. This paper supplies a technical treatment to the pitfalls of
obscure privacy, and establishes transparent privacy as a prerequisite to
drawing correct statistical inference. It advocates conceiving transparent
privacy as a dynamic component that can improve data quality from the total
survey error perspective, and discusses the limited statistical usability of
mere procedural transparency which may arise when dealing with mandated
invariants. Transparent privacy is the only viable path towards principled
inference from privatized data releases. Its arrival marks great progress
towards improved reproducibility, accountability and public trust.
| stat |
Doubly Robust Covariate Shift Regression with Semi-nonparametric
Nuisance Models | In contemporary statistical learning, covariate shift correction plays an
important role when distribution of the testing data is shifted from the
training data. Importance weighting is used to adjust for this but is not
robust to model misspecifcation or excessive estimation error. In this paper,
we propose a doubly robust covariate shift regression approach that introduces
an imputation model for the targeted response, and uses it to augment the
importance weighting equation. With a novel semi-nonparametric construction for
the two nuisance models, our method is less prone to the curse of
dimensionality compared to the nonparametric approaches, and is less prone to
model mis-specification than the parametric approach. To remove the overfitting
bias of the nonparametric components under potential model mis-specification,
we construct calibrated moment estimating equations for the semi-nonparametric
models. We show that our estimator is root-n consistent when at least one
nuisance model is correctly specified, estimation for the parametric part of
the nuisance models achieves parametric rate, and the nonparametric components
are rate doubly robust. Simulation studies demonstrate that our method is more
robust and efficient than existing parametric and fully nonparametric (machine
learning) estimators under various configurations. We also examine the utility
of our method through a real example about transfer learning of phenotyping
algorithm for bipolar disorder. Finally, we propose ways to improve the
(intrinsic) efficiency of our estimator and to incorporate high dimensional or
machine learning models with our proposed framework.
| stat |
Improved Confidence Regions in Meta-analysis of Diagnostic Test Accuracy | Meta-analyses of diagnostic test accuracy (DTA) studies have been gathering
attention in research in clinical epidemiology and health technology
development, and bivariate random-effects model is becoming a standard tool.
However, standard inference methods usually underestimate statistical errors
and possibly provide highly overconfident results under realistic situations
since they ignore the variability in the estimation of variance parameters. To
overcome the difficulty, a new improved inference method, namely, an accurate
confidence region for the meta-analysis of DTA, by asymptotically expanding the
coverage probability of the standard confidence region. The advantage of the
proposed confidence region is that it holds a relatively simple expression and
does not require any repeated calculations such as Bootstrap or Monte Carlo
methods to compute the region, thereby the proposed method can be easily
carried out in practical applications. The effectiveness of the proposed method
is demonstrated through simulation studies and an application to meta-analysis
of screening test accuracy for alcohol problems.
| stat |
Comparing the Performance of Statistical Adjustment Methods By
Recovering the Experimental Benchmark from the REFLUX Trial | Much evidence in comparative effectiveness research is based on observational
studies. Researchers who conduct observational studies typically assume that
there are no unobservable differences between the treated and control groups.
Treatment effects are estimated after adjusting for observed differences
between treated and controls. However, treatment effect estimates may be biased
due to model misspecification. That is, if the method of treatment effect
estimation imposes unduly strong functional form assumptions, treatment effect
estimates may be significantly biased. In this study, we compare the
performance of a wide variety of treatment effect estimation methods. We do so
within the context of the REFLUX study from the UK. In REFLUX, after study
qualification, participants were enrolled in either a randomized trial arm or
patient preference arm. In the randomized trial, patients were randomly
assigned to either surgery or medical management. In the patient preference
arm, participants selected to either have surgery or medical management. We
attempt to recover the treatment effect estimate from the randomized trial arm
using the data from the patient preference arm of the study. We vary the method
of treatment effect estimation and record which methods are successful and
which are not. We apply over 20 different methods including standard regression
models as well as advanced machine learning methods. We find that simple
propensity score matching methods perform the worst. We also find significant
variation in performance across methods. The wide variation in performance
suggests analysts should use multiple methods of estimation as a robustness
check.
| stat |
Estimating Monte Carlo variance from multiple Markov chains | The ever-increasing power of the personal computer has led to easy parallel
implementations of Markov chain Monte Carlo (MCMC). However, almost all work in
estimating the variance of Monte Carlo averages, including the efficient batch
means (BM) estimator, focuses on a single-chain MCMC run. We demonstrate that
simply averaging covariance matrix estimators from multiple chains (average BM)
can yield critical underestimates in small sample sizes, especially for slow
mixing Markov chains. We propose a multivariate replicated batch means (RBM)
estimator that utilizes information from parallel chains, thereby correcting
for the underestimation. Under weak conditions on the mixing rate of the
process, the RBM and ABM estimator are both strongly consistent and exhibit
similar large-sample bias and variance. However, in small runs the RBM
estimator can be dramatically superior. This is demonstrated through a variety
of examples, including a two-variable Gibbs sampler for a bivariate Gaussian
target distribution. Here, we obtain a closed-form expression for the
asymptotic covariance matrix of the Monte Carlo estimator, a useful result for
benchmarking in the future.
| stat |
Comparison of ARIMA, ETS, NNAR and hybrid models to forecast the second
wave of COVID-19 hospitalizations in Italy | Coronavirus disease (COVID-19) is a severe ongoing novel pandemic that has
emerged in Wuhan, China, in December 2019. As of October 13, the outbreak has
spread rapidly across the world, affecting over 38 million people, and causing
over 1 million deaths. In this article, I analysed several time series
forecasting methods to predict the spread of COVID-19 second wave in Italy,
over the period after October 13, 2020. I used an autoregressive model (ARIMA),
an exponential smoothing state space model (ETS), a neural network
autoregression model (NNAR), and the following hybrid combinations of them:
ARIMA-ETS, ARIMA-NNAR, ETS-NNAR, and ARIMA-ETS-NNAR. About the data, I
forecasted the number of patients hospitalized with mild symptoms, and in
intensive care units (ICU). The data refer to the period February 21,
2020-October 13, 2020 and are extracted from the website of the Italian
Ministry of Health (www.salute.gov.it). The results show that i) the hybrid
models, except for ARIMA-ETS, are better at capturing the linear and non-linear
epidemic patterns, by outperforming the respective single models; and ii) the
number of COVID-19-related hospitalized with mild symptoms and in ICU will
rapidly increase in the next weeks, by reaching the peak in about 50-60 days,
i.e. in mid-December 2020, at least. To tackle the upcoming COVID-19 second
wave, on one hand, it is necessary to hire healthcare workers and implement
sufficient hospital facilities, protective equipment, and ordinary and
intensive care beds; and on the other hand, it may be useful to enhance social
distancing by improving public transport and adopting the double-shifts
schooling system, for example.
| stat |
Regularity as Regularization: Smooth and Strongly Convex Brenier
Potentials in Optimal Transport | Estimating Wasserstein distances between two high-dimensional densities
suffers from the curse of dimensionality: one needs an exponential (wrt
dimension) number of samples to ensure that the distance between two empirical
measures is comparable to the distance between the original densities.
Therefore, optimal transport (OT) can only be used in machine learning if it is
substantially regularized. On the other hand, one of the greatest achievements
of the OT literature in recent years lies in regularity theory: Caffarelli
showed that the OT map between two well behaved measures is Lipschitz, or
equivalently when considering 2-Wasserstein distances, that Brenier convex
potentials (whose gradient yields an optimal map) are smooth. We propose in
this work to draw inspiration from this theory and use regularity as a
regularization tool. We give algorithms operating on two discrete measures that
can recover nearly optimal transport maps with small distortion, or
equivalently, nearly optimal Brenier potentials that are strongly convex and
smooth. The problem boils down to solving alternatively a convex QCQP and a
discrete OT problem, granting access to the values and gradients of the Brenier
potential not only on sampled points, but also out of sample at the cost of
solving a simpler QCQP for each evaluation. We propose algorithms to estimate
and evaluate transport maps with desired regularity properties, benchmark their
statistical performance, apply them to domain adaptation and visualize their
action on a color transfer task.
| stat |
Adjusting for Partial Compliance in SMARTs: a Bayesian Semiparametric
Approach | The cyclical and heterogeneous nature of many substance use disorders
highlights the need to adapt the type or the dose of treatment to accommodate
the specific and changing needs of individuals. The Adaptive Treatment for
Alcohol and Cocaine Dependence study (ENGAGE) is a multi-stage randomized trial
that aimed to provide longitudinal data for constructing treatment strategies
to improve patients' engagement in therapy. However, the high rate of
noncompliance and lack of analytic tools to account for noncompliance have
impeded researchers from using the data to achieve the main goal of the trial.
We overcome this issue by defining our target parameter as the mean outcome
under different treatment strategies for given potential compliance strata and
propose a Bayesian semiparametric model to estimate this quantity. While it
adds substantial complexities to the analysis, one important feature of our
work is that we consider partial rather than binary compliance classes which is
more relevant in longitudinal studies. We assess the performance of our method
through comprehensive simulation studies. We illustrate its application on the
ENGAGE study and demonstrate that the optimal treatment strategy depends on
compliance strata.
| stat |
Informed Pooled Testing with Quantitative Assays | Pooled testing is widely used for screening for viral or bacterial infections
with low prevalence when individual testing is not cost-efficient. Pooled
testing with qualitative assays that give binary results has been well-studied.
However, characteristics of pooling with quantitative assays were mostly
demonstrated using simulations or empirical studies. We investigate properties
of three pooling strategies with quantitative assays: traditional two-stage
mini-pooling (MP) (Dorfman, 1943), mini-pooling with deconvolution algorithm
(MPA) (May et al., 2010), and marker-assisted MPA (mMPA) (Liu et al., 2017).
MPA and mMPA test individuals in a sequence after a positive pool and implement
a deconvolution algorithm to determine when testing can cease to ascertain all
individual statuses. mMPA uses information from other available markers to
determine an optimal order for individual testings. We derive and compare the
general statistical properties of the three pooling methods. We show that with
a proper pool size, MP, MPA, and mMPA can be more cost-efficient than
individual testing, and mMPA is superior to MPA and MP. For diagnostic
accuracy, mMPA and MPA have higher specificity and positive predictive value
but lower sensitivity and negative predictive value than MP and individual
testing. Included in this paper are applications to various simulations and an
application for HIV treatment monitoring.
| stat |
An ordinal measure of interrater absolute agreement | A measure of interrater absolute agreement for ordinal scales is proposed
capitalizing on the dispersion index for ordinal variables proposed by Giuseppe
Leti. The procedure allows to avoid the problem of restriction of variance that
sometimes affect traditional measures of interrater agreement in different
fields of application. An unbiased estimator of the proposed measure is
introduced and its sampling properties are investigated. In order to construct
confidence intervals for interrater absolute agreement both asymptotic results
and bootstrapping methods are used and their performance is evaluated.
Simulated data are employed to demonstrate the accuracy and practical utility
of the new procedure for assessing agreement. Finally, an application to a real
case is provided.
| stat |
Combined Tail Estimation Using Censored Data and Expert Information | We study tail estimation in Pareto-like settings for datasets with a high
percentage of randomly right-censored data, and where some expert information
on the tail index is available for the censored observations. This setting
arises for instance naturally for liability insurance claims, where actuarial
experts build reserves based on the specificity of each open claim, which can
be used to improve the estimation based on the already available data points
from closed claims. Through an entropy-perturbed likelihood we derive an
explicit estimator and establish a close analogy with Bayesian methods.
Embedded in an extreme value approach, asymptotic normality of the estimator is
shown, and when the expert is clair-voyant, a simple combination formula can be
deduced, bridging the classical statistical approach with the expert
information. Following the aforementioned combination formula, a combination of
quantile estimators can be naturally defined. In a simulation study, the
estimator is shown to often outperform the Hill estimator for censored
observations and recent Bayesian solutions, some of which require more
information than usually available. Finally we perform a case study on a motor
third-party liability insurance claim dataset, where Hill-type and quantile
plots incorporate ultimate values into the estimation procedure in an intuitive
manner.
| stat |
A permutation-based Bayesian approach for inverse covariance estimation | Covariance estimation and selection for multivariate datasets in a
high-dimensional regime is a fundamental problem in modern statistics. Gaussian
graphical models are a popular class of models used for this purpose. Current
Bayesian methods for inverse covariance matrix estimation under Gaussian
graphical models require the underlying graph and hence the ordering of
variables to be known. However, in practice, such information on the true
underlying model is often unavailable. We therefore propose a novel
permutation-based Bayesian approach to tackle the unknown variable ordering
issue. In particular, we utilize multiple maximum a posteriori estimates under
the DAG-Wishart prior for each permutation, and subsequently construct the
final estimate of the inverse covariance matrix. The proposed estimator has
smaller variability and yields order-invariant property. We establish posterior
convergence rates under mild assumptions and illustrate that our method
outperforms existing approaches in estimating the inverse covariance matrices
via simulation studies.
| stat |
Unbiased MLMC stochastic gradient-based optimization of Bayesian
experimental designs | In this paper we propose an efficient stochastic optimization algorithm to
search for Bayesian experimental designs such that the expected information
gain is maximized. The gradient of the expected information gain with respect
to experimental design parameters is given by a nested expectation, for which
the standard Monte Carlo method using a fixed number of inner samples yields a
biased estimator. In this paper, applying the idea of randomized multilevel
Monte Carlo (MLMC) methods, we introduce an unbiased Monte Carlo estimator for
the gradient of the expected information gain with finite expected squared
$\ell_2$-norm and finite expected computational cost per sample. Our unbiased
estimator can be combined well with stochastic gradient descent algorithms,
which results in our proposal of an optimization algorithm to search for an
optimal Bayesian experimental design. Numerical experiments confirm that our
proposed algorithm works well not only for a simple test problem but also for a
more realistic pharmacokinetic problem.
| stat |
A Hierarchical Meta-Analysis for Settings Involving Multiple Outcomes
across Multiple Cohorts | Evidence from animal models and epidemiological studies has linked prenatal
alcohol exposure (PAE) to a broad range of long-term cognitive and behavioral
deficits. However, there is virtually no information in the scientific
literature regarding the levels of PAE associated with an increased risk of
clinically significant adverse effects. During the period from 1975-1993,
several prospective longitudinal cohort studies were conducted in the U.S., in
which maternal reports regarding alcohol use were obtained during pregnancy and
the cognitive development of the offspring was assessed from early childhood
through early adulthood. The sample sizes in these cohorts did not provide
sufficient power to examine effects associated with different levels and
patterns of PAE. To address this critical public health issue, we have
developed a hierarchical meta-analysis to synthesize information regarding the
effects of PAE on cognition, integrating data on multiple endpoints from six
U.S. longitudinal cohort studies. Our approach involves estimating the
dose-response coefficients for each endpoint and then pooling these correlated
dose-response coefficients to obtain an estimated `global' effect of exposure
on cognition. In the first stage, we use individual participant data to derive
estimates of the effects of PAE by fitting regression models that adjust for
potential confounding variables using propensity scores. The correlation matrix
characterizing the dependence between the endpoint-specific dose-response
coefficients estimated within each cohort is then run, while accommodating
incomplete information on some endpoints. We also compare and discuss
inferences based on the proposed approach to inferences based on a full
multivariate analysis
| stat |
Unsupervised Learning of GMM with a Uniform Background Component | Gaussian Mixture Models are one of the most studied and mature models in
unsupervised learning. However, outliers are often present in the data and
could influence the cluster estimation. In this paper, we study a new model
that assumes that data comes from a mixture of a number of Gaussians as well as
a uniform ``background'' component assumed to contain outliers and other
non-interesting observations. We develop a novel method based on robust loss
minimization that performs well in clustering such GMM with a uniform
background. We give theoretical guarantees for our clustering algorithm to
obtain best clustering results with high probability. Besides, we show that the
result of our algorithm does not depend on initialization or local optima, and
the parameter tuning is an easy task. By numeric simulations, we demonstrate
that our algorithm enjoys high accuracy and achieves the best clustering
results given a large enough sample size. Finally, experimental comparisons
with typical clustering methods on real datasets witness the potential of our
algorithm in real applications.
| stat |
Randomized Exploration for Non-Stationary Stochastic Linear Bandits | We investigate two perturbation approaches to overcome conservatism that
optimism based algorithms chronically suffer from in practice. The first
approach replaces optimism with a simple randomization when using confidence
sets. The second one adds random perturbations to its current estimate before
maximizing the expected reward. For non-stationary linear bandits, where each
action is associated with a $d$-dimensional feature and the unknown parameter
is time-varying with total variation $B_T$, we propose two randomized
algorithms, Discounted Randomized LinUCB (D-RandLinUCB) and Discounted Linear
Thompson Sampling (D-LinTS) via the two perturbation approaches. We highlight
the statistical optimality versus computational efficiency trade-off between
them in that the former asymptotically achieves the optimal dynamic regret
$\tilde{\mathcal{O}}(d ^{2/3}B_T^{1/3} T^{2/3})$, but the latter is
oracle-efficient with an extra logarithmic factor in the number of arms
compared to minimax-optimal dynamic regret. In a simulation study, both
algorithms show outstanding performance in tackling conservatism issue that
Discounted LinUCB struggles with.
| stat |
Regularized Optimal Transport for Dynamic Semi-supervised Learning | Semi-supervised learning provides an effective paradigm for leveraging
unlabeled data to improve a model's performance. Among the many strategies
proposed, graph-based methods have shown excellent properties, in particular
since they allow to solve directly the transductive tasks according to Vapnik's
principle and they can be extended efficiently for inductive tasks. In this
paper, we propose a novel approach for the transductive semi-supervised
learning, using a complete bipartite edge-weighted graph. The proposed approach
uses the regularized optimal transport between empirical measures defined on
labelled and unlabelled data points in order to obtain an affinity matrix from
the optimal transport plan. This matrix is further used to propagate labels
through the vertices of the graph in an incremental process ensuring the
certainty of the predictions by incorporating a certainty score based on
Shannon's entropy. We also analyze the convergence of our approach and we
derive an efficient way to extend it for out-of-sample data. Experimental
analysis was used to compare the proposed approach with other label propagation
algorithms on 12 benchmark datasets, for which we surpass state-of-the-art
results. We release our code.
| stat |
Marginally Interpretable Linear Transformation Models for Clustered
Observations | Clustered observations are ubiquitous in controlled and observational studies
and arise naturally in multicenter trials or longitudinal surveys. I present
two novel models for the analysis of clustered observations where the marginal
distributions are described by a linear transformation model and the
correlations by a joint multivariate normal distribution. Both models provide
analytic formulae for the marginal distributions, one of which features
directly interpretable parameters. Owing to the richness of transformation
models, the techniques are applicable to any type of response variable,
including bounded, skewed, binary, ordinal, or survival responses. I present
re-analyses of five applications from different domains, including models for
non-normal and discrete responses, and explain how specific models for the
estimation of marginal distributions can be defined within this novel modelling
framework and how the results can be interpreted in a marginal way.
| stat |
Bayesian Quantile Matching Estimation | Due to increased awareness of data protection and corresponding laws many
data, especially involving sensitive personal information, are not publicly
accessible. Accordingly, many data collecting agencies only release aggregated
data, e.g. providing the mean and selected quantiles of population
distributions. Yet, research and scientific understanding, e.g. for medical
diagnostics or policy advice, often relies on data access. To overcome this
tension, we propose a Bayesian method for learning from quantile information.
Being based on order statistics of finite samples our method adequately and
correctly reflects the uncertainty of empirical quantiles. After outlining the
theory, we apply our method to simulated as well as real world examples.
| stat |
PCAS: Pruning Channels with Attention Statistics for Deep Network
Compression | Compression techniques for deep neural networks are important for
implementing them on small embedded devices. In particular, channel-pruning is
a useful technique for realizing compact networks. However, many conventional
methods require manual setting of compression ratios in each layer. It is
difficult to analyze the relationships between all layers, especially for
deeper models. To address these issues, we propose a simple channel-pruning
technique based on attention statistics that enables to evaluate the importance
of channels. We improved the method by means of a criterion for automatic
channel selection, using a single compression ratio for the entire model in
place of per-layer model analysis. The proposed approach achieved superior
performance over conventional methods with respect to accuracy and the
computational costs for various models and datasets. We provide analysis
results for behavior of the proposed criterion on different datasets to
demonstrate its favorable properties for channel pruning.
| stat |
Bayesian Spatial Binary Regression for Label Fusion in Structural
Neuroimaging | Many analyses of neuroimaging data involve studying one or more regions of
interest (ROIs) in a brain image. In order to do so, each ROI must first be
identified. Since every brain is unique, the location, size, and shape of each
ROI varies across subjects. Thus, each ROI in a brain image must either be
manually identified or (semi-) automatically delineated, a task referred to as
segmentation. Automatic segmentation often involves mapping a previously
manually segmented image to a new brain image and propagating the labels to
obtain an estimate of where each ROI is located in the new image. A more recent
approach to this problem is to propagate labels from multiple manually
segmented atlases and combine the results using a process known as label
fusion. To date, most label fusion algorithms either employ voting procedures
or impose prior structure and subsequently find the maximum a posteriori
estimator (i.e., the posterior mode) through optimization. We propose using a
fully Bayesian spatial regression model for label fusion that facilitates
direct incorporation of covariate information while making accessible the
entire posterior distribution. We discuss the implementation of our model via
Markov chain Monte Carlo and illustrate the procedure through both simulation
and application to segmentation of the hippocampus, an anatomical structure
known to be associated with Alzheimer's disease.
| stat |
Two equalities expressing the determinant of a matrix in terms of
expectations over matrix-vector products | We introduce two equations expressing the inverse determinant of a full rank
matrix $\mathbf{A} \in \mathbb{R}^{n \times n}$ in terms of expectations over
matrix-vector products. The first relationship is $|\mathrm{det}
(\mathbf{A})|^{-1} = \mathbb{E}_{\mathbf{s} \sim \mathcal{S}^{n-1}}\bigl[\,
\Vert \mathbf{As}\Vert^{-n} \bigr]$, where expectations are over vectors drawn
uniformly on the surface of an $n$-dimensional radius one hypersphere. The
second relationship is $|\mathrm{det}(\mathbf{A})|^{-1} =
\mathbb{E}_{\mathbf{x} \sim q}[\,p(\mathbf{Ax}) /\, q(\mathbf{x})]$, where $p$
and $q$ are smooth distributions, and $q$ has full support.
| stat |
Noise Contrastive Meta-Learning for Conditional Density Estimation using
Kernel Mean Embeddings | Current meta-learning approaches focus on learning functional representations
of relationships between variables, i.e. on estimating conditional expectations
in regression. In many applications, however, we are faced with conditional
distributions which cannot be meaningfully summarized using expectation only
(due to e.g. multimodality). Hence, we consider the problem of conditional
density estimation in the meta-learning setting. We introduce a novel technique
for meta-learning which combines neural representation and noise-contrastive
estimation with the established literature of conditional mean embeddings into
reproducing kernel Hilbert spaces. The method is validated on synthetic and
real-world problems, demonstrating the utility of sharing learned
representations across multiple conditional density estimation tasks.
| stat |
Non-linear regression models for behavioral and neural data analysis | Regression models are popular tools in empirical sciences to infer the
influence of a set of variables onto a dependent variable given an experimental
dataset. In neuroscience and cognitive psychology, Generalized Linear Models
(GLMs) -including linear regression, logistic regression, and Poisson GLM- is
the regression model of choice to study the factors that drive participant's
choices, reaction times and neural activations. These methods are however
limited as they only capture linear contributions of each regressors. Here, we
introduce an extension of GLMs called Generalized Unrestricted Models (GUMs),
which allows to infer a much richer set of contributions of the regressors to
the dependent variable, including possible interactions between the regressors.
In a GUM, each regressor is passed through a linear or nonlinear function, and
the contribution of the different resulting transformed regressors can be
summed or multiplied to generate a predictor for the dependent variable. We
propose a Bayesian treatment of these models in which we endow functions with
Gaussian Process priors, and we present two methods to compute a posterior over
the functions given a dataset: the Laplace method and a sparse variational
approach, which scales better for large dataset. For each method, we assess the
quality of the model estimation and we detail how the hyperparameters (defining
for example the expected smoothness of the function) can be fitted. Finally, we
illustrate the power of the method on a behavioral dataset where subjects
reported the average perceived orientation of a series of gratings. The method
allows to recover the mapping of the grating angle onto perceptual evidence for
each subject, as well as the impact of the grating based on its position.
Overall, GUMs provides a very rich and flexible framework to run nonlinear
regression analysis in neuroscience, psychology, and beyond.
| stat |
Big Variates: Visualizing and identifying key variables in a
multivariate world | Big Data involves both a large number of events but also many variables. This
paper will concentrate on the challenge presented by the large number of
variables in a Big Dataset. It will start with a brief review of exploratory
data visualisation for large dimensional datasets and the use of parallel
coordinates. This motivates the use of information theoretic ideas to
understand multivariate data. Two key information-theoretic statistics
(Similarity Index and Class Distance Indicator) will be described which are
used to identify the key variables and then guide the user in a subsequent
machine learning analysis. Key to the approach is a novel algorithm to
histogram data which quantifies the information content of the data. The Class
Distance Indicator also sets a limit on the classification performance of
machine learning algorithms for the specific dataset.
| stat |
Evaluation Uncertainty in Data-Driven Self-Driving Testing | Safety evaluation of self-driving technologies has been extensively studied.
One recent approach uses Monte Carlo based evaluation to estimate the
occurrence probabilities of safety-critical events as safety measures. These
Monte Carlo samples are generated from stochastic input models constructed
based on real-world data. In this paper, we propose an approach to assess the
impact on the probability estimates from the evaluation procedures due to the
estimation error caused by data variability. Our proposed method merges the
classical bootstrap method for estimating input uncertainty with a likelihood
ratio based scheme to reuse experiment outputs. This approach is economical and
efficient in terms of implementation costs in assessing input uncertainty for
the evaluation of self-driving technology. We use an example in autonomous
vehicle (AV) safety evaluation to demonstrate the proposed approach as a
diagnostic tool for the quality of the fitted input model.
| stat |
Rank-normalization, folding, and localization: An improved $\widehat{R}$
for assessing convergence of MCMC | Markov chain Monte Carlo is a key computational tool in Bayesian statistics,
but it can be challenging to monitor the convergence of an iterative stochastic
algorithm. In this paper we show that the convergence diagnostic $\widehat{R}$
of Gelman and Rubin (1992) has serious flaws. Traditional $\widehat{R}$ will
fail to correctly diagnose convergence failures when the chain has a heavy tail
or when the variance varies across the chains. In this paper we propose an
alternative rank-based diagnostic that fixes these problems. We also introduce
a collection of quantile-based local efficiency measures, along with a
practical approach for computing Monte Carlo error estimates for quantiles. We
suggest that common trace plots should be replaced with rank plots from
multiple chains. Finally, we give recommendations for how these methods should
be used in practice.
| stat |
Multivariate postprocessing methods for high-dimensional seasonal
weather forecasts | Seasonal weather forecasts are crucial for long-term planning in many
practical situations and skillful forecasts may have substantial economic and
humanitarian implications. Current seasonal forecasting models require
statistical postprocessing of the output to correct systematic biases and
unrealistic uncertainty assessments. We propose a multivariate postprocessing
approach utilizing covariance tapering, combined with a dimension reduction
step based on principal component analysis for efficient computation. Our
proposed technique can correctly and efficiently handle non-stationary,
non-isotropic and negatively correlated spatial error patterns, and is
applicable on a global scale. Further, a moving average approach to marginal
postprocessing is shown to flexibly handle trends in biases caused by global
warming, and short training periods. In an application to global sea surface
temperature forecasts issued by the Norwegian Climate Prediction Model
(NorCPM), our proposed methodology is shown to outperform known reference
methods.
| stat |
Higher Order Generalization Error for First Order Discretization of
Langevin Diffusion | We propose a novel approach to analyze generalization error for
discretizations of Langevin diffusion, such as the stochastic gradient Langevin
dynamics (SGLD). For an $\epsilon$ tolerance of expected generalization error,
it is known that a first order discretization can reach this target if we run
$\Omega(\epsilon^{-1} \log (\epsilon^{-1}) )$ iterations with
$\Omega(\epsilon^{-1})$ samples. In this article, we show that with additional
smoothness assumptions, even first order methods can achieve arbitrarily
runtime complexity. More precisely, for each $N>0$, we provide a sufficient
smoothness condition on the loss function such that a first order
discretization can reach $\epsilon$ expected generalization error given
$\Omega( \epsilon^{-1/N} \log (\epsilon^{-1}) )$ iterations with
$\Omega(\epsilon^{-1})$ samples.
| stat |
Uncertainty quantification for epidemiological forecasts of COVID-19
through combinations of model predictions | A common statistical problem is prediction, or forecasting, in the presence
of an ensemble of multiple candidate models. For example, multiple candidate
models may be available to predict case numbers in a disease epidemic,
resulting from different modelling approaches (e.g. mechanistic or empirical)
or differing assumptions about spatial or age mixing. Alternative models
capture genuine uncertainty in scientific understanding of disease dynamics,
and/or different simplifying assumptions underpinning each model derivation.
While the analysis of multi-model ensembles can be computationally challenging,
accounting for this 'structural uncertainty' can improve forecast accuracy and
reduce the risk of over-estimated confidence. In this paper we look at
combining epidemiological forecasts for COVID-19 daily deaths, hospital
admissions, and hospital and ICU occupancy, in order to improve the predictive
accuracy of the short term forecasts. We combining models via combinations of
individual predictive densities with weights chosen via application of
predictive scoring, as commonly applied in meteorological and economic
forecasting.
| stat |
$\mathcal{G}$-SGD: Optimizing ReLU Neural Networks in its Positively
Scale-Invariant Space | It is well known that neural networks with rectified linear units (ReLU)
activation functions are positively scale-invariant. Conventional algorithms
like stochastic gradient descent optimize the neural networks in the vector
space of weights, which is, however, not positively scale-invariant. This
mismatch may lead to problems during the optimization process. Then, a natural
question is: \emph{can we construct a new vector space that is positively
scale-invariant and sufficient to represent ReLU neural networks so as to
better facilitate the optimization process }? In this paper, we provide our
positive answer to this question. First, we conduct a formal study on the
positive scaling operators which forms a transformation group, denoted as
$\mathcal{G}$. We show that the value of a path (i.e. the product of the
weights along the path) in the neural network is invariant to positive scaling
and prove that the value vector of all the paths is sufficient to represent the
neural networks under mild conditions. Second, we show that one can identify
some basis paths out of all the paths and prove that the linear span of their
value vectors (denoted as $\mathcal{G}$-space) is an invariant space with lower
dimension under the positive scaling group. Finally, we design stochastic
gradient descent algorithm in $\mathcal{G}$-space (abbreviated as
$\mathcal{G}$-SGD) to optimize the value vector of the basis paths of neural
networks with little extra cost by leveraging back-propagation. Our experiments
show that $\mathcal{G}$-SGD significantly outperforms the conventional SGD
algorithm in optimizing ReLU networks on benchmark datasets.
| stat |
Doubly Robust Thompson Sampling for linear payoffs | A challenging aspect of the bandit problem is that a stochastic reward is
observed only for the chosen arm and the rewards of other arms remain missing.
Since the arm choice depends on the past context and reward pairs, the contexts
of chosen arms suffer from correlation and render the analysis difficult. We
propose a novel multi-armed contextual bandit algorithm called Doubly Robust
(DR) Thompson Sampling (TS) that applies the DR technique used in missing data
literature to TS. The proposed algorithm improves the bound of TS by a factor
of $\sqrt{d}$, where $d$ is the dimension of the context. A benefit of the
proposed method is that it uses all the context data, chosen or not chosen,
thus allowing to circumvent the technical definition of unsaturated arms used
in theoretical analysis of TS. Empirical studies show the advantage of the
proposed algorithm over TS.
| stat |
A generalized double robust Bayesian model averaging approach to causal
effect estimation with application to the Study of Osteoporotic Fractures | Analysts often use data-driven approaches to supplement their substantive
knowledge when selecting covariates for causal effect estimation. Multiple
variable selection procedures tailored for causal effect estimation have been
devised in recent years, but additional developments are still required to
adequately address the needs of data analysts. In this paper, we propose a
Generalized Bayesian Causal Effect Estimation (GBCEE) algorithm to perform
variable selection and produce double robust estimates of causal effects for
binary or continuous exposures and outcomes. GBCEE employs a prior distribution
that targets the selection of true confounders and predictors of the outcome
for the unbiased estimation of causal effects with reduced standard errors.
Double robust estimators provide some robustness against model
misspecification, whereas the Bayesian machinery allows GBCEE to directly
produce inferences for its estimate. GBCEE was compared to multiple
alternatives in various simulation scenarios and was observed to perform
similarly or to outperform double robust alternatives. Its ability to directly
produce inferences is also an important advantage from a computational
perspective. The method is finally illustrated for the estimation of the effect
of meeting physical activity recommendations on the risk of hip or upper-leg
fractures among elderly women in the Study of Osteoporotic Fractures. The 95%
confidence interval produced by GBCEE is 61% shorter than that of a double
robust estimator adjusting for all potential confounders in this illustration.
| stat |
Total Variation Regularization for Compartmental Epidemic Models with
Time-Varying Dynamics | Compartmental epidemic models are among the most popular ones in
epidemiology. For the parameters (e.g., the transmission rate) characterizing
these models, the majority of researchers simplify them as constants, while
some others manage to detect their continuous variations. In this paper, we aim
at capturing, on the other hand, discontinuous variations, which better
describe the impact of many noteworthy events, such as city lockdowns, the
opening of field hospitals, and the mutation of the virus, whose effect should
be instant. To achieve this, we balance the model's likelihood by total
variation, which regulates the temporal variations of the model parameters. To
infer these parameters, instead of using Monte Carlo methods, we design a novel
yet straightforward optimization algorithm, dubbed Iterated Nelder--Mead, which
repeatedly applies the Nelder--Mead algorithm. Experiments conducted on the
simulated data demonstrate that our approach can reproduce these
discontinuities and precisely depict the epidemics.
| stat |
Weston-Watkins Hinge Loss and Ordered Partitions | Multiclass extensions of the support vector machine (SVM) have been
formulated in a variety of ways. A recent empirical comparison of nine such
formulations [Do\v{g}an et al. 2016] recommends the variant proposed by Weston
and Watkins (WW), despite the fact that the WW-hinge loss is not calibrated
with respect to the 0-1 loss. In this work we introduce a novel discrete loss
function for multiclass classification, the ordered partition loss, and prove
that the WW-hinge loss is calibrated with respect to this loss. We also argue
that the ordered partition loss is maximally informative among discrete losses
satisfying this property. Finally, we apply our theory to justify the empirical
observation made by Do\v{g}an et al. that the WW-SVM can work well even under
massive label noise, a challenging setting for multiclass SVMs.
| stat |
Marginal false discovery rate control for likelihood-based penalized
regression models | The popularity of penalized regression in high-dimensional data analysis has
led to a demand for new inferential tools for these models. False discovery
rate control is widely used in high-dimensional hypothesis testing, but has
only recently been considered in the context of penalized regression. Almost
all of this work, however, has focused on lasso-penalized linear regression. In
this paper, we derive a general method for controlling the marginal false
discovery rate that can be applied to any penalized likelihood-based model,
such as logistic regression and Cox regression. Our approach is fast, flexible
and can be used with a variety of penalty functions including lasso, elastic
net, MCP, and MNet. We derive theoretical results under which the proposed
method is valid, and use simulation studies to demonstrate that the approach is
reasonably robust, albeit slightly conservative, when these assumptions are
violated. Despite being conservative, we show that our method often offers more
power to select causally important features than existing approaches. Finally,
the practical utility of the method is demonstrated on gene expression data
sets with binary and time-to-event outcomes.
| stat |
Bayesian Semiparametric Longitudinal Drift-Diffusion Mixed Models for
Tone Learning in Adults | Understanding how adult humans learn non-native speech categories such as
tone information has shed novel insights into the mechanisms underlying
experience-dependent brain plasticity. Scientists have traditionally examined
these questions using longitudinal learning experiments under a multi-category
decision making paradigm. Drift-diffusion processes are popular in such
contexts for their ability to mimic underlying neural mechanisms. Motivated by
these problems, we develop a novel Bayesian semiparametric inverse Gaussian
drift-diffusion mixed model for multi-alternative decision making in
longitudinal settings. We design a Markov chain Monte Carlo algorithm for
posterior computation. We evaluate the method's empirical performances through
synthetic experiments. Applied to our motivating longitudinal tone learning
study, the method provides novel insights into how the biologically
interpretable model parameters evolve with learning, differ between
input-response tone combinations, and differ between well and poorly performing
adults.
| stat |
Identifying the Influential Inputs for Network Output Variance Using
Sparse Polynomial Chaos Expansion | Sensitivity analysis (SA) is an important aspect of process automation. It
often aims to identify the process inputs that influence the process output's
variance significantly. Existing SA approaches typically consider the
input-output relationship as a black-box and conduct extensive random sampling
from the actual process or its high-fidelity simulation model to identify the
influential inputs. In this paper, an alternate, novel approach is proposed
using a sparse polynomial chaos expansion-based model for a class of
input-output relationships represented as directed acyclic networks. The model
exploits the relationship structure by recursively relating a network node to
its direct predecessors to trace the output variance back to the inputs. It,
thereby, estimates the Sobol indices, which measure the influence of each input
on the output variance, accurately and efficiently. Theoretical analysis
establishes the validity of the model as the prediction of the network output
converges in probability to the true output under certain regularity
conditions. Empirical evaluation on two manufacturing processes shows that the
model estimates the Sobol indices accurately with far fewer observations than a
state-of-the-art Monte Carlo sampling method.
| stat |
Bayesian EWMA and CUSUM Control Charts Under Different Loss Functions | The Exponentially Weighted Moving Average (EWMA) and Cumulative Sum (CUSUM)
control charts have been used in profile monitoring to track drift shifts that
occur in a monitored process. We construct Bayesian EWMA and Bayesian CUSUM
charts informed by posterior and posterior predictive distributions using
different loss functions, prior distributions, and likelihood distributions. A
simulation study is performed, and the performance of the charts are evaluated
via average run length (ARL), standard deviation of the run length (SDRL),
average time to signal (ATS), and standard deviation of time to signal (SDTS).
A sensitivity analysis is conducted using choices for the smoothing parameter,
out-of-control shift size, and hyper-parameters of the distribution. Based on
obtained results, we provide recommendations for use of the Bayesian EWMA and
Bayesian CUSUM control charts.
| stat |
Asynchronous Batch Bayesian Optimisation with Improved Local
Penalisation | Batch Bayesian optimisation (BO) has been successfully applied to
hyperparameter tuning using parallel computing, but it is wasteful of
resources: workers that complete jobs ahead of others are left idle. We address
this problem by developing an approach, Penalising Locally for Asynchronous
Bayesian Optimisation on $k$ workers (PLAyBOOK), for asynchronous parallel BO.
We demonstrate empirically the efficacy of PLAyBOOK and its variants on
synthetic tasks and a real-world problem. We undertake a comparison between
synchronous and asynchronous BO, and show that asynchronous BO often
outperforms synchronous batch BO in both wall-clock time and number of function
evaluations.
| stat |
An efficient methodology to estimate the parameters of a two-dimensional
chirp signal model | In various capacities of statistical signal processing two-dimensional (2-D)
chirp models have been considered significantly, particularly in image
processing$-$ to model gray-scale and texture images, magnetic resonance
imaging, optical imaging etc. In this paper we address the problem of
estimation of the unknown parameters of a 2-D chirp model under the assumption
that the errors are independently and identically distributed (i.i.d.). The key
attribute of the proposed estimation procedure is that it is computationally
more efficient than the least squares estimation method. Moreover, the proposed
estimators are observed to have the same asymptotic properties as the least
squares estimators, thus providing computational effectiveness without any
compromise on the efficiency of the estimators. We extend the propounded
estimation method to provide a sequential procedure to estimate the unknown
parameters of a 2-D chirp model with multiple components and under the
assumption of i.i.d. errors we study the large sample properties of these
sequential estimators. Simulation studies and a synthetic data analysis show
that the proposed estimators perform satisfactorily.
| stat |
A Framework for Adaptive MCMC Targeting Multimodal Distributions | We propose a new Monte Carlo method for sampling from multimodal
distributions. The idea of this technique is based on splitting the task into
two: finding the modes of a target distribution $\pi$ and sampling, given the
knowledge of the locations of the modes. The sampling algorithm relies on steps
of two types: local ones, preserving the mode; and jumps to regions associated
with different modes. Besides, the method learns the optimal parameters of the
algorithm while it runs, without requiring user intervention. Our technique
should be considered as a flexible framework, in which the design of moves can
follow various strategies known from the broad MCMC literature.
In order to design an adaptive scheme that facilitates both local and jump
moves, we introduce an auxiliary variable representing each mode and we define
a new target distribution $\tilde{\pi}$ on an augmented state space
$\mathcal{X}~\times~\mathcal{I}$, where $\mathcal{X}$ is the original state
space of $\pi$ and $\mathcal{I}$ is the set of the modes. As the algorithm runs
and updates its parameters, the target distribution $\tilde{\pi}$ also keeps
being modified. This motivates a new class of algorithms, Auxiliary Variable
Adaptive MCMC. We prove general ergodic results for the whole class before
specialising to the case of our algorithm.
| stat |
Robust Particle Filtering via Bayesian Nonparametric Outlier Modeling | This paper is concerned with the online estimation of a nonlinear dynamic
system from a series of noisy measurements. The focus is on cases wherein
outliers are present in-between normal noises. We assume that the outliers
follow an unknown generating mechanism which deviates from that of normal
noises, and then model the outliers using a Bayesian nonparametric model called
Dirichlet process mixture (DPM). A sequential particle-based algorithm is
derived for posterior inference for the outlier model as well as the state of
the system to be estimated. The resulting algorithm is termed DPM based robust
PF (DPM-RPF). The nonparametric feature makes this algorithm allow the data to
"speak for itself" to determine the complexity and structure of the outlier
model. Simulation results show that it performs remarkably better than two
state-of-the-art methods especially when outliers appear frequently along time.
| stat |
Can You Trust Your Model's Uncertainty? Evaluating Predictive
Uncertainty Under Dataset Shift | Modern machine learning methods including deep learning have achieved great
success in predictive accuracy for supervised learning tasks, but may still
fall short in giving useful estimates of their predictive {\em uncertainty}.
Quantifying uncertainty is especially critical in real-world settings, which
often involve input distributions that are shifted from the training
distribution due to a variety of factors including sample bias and
non-stationarity. In such settings, well calibrated uncertainty estimates
convey information about when a model's output should (or should not) be
trusted. Many probabilistic deep learning methods, including Bayesian-and
non-Bayesian methods, have been proposed in the literature for quantifying
predictive uncertainty, but to our knowledge there has not previously been a
rigorous large-scale empirical comparison of these methods under dataset shift.
We present a large-scale benchmark of existing state-of-the-art methods on
classification problems and investigate the effect of dataset shift on accuracy
and calibration. We find that traditional post-hoc calibration does indeed fall
short, as do several other previous methods. However, some methods that
marginalize over models give surprisingly strong results across a broad
spectrum of tasks.
| stat |
Exactly Computing the Local Lipschitz Constant of ReLU Networks | The local Lipschitz constant of a neural network is a useful metric with
applications in robustness, generalization, and fairness evaluation. We provide
novel analytic results relating the local Lipschitz constant of nonsmooth
vector-valued functions to a maximization over the norm of the generalized
Jacobian. We present a sufficient condition for which backpropagation always
returns an element of the generalized Jacobian, and reframe the problem over
this broad class of functions. We show strong inapproximability results for
estimating Lipschitz constants of ReLU networks, and then formulate an
algorithm to compute these quantities exactly. We leverage this algorithm to
evaluate the tightness of competing Lipschitz estimators and the effects of
regularized training on the Lipschitz constant.
| stat |
A method to identify geochemical mineralization on linear transect | Mineral exploration in biogeochemistry is related to the detection of
anomalies in soil, which is driven by many factors and thus a complex problem.
Mik\v{s}ov\'a, Rieser, and Filzmoser (2019) have introduced a method for the
identification of spatial patterns with increased element concentrations in
samples along a linear sampling transect. This procedure is based on fitting
Generalized Additive Models (GAMs) to the concentration data, and computing a
curvature measure from the pairwise log-ratios of these fits. The higher the
curvature, the more likely one or both elements of the pair indicate local
mineralization. This method is applied on two geochemical data sets which have
been collected specifically for the purpose of mineral exploration. The aim is
to test the technique for its ability to identify pathfinder elements to detect
mineralized zones, and to verify whether the method can indicate which sampling
material is best suited for this purpose.
Reference: Mik\v{s}ov\'a D., Rieser C., Filzmoser P. (2019). "Identification
of mineralization in geochemistry along a transect based on the spatial
curvature of log-ratios." arXiv, (1912.02867).
| stat |
High-dimensional, multiscale online changepoint detection | We introduce a new method for high-dimensional, online changepoint detection
in settings where a $p$-variate Gaussian data stream may undergo a change in
mean. The procedure works by performing likelihood ratio tests against simple
alternatives of different scales in each coordinate, and then aggregating test
statistics across scales and coordinates. The algorithm is online in the sense
that both its storage requirements and worst-case computational complexity per
new observation are independent of the number of previous observations; in
practice, it may even be significantly faster than this. We prove that the
patience, or average run length under the null, of our procedure is at least at
the desired nominal level, and provide guarantees on its response delay under
the alternative that depend on the sparsity of the vector of mean change.
Simulations confirm the practical effectiveness of our proposal, which is
implemented in the R package 'ocd', and we also demonstrate its utility on a
seismology data set.
| stat |
Efficient testing and effect size estimation for set-based genetic
association inference via semiparametric multilevel mixture modeling:
Application to a genome-wide association study of coronary artery disease | In genetic association studies, rare variants with extremely small allele
frequency play a crucial role in complex traits, and the set-based testing
methods that jointly assess the effects of groups of single nucleotide
polymorphisms (SNPs) were developed to improve powers for the association
tests. However, the powers of these tests are still severely limited due to the
extremely small allele frequency, and precise estimations for the effect sizes
of individual SNPs are substantially impossible. In this article, we provide an
efficient set-based inference framework that addresses the two important issues
simultaneously based on a Bayesian semiparametric multilevel mixture model. We
propose to use the multilevel hierarchical model that incorporate the
variations in set-specific effects and variant-specific effects, and to apply
the optimal discovery procedure (ODP) that achieves the largest overall power
in multiple significance testing. In addition, we provide Bayesian optimal
"set-based" estimator of the empirical distribution of effect sizes. Efficiency
of the proposed methods is demonstrated through application to a genome-wide
association study of coronary artery disease (CAD), and through simulation
studies. These results suggested there could be a lot of rare variants with
large effect sizes for CAD, and the number of significant sets detected by the
ODP was much greater than those by existing methods.
| stat |
A goodness-of-fit test for the functional linear model with functional
response | The Functional Linear Model with Functional Response (FLMFR) is one of the
most fundamental models to assess the relation between two functional random
variables. In this paper, we propose a novel goodness-of-fit test for the FLMFR
against a general, unspecified, alternative. The test statistic is formulated
in terms of a Cram\'er-von Mises norm over a doubly-projected empirical process
which, using geometrical arguments, yields an easy-to-compute weighted
quadratic norm. A resampling procedure calibrates the test through a wild
bootstrap on the residuals and the use of convenient computational procedures.
As a sideways contribution, and since the statistic requires a reliable
estimator of the FLMFR, we discuss and compare several regularized estimators,
providing a new one specifically convenient for our test. The finite sample
behavior of the test is illustrated via a simulation study. Also, the new
proposal is compared with previous significance tests. Two novel real datasets
illustrate the application of the new test.
| stat |
Data Generation in Low Sample Size Setting Using Manifold Sampling and a
Geometry-Aware VAE | While much efforts have been focused on improving Variational Autoencoders
through richer posterior and prior distributions, little interest was shown in
amending the way we generate the data. In this paper, we develop two non
\emph{prior-dependent} generation procedures based on the geometry of the
latent space seen as a Riemannian manifold. The first one consists in sampling
along geodesic paths which is a natural way to explore the latent space while
the second one consists in sampling from the inverse of the metric volume
element which is easier to use in practice. Both methods are then compared to
\emph{prior-based} methods on various data sets and appear well suited for a
limited data regime. Finally, the latter method is used to perform data
augmentation in a small sample size setting and is validated across various
standard and \emph{real-life} data sets. In particular, this scheme allows to
greatly improve classification results on the OASIS database where balanced
accuracy jumps from 80.7% for a classifier trained with the raw data to 89.1%
when trained only with the synthetic data generated by our method. Such results
were also observed on 4 standard data sets.
| stat |
Consistency issues in Gaussian Mixture Models reduction algorithms | In many contexts Gaussian Mixtures (GM) are used to approximate probability
distributions, possibly time-varying. In some applications the number of GM
components exponentially increases over time, and reduction procedures are
required to keep them reasonably limited. The GM reduction (GMR) problem can be
formulated by choosing different measures of the dissimilarity of GMs before
and after reduction, like the Kullback-Leibler Divergence (KLD) and the
Integral Squared Error (ISE). Since in no case the solution is obtained in
closed form, many approximate GMR algorithms have been proposed in the past
three decades, although none of them provides optimality guarantees. In this
work we discuss the importance of the choice of the dissimilarity measure and
the issue of consistency of all steps of a reduction algorithm with the chosen
measure. Indeed, most of the existing GMR algorithms are composed by several
steps which are not consistent with a unique measure, and for this reason may
produce reduced GMs far from optimality. In particular, the use of the KLD, of
the ISE and normalized ISE is discussed and compared in this perspective.
| stat |
Approximately Optimal Subset Selection for Statistical Design and
Modelling | We study the problem of optimal subset selection from a set of correlated
random variables. In particular, we consider the associated combinatorial
optimization problem of maximizing the determinant of a symmetric positive
definite matrix that characterizes the chosen subset. This problem arises in
many domains, such as experimental designs, regression modeling, and
environmental statistics. We establish an efficient polynomial-time algorithm
using Determinantal Point Process for approximating the optimal solution to the
problem. We demonstrate the advantages of our methods by presenting
computational results for both synthetic and real data sets.
| stat |
A Bootstrap Lasso + Partial Ridge Method to Construct Confidence
Intervals for Parameters in High-dimensional Sparse Linear Models | Constructing confidence intervals for the coefficients of high-dimensional
sparse linear models remains a challenge, mainly because of the complicated
limiting distributions of the widely used estimators, such as the lasso.
Several methods have been developed for constructing such intervals. Bootstrap
lasso+ols is notable for its technical simplicity, good interpretability, and
performance that is comparable with that of other more complicated methods.
However, bootstrap lasso+ols depends on the beta-min assumption, a theoretic
criterion that is often violated in practice. Thus, we introduce a new method,
called bootstrap lasso+partial ridge, to relax this assumption. Lasso+partial
ridge is a two-stage estimator. First, the lasso is used to select features.
Then, the partial ridge is used to refit the coefficients. Simulation results
show that bootstrap lasso+partial ridge outperforms bootstrap lasso+ols when
there exist small, but nonzero coefficients, a common situation that violates
the beta-min assumption. For such coefficients, the confidence intervals
constructed using bootstrap lasso+partial ridge have, on average, $50\%$ larger
coverage probabilities than those of bootstrap lasso+ols. Bootstrap
lasso+partial ridge also has, on average, $35\%$ shorter confidence interval
lengths than those of the de-sparsified lasso methods, regardless of whether
the linear models are misspecified. Additionally, we provide theoretical
guarantees for bootstrap lasso+partial ridge under appropriate conditions, and
implement it in the R package "HDCI."
| stat |
Bayesian Wavelet Shrinkage with Beta Priors | In wavelet shrinkage and thresholding, most of the standard techniques do not
consider information that wavelet coefficients might be bounded, although
information about bounded energy in signals can be readily available. To
address this, we present a Bayesian approach for shrinkage of bounded wavelet
coefficients in the context of non-parametric regression. We propose the use of
a zero-contaminated beta distribution with a support symmetric around zero as
the prior distribution for the location parameter in the wavelet domain in
models with additive gaussian errors. The hyperparameters of the proposed model
are closely related to the shrinkage level, which facilitates their elicitation
and interpretation. For signals with a low signal-to-noise ratio, the
associated Bayesian shrinkage rules provide significant improvement in
performance in simulation studies when compared with standard techniques.
| stat |
Nested sampling cross-checks using order statistics | Nested sampling (NS) is an invaluable tool in data analysis in modern
astrophysics, cosmology, gravitational wave astronomy and particle physics. We
identify a previously unused property of NS related to order statistics: the
insertion indexes of new live points into the existing live points should be
uniformly distributed. This observation enabled us to create a novel
cross-check of single NS runs. The tests can detect when an NS run failed to
sample new live points from the constrained prior and plateaus in the
likelihood function, which break an assumption of NS and thus leads to
unreliable results. We applied our cross-check to NS runs on toy functions with
known analytic results in 2 - 50 dimensions, showing that our approach can
detect problematic runs on a variety of likelihoods, settings and dimensions.
As an example of a realistic application, we cross-checked NS runs performed in
the context of cosmological model selection. Since the cross-check is simple,
we recommend that it become a mandatory test for every applicable NS run.
| stat |
Maximum Relevance and Minimum Redundancy Feature Selection Methods for a
Marketing Machine Learning Platform | In machine learning applications for online product offerings and marketing
strategies, there are often hundreds or thousands of features available to
build such models. Feature selection is one essential method in such
applications for multiple objectives: improving the prediction accuracy by
eliminating irrelevant features, accelerating the model training and prediction
speed, reducing the monitoring and maintenance workload for feature data
pipeline, and providing better model interpretation and diagnosis capability.
However, selecting an optimal feature subset from a large feature space is
considered as an NP-complete problem. The mRMR (Minimum Redundancy and Maximum
Relevance) feature selection framework solves this problem by selecting the
relevant features while controlling for the redundancy within the selected
features. This paper describes the approach to extend, evaluate, and implement
the mRMR feature selection methods for classification problem in a marketing
machine learning platform at Uber that automates creation and deployment of
targeting and personalization models at scale. This study first extends the
existing mRMR methods by introducing a non-linear feature redundancy measure
and a model-based feature relevance measure. Then an extensive empirical
evaluation is performed for eight different feature selection methods, using
one synthetic dataset and three real-world marketing datasets at Uber to cover
different use cases. Based on the empirical results, the selected mRMR method
is implemented in production for the marketing machine learning platform. A
description of the production implementation is provided and an online
experiment deployed through the platform is discussed.
| stat |
Finding your feet: A Gaussian process model for estimating the abilities
of batsmen in Test cricket | In the sport of cricket, player batting ability is traditionally measured
using the batting average. However, the batting average fails to measure both
short-term changes in ability that occur during an innings, and long-term
changes that occur between innings, due to the likes of age and experience in
various match conditions. We derive and fit a Bayesian parametric model that
employs a Gaussian process to measure and predict how the batting abilities of
cricket players vary and fluctuate over the course of entire playing careers.
The results allow us to better quantify and predict player batting ability,
compared with both traditional cricket statistics, such as the batting average,
and more complex models, such as the official International Cricket Council
ratings.
| stat |
Optimal reinsurance for risk over surplus ratios | Optimal reinsurance when Value at Risk and expected surplus is balanced
through their ratio is studied, and it is demonstrated how results for
risk-adjusted surplus can be utilized. Simplifications for large portfolios are
derived, and this large-portfolio study suggests a new condition on the
reinsurance pricing regime which is crucial for the results obtained. One or
two-layer contracts now become optimal for both risk-adjusted surplus and the
risk over expected surplus ratio, but there is no second layer when portfolios
are large or when reinsurance prices are below some threshold. Simple
approximations of the optimum portfolio are considered, and their degree of
degradation compared to the optimum is studied which leads to theoretical
degradation rates as the number of policies grows. The theory is supported by
numerical experiments which suggest that the shape of the claim severity
distributions may not be of primary importance when designing an optimal
reinsurance program. It is argued that the approach can be applied to
Conditional Value at Risk as well.
| stat |
Analysis of the AOK Lower Saxony hospitalisation records data (years
2008 -- 2015) | Multidrug-resistant Enterobacteriaceae (MDR-E) have become a major public
health threat in many European countries. While traditional infection control
strategies primarily target the containment of intra-hospital transmission,
there is growing evidence highlighting the importance of inter-hospital patient
traffic for the spread of MDR-E within healthcare systems.
Our aim is to propose a network model, which will reflect patient traffic in
various European healthcare systems and will thus provide the framework to
study systematically transmission dynamics of MDR-E and the effectiveness of
infection control strategies to contain their spread within and potentially
across healthcare systems. However, to do that first we need to analyse real
patients data and base on that propose network model reflecting the complexity
of the real hospital network connections and dynamics of patient transfers
between healthcare facilities.
| stat |
Best Practices in Scientific Computing | The world is becoming increasingly complex, both in terms of the rich sources
of data we have access to as well as in terms of the statistical and
computational methods we can use on those data. These factors create an
ever-increasing risk for errors in our code and sensitivity in our findings to
data preparation and execution of complex statistical and computing methods.
The consequences of coding and data mistakes can be substantial. Openness
(e.g., providing others with data code) and transparency (e.g., requiring that
data processing and code follow standards) are two key solutions to help
alleviate concerns about replicability and errors. In this paper, we describe
the key steps for implementing a code quality assurance (QA) process for
researchers to follow to improve their coding practices throughout a project to
assure the quality of the final data, code, analyses and ultimately the
results. These steps include: (i) adherence to principles for code writing and
style that follow best practices, (ii) clear written documentation that
describes code, workflow and key analytic decisions; (iii) careful version
control, (iv) good data management; and (iv) regular testing and review.
Following all these steps will greatly improve the ability of a study to assure
results are accurate and reproducible. The responsibility for code QA falls not
only on individual researchers but institutions, journals, and funding agencies
as well.
| stat |
Bayesian Set of Best Dynamic Treatment Regimes and Sample Size
Determination for SMARTs with Binary Outcomes | One of the main goals of sequential, multiple assignment, randomized trials
(SMART) is to find the most efficacious design embedded dynamic treatment
regimes. The analysis method known as multiple comparisons with the best (MCB)
allows comparison between dynamic treatment regimes and identification of a set
of optimal regimes in the frequentist setting for continuous outcomes, thereby,
directly addressing the main goal of a SMART. In this paper, we develop a
Bayesian generalization to MCB for SMARTs with binary outcomes. Furthermore, we
show how to choose the sample size so that the inferior embedded DTRs are
screened out with a specified power. We compare log-odds between different DTRs
using their exact distribution without relying on asymptotic normality in
either the analysis or the power calculation. We conduct extensive simulation
studies under two SMART designs and illustrate our method's application to the
Adaptive Treatment for Alcohol and Cocaine Dependence (ENGAGE) trial.
| stat |
John's Walk | We present an affine-invariant random walk for drawing uniform random samples
from a convex body $\mathcal{K} \subset \mathbb{R}^n$ that uses maximum volume
inscribed ellipsoids, known as John's ellipsoids, for the proposal
distribution. Our algorithm makes steps using uniform sampling from the John's
ellipsoid of the symmetrization of $\mathcal{K}$ at the current point. We show
that from a warm start, the random walk mixes in $\widetilde{O}(n^7)$ steps
where the log factors depend only on constants associated with the warm start
and desired total variation distance to uniformity. We also prove polynomial
mixing bounds starting from any fixed point $x$ such that for any chord $pq$ of
$\mathcal{K}$ containing $x$, $\left|\log \frac{|p-x|}{|q-x|}\right|$ is
bounded above by a polynomial in $n$.
| stat |
Combining independent p-values in replicability analysis: A comparative
study | Given a family of null hypotheses $H_{1},\ldots,H_{s}$, we are interested in
the hypothesis $H_{s}^{\gamma}$ that at most $\gamma-1$ of these null
hypotheses are false. Assuming that the corresponding $p$-values are
independent, we are investigating combined $p$-values that are valid for
testing $H_{s}^{\gamma}$. In various settings in which $H_{s}^{\gamma}$ is
false, we determine which combined $p$-value works well in which setting. Via
simulations, we find that the Stouffer method works well if the null $p$-values
are uniformly distributed and the signal strength is low, and the Fisher method
works better if the null $p$-values are conservative, i.e. stochastically
larger than the uniform distribution. The minimum method works well if the
evidence for the rejection of $H_{s}^{\gamma}$ is focused on only a few
non-null $p$-values, especially if the null $p$-values are conservative.
Methods that incorporate the combination of $e$-values work well if the null
hypotheses $H_{1},\ldots,H_{s}$ are simple.
| stat |
A Simple Convergence Proof of Adam and Adagrad | We provide a simple proof of convergence covering both the Adam and Adagrad
adaptive optimization algorithms when applied to smooth (possibly non-convex)
objective functions with bounded gradients. We show that in expectation, the
squared norm of the objective gradient averaged over the trajectory has an
upper-bound which is explicit in the constants of the problem, parameters of
the optimizer and the total number of iterations $N$. This bound can be made
arbitrarily small: Adam with a learning rate $\alpha=1/\sqrt{N}$ and a momentum
parameter on squared gradients $\beta_2=1-1/N$ achieves the same rate of
convergence $O(\ln(N)/\sqrt{N})$ as Adagrad. Finally, we obtain the tightest
dependency on the heavy ball momentum among all previous convergence bounds for
non-convex Adam and Adagrad, improving from $O((1-\beta_1)^{-3})$ to
$O((1-\beta_1)^{-1})$. Our technique also improves the best known dependency
for standard SGD by a factor $1 - \beta_1$.
| stat |
Bootstrap Based Inference for Sparse High-Dimensional Time Series Models | Fitting sparse models to high-dimensional time series is an important area of
statistical inference. In this paper we consider sparse vector autoregressive
models and develop appropriate bootstrap methods to infer properties of such
processes. Our bootstrap methodology generates pseudo time series using a
model-based bootstrap procedure which involves an estimated, sparsified version
of the underlying vector autoregressive model. Inference is performed using
so-called de-sparsified or de-biased estimators of the autoregressive model
parameters. We derive the asymptotic distribution of such estimators in the
time series context and establish asymptotic validity of the bootstrap
procedure proposed for estimation and, appropriately modified, for testing
purposes. In particular we focus on testing that large groups of autoregressive
coefficients equal zero. Our theoretical results are complemented by
simulations which investigate the finite sample performance of the bootstrap
methodology proposed. A real-life data application is also presented.
| stat |
Improving forecasting with sub-seasonal time series patterns | Time series forecasting plays an increasingly important role in modern
business decisions. In today's data-rich environment, people often aim to
choose the optimal forecasting model for their data. However, identifying the
optimal model often requires professional knowledge and experience, making
accurate forecasting a challenging task. To mitigate the importance of model
selection, we propose a simple and reliable algorithm and successfully improve
the forecasting performance. Specifically, we construct multiple time series
with different sub-seasons from the original time series. These derived series
highlight different sub-seasonal patterns of the original series, making it
possible for the forecasting methods to capture diverse patterns and components
of the data. Subsequently, we make forecasts for these multiple series
separately with classical statistical models (ETS or ARIMA). Finally, the
forecasts of these multiple series are combined with equal weights. We evaluate
our approach on the widely-used forecasting competition datasets (M1, M3, and
M4), in terms of both point forecasts and prediction intervals. We observe
improvements in performance compared with the benchmarks. Our approach is
particularly suitable and robust for the datasets with higher frequencies. To
demonstrate the practical value of our proposition, we showcase the performance
improvements from our approach on hourly load data.
| stat |
A weighted Discrepancy Bound of quasi-Monte Carlo Importance Sampling | Importance sampling Monte-Carlo methods are widely used for the approximation
of expectations with respect to partially known probability measures. In this
paper we study a deterministic version of such an estimator based on
quasi-Monte Carlo. We obtain an explicit error bound in terms of the
star-discrepancy for this method.
| stat |
Causal learning with sufficient statistics: an information bottleneck
approach | The inference of causal relationships using observational data from partially
observed multivariate systems with hidden variables is a fundamental question
in many scientific domains. Methods extracting causal information from
conditional independencies between variables of a system are common tools for
this purpose, but are limited in the lack of independencies. To surmount this
limitation, we capitalize on the fact that the laws governing the generative
mechanisms of a system often result in substructures embodied in the generative
functional equation of a variable, which act as sufficient statistics for the
influence that other variables have on it. These functional sufficient
statistics constitute intermediate hidden variables providing new conditional
independencies to be tested. We propose to use the Information Bottleneck
method, a technique commonly applied for dimensionality reduction, to find
underlying sufficient sets of statistics. Using these statistics we formulate
new additional rules of causal orientation that provide causal information not
obtainable from standard structure learning algorithms, which exploit only
conditional independencies between observable variables. We validate the use of
sufficient statistics for structure learning both with simulated systems built
to contain specific sufficient statistics and with benchmark data from
regulatory rules previously and independently proposed to model biological
signal transduction networks.
| stat |
Cultural evolution in Vietnam's early 20th century: a Bayesian networks
analysis of Franco-Chinese house designs | The study of cultural evolution has taken on an increasingly
interdisciplinary and diverse approach in explicating phenomena of cultural
transmission and adoptions. Inspired by this computational movement, this study
uses Bayesian networks analysis, combining both the frequentist and the
Hamiltonian Markov chain Monte Carlo (MCMC) approach, to investigate the highly
representative elements in the cultural evolution of a Vietnamese city's
architecture in the early 20th century. With a focus on the fa\c{c}ade design
of 68 old houses in Hanoi's Old Quarter (based on 78 data lines extracted from
248 photos), the study argues that it is plausible to look at the aesthetics,
architecture and designs of the house fa\c{c}ade to find traces of cultural
evolution in Vietnam, which went through more than six decades of French
colonization and centuries of sociocultural influence from China. The in-depth
technical analysis, though refuting the presumed model on the probabilistic
dependency among the variables, yields several results, the most notable of
which is the strong influence of Buddhism over the decorations of the house
fa\c{c}ade. Particularly, in the top 5 networks with the best Bayesian
Information criterion (BIC) scores and p<0.05, the variable for decorations
(DC) always has a direct probabilistic dependency on the variable B for
Buddhism. The paper then checks the robustness of these models using
Hamiltonian MCMC method and find the posterior distributions of the models'
coefficients all satisfy the technical requirement. Finally, this study
suggests integrating Bayesian statistics in social sciences in general and for
study of cultural evolution and architectural transformation in particular.
| stat |
Causal variance decompositions for institutional comparisons in
healthcare | There is increasing interest in comparing institutions delivering healthcare
in terms of disease-specific quality indicators (QIs) that capture processes or
outcomes showing variations in the care provided. Such comparisons can be
framed in terms of causal models, where adjusting for patient case-mix is
analogous to controlling for confounding, and exposure is being treated in a
given hospital, for instance. Our goal here is to help identifying good QIs
rather than comparing hospitals in terms of an already chosen QI, and so we
focus on the presence and magnitude of overall variation in care between the
hospitals rather than the pairwise differences between any two hospitals. We
consider how the observed variation in care received at patient level can be
decomposed into that causally explained by the hospital performance adjusting
for the case-mix, the case-mix itself, and residual variation. For this
purpose, we derive a three-way variance decomposition, with particular
attention to its causal interpretation in terms of potential outcome variables.
We propose model-based estimators for the decomposition, accommodating
different link functions and either fixed or random effect models. We evaluate
their performance in a simulation study and demonstrate their use in a real
data application.
| stat |
Design Principles for Data Analysis | The data science revolution has led to an increased interest in the practice
of data analysis. While much has been written about statistical thinking, a
complementary form of thinking that appears in the practice of data analysis is
design thinking -- the problem-solving process to understand the people for
whom a product is being designed. For a given problem, there can be significant
or subtle differences in how a data analyst (or producer of a data analysis)
constructs, creates, or designs a data analysis, including differences in the
choice of methods, tooling, and workflow. These choices can affect the data
analysis products themselves and the experience of the consumer of the data
analysis. Therefore, the role of a producer can be thought of as designing the
data analysis with a set of design principles. Here, we introduce design
principles for data analysis and describe how they can be mapped to data
analyses in a quantitative, objective and informative manner. We also provide
empirical evidence of variation of principles within and between both producers
and consumers of data analyses. Our work leads to two insights: it suggests a
formal mechanism to describe data analyses based on the design principles for
data analysis, and it provides a framework to teach students how to build data
analyses using formal design principles.
| stat |
CLAIMED: A CLAssification-Incorporated Minimum Energy Design to explore
a multivariate response surface with feasibility constraints | Motivated by the problem of optimization of force-field systems in physics
using large-scale computer simulations, we consider exploration of a
deterministic complex multivariate response surface. The objective is to find
input combinations that generate output close to some desired or "target"
vector. In spite of reducing the problem to exploration of the input space with
respect to a one-dimensional loss function, the search is nontrivial and
challenging due to infeasible input combinations, high dimensionalities of the
input and output space and multiple "desirable" regions in the input space and
the difficulty of emulating the objective function well with a surrogate model.
We propose an approach that is based on combining machine learning techniques
with smart experimental design ideas to locate multiple good regions in the
input space.
| stat |
Computation of the expected value of a function of a chi-distributed
random variable | We consider the problem of numerically evaluating the expected value of a
smooth bounded function of a chi-distributed random variable, divided by the
square root of the number of degrees of freedom. This problem arises in the
contexts of simultaneous inference, the selection and ranking of populations
and in the evaluation of multivariate t probabilities. It also arises in the
assessment of the coverage probability and expected volume properties of the
some non-standard confidence regions. We use a transformation put forward by
Mori, followed by the application of the trapezoidal rule. This rule has the
remarkable property that, for suitable integrands, it is exponentially
convergent. We use it to create a nested sequence of quadrature rules, for the
estimation of the approximation error, so that previous evaluations of the
integrand are not wasted. The application of the trapezoidal rule requires the
approximation of an infinite sum by a finite sum. We provide a new easily
computed upper bound on the error of this approximation. Our overall conclusion
is that this method is a very suitable candidate for the computation of the
coverage and expected volume properties of non-standard confidence regions.
| stat |
Bayesian Approximations to Hidden Semi-Markov Models | We propose a Bayesian hidden Markov model for analyzing time series and
sequential data where a special structure of the transition probability matrix
is embedded to model explicit-duration semi-Markovian dynamics. Our formulation
allows for the development of highly flexible and interpretable models that can
integrate available prior information on state durations while keeping a
moderate computational cost to perform efficient posterior inference. We show
the benefits of choosing a Bayesian approach over its frequentist counterpart,
in terms of incorporation of prior information, quantification of uncertainty,
model selection and out-of-sample forecasting. The use of our methodology is
illustrated in an application relevant to e-Health, where we investigate
rest-activity rhythms using telemetric activity data collected via a wearable
sensing device.
| stat |