title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Trees, Forests, Chickens, and Eggs: When and Why to Prune Trees in a
Random Forest | Due to their long-standing reputation as excellent off-the-shelf predictors,
random forests continue remain a go-to model of choice for applied
statisticians and data scientists. Despite their widespread use, however, until
recently, little was known about their inner-workings and about which aspects
of the procedure were driving their success. Very recently, two competing
hypotheses have emerged -- one based on interpolation and the other based on
regularization. This work argues in favor of the latter by utilizing the
regularization framework to reexamine the decades-old question of whether
individual trees in an ensemble ought to be pruned. Despite the fact that
default constructions of random forests use near full depth trees in most
popular software packages, here we provide strong evidence that tree depth
should be seen as a natural form of regularization across the entire procedure.
In particular, our work suggests that random forests with shallow trees are
advantageous when the signal-to-noise ratio in the data is low. In building up
this argument, we also critique the newly popular notion of "double descent" in
random forests by drawing parallels to U-statistics and arguing that the
noticeable jumps in random forest accuracy are the result of simple averaging
rather than interpolation.
| stat |
Data Generators for Learning Systems Based on RBF Networks | There are plenty of problems where the data available is scarce and
expensive. We propose a generator of semi-artificial data with similar
properties to the original data which enables development and testing of
different data mining algorithms and optimization of their parameters. The
generated data allow a large scale experimentation and simulations without
danger of overfitting. The proposed generator is based on RBF networks, which
learn sets of Gaussian kernels. These Gaussian kernels can be used in a
generative mode to generate new data from the same distributions. To assess
quality of the generated data we evaluated the statistical properties of the
generated data, structural similarity and predictive similarity using
supervised and unsupervised learning techniques. To determine usability of the
proposed generator we conducted a large scale evaluation using 51 UCI data
sets. The results show a considerable similarity between the original and
generated data and indicate that the method can be useful in several
development and simulation scenarios. We analyze possible improvements in
classification performance by adding different amounts of generated data to the
training set, performance on high dimensional data sets, and conditions when
the proposed approach is successful.
| stat |
Improving Graduation Rate Estimates Using Regularly Updating Multi-Level
Absorbing Markov Chains | American universities use a procedure based on a rolling six-year graduation
rate to calculate statistics regarding their students' final educational
outcomes (graduating or not graduating). As~an alternative to the six-year
graduation rate method, many studies have applied absorbing Markov chains for
estimating graduation rates. In both cases, a frequentist approach is used.
For~the standard six-year graduation rate method, the frequentist approach
corresponds to counting the number of students who finished their program
within six years and dividing by the number of students who entered that year.
In the case of absorbing Markov chains, the frequentist approach is used to
compute the underlying transition matrix, which is then used to estimate the
graduation rate. In this paper, we apply a sensitivity analysis to compare the
performance of the standard six-year graduation rate method with that of
absorbing Markov chains. Through the analysis, we highlight significant
limitations with regards to the estimation accuracy of both approaches when
applied to small sample sizes or cohorts at a university. Additionally, we note
that the Absorbing Markov chain method introduces a significant bias, which
leads to an underestimation of the true graduation rate. To~overcome both these
challenges, we propose and evaluate the use of a regularly updating multi-level
absorbing Markov chain (RUML-AMC) in which the transition matrix is updated
year to year. We empirically demonstrate that the proposed RUML-AMC approach
nearly eliminates estimation bias while reducing the estimation variation by
more than 40%, especially for populations with small sample sizes.
| stat |
Boosting Random Forests to Reduce Bias; One-Step Boosted Forest and its
Variance Estimate | In this paper we propose using the principle of boosting to reduce the bias
of a random forest prediction in the regression setting. From the original
random forest fit we extract the residuals and then fit another random forest
to these residuals. We call the sum of these two random forests a
\textit{one-step boosted forest}. We show with simulated and real data that the
one-step boosted forest has a reduced bias compared to the original random
forest. The paper also provides a variance estimate of the one-step boosted
forest by an extension of the infinitesimal Jackknife estimator. Using this
variance estimate we can construct prediction intervals for the boosted forest
and we show that they have good coverage probabilities. Combining the bias
reduction and the variance estimate we show that the one-step boosted forest
has a significant reduction in predictive mean squared error and thus an
improvement in predictive performance. When applied on datasets from the UCI
database, one-step boosted forest performs better than random forest and
gradient boosting machine algorithms. Theoretically we can also extend such a
boosting process to more than one step and the same principles outlined in this
paper can be used to find variance estimates for such predictors. Such boosting
will reduce bias even further but it risks over-fitting and also increases the
computational burden.
| stat |
Measurement errors in the binary instrumental variable model | Instrumental variable methods can identify causal effects even when the
treatment and outcome are confounded. We study the problem of imperfect
measurements of the binary instrumental variable, treatment or outcome. We
first consider non-differential measurement errors, that is, the mis-measured
variable does not depend on other variables given its true value. We show that
the measurement error of the instrumental variable does not bias the estimate,
the measurement error of the treatment biases the estimate away from zero, and
the measurement error of the outcome biases the estimate toward zero. Moreover,
we derive sharp bounds on the causal effects without additional assumptions.
These bounds are informative because they exclude zero. We then consider
differential measurement errors, and focus on sensitivity analyses in those
settings.
| stat |
Estimating an Extreme Bayesian Network via Scalings | Recursive max-linear vectors model causal dependence between its components
by expressing each node variable as a max-linear function of its parental nodes
in a directed acyclic graph and some exogenous innovation. Motivated by extreme
value theory, innovations are assumed to have regularly varying distribution
tails. We propose a scaling technique in order to determine a causal order of
the node variables. All dependence parameters are then estimated from the
estimated scalings. Furthermore, we prove asymptotic normality of the estimated
scalings and dependence parameters based on asymptotic normality of the
empirical spectral measure. Finally, we apply our structure learning and
estimation algorithm to financial data and food dietary interview data.
| stat |
Sparsistent filtering of comovement networks from high-dimensional data | Network filtering is an important form of dimension reduction to isolate the
core constituents of large and interconnected complex systems. We introduce a
new technique to filter large dimensional networks arising out of dynamical
behavior of the constituent nodes, exploiting their spectral properties. As
opposed to the well known network filters that rely on preserving key
topological properties of the realized network, our method treats the spectrum
as the fundamental object and preserves spectral properties. Applying
asymptotic theory for high dimensional data for the filter, we show that it can
be tuned to interpolate between zero filtering to maximal filtering that
induces sparsity and consistency while having the least spectral distance from
a linear shrinkage estimator. We apply our proposed filter to covariance
networks constructed from financial data, to extract the key subnetwork
embedded in the full sample network.
| stat |
A-ComVar: A Flexible Extension of Common Variance Design | We consider nonregular fractions of factorial experiments for a class of
linear models. These models have a common general mean and main effects,
however they may have different 2-factor interactions. Here we assume for
simplicity that 3-factor and higher order interactions are negligible. In the
absence of a priori knowledge about which interactions are important, it is
reasonable to prefer a design that results in equal variance for the estimates
of all interaction effects to aid in model discrimination. Such designs are
called common variance designs and can be quite challenging to identify without
performing an exhaustive search of possible designs. In this work, we introduce
an extension of common variance designs called approximate common variance, or
A-ComVar designs. We develop a numerical approach to finding A-ComVar designs
that is much more efficient than an exhaustive search. We present the types of
A-ComVar designs that can be found for different number of factors, runs, and
interactions. We further demonstrate the competitive performance of both common
variance and A-ComVar designs using several comparisons to other popular
designs in the literature.
| stat |
Iterative Construction of Gaussian Process Surrogate Models for Bayesian
Inference | A new algorithm is developed to tackle the issue of sampling non-Gaussian
model parameter posterior probability distributions that arise from solutions
to Bayesian inverse problems. The algorithm aims to mitigate some of the
hurdles faced by traditional Markov Chain Monte Carlo (MCMC) samplers, through
constructing proposal probability densities that are both, easy to sample and
that provide a better approximation to the target density than a simple
Gaussian proposal distribution would. To achieve that, a Gaussian proposal
distribution is augmented with a Gaussian Process (GP) surface that helps
capture non-linearities in the log-likelihood function. In order to train the
GP surface, an iterative approach is adopted for the optimal selection of
points in parameter space. Optimality is sought by maximizing the information
gain of the GP surface using a minimum number of forward model simulation runs.
The accuracy of the GP-augmented surface approximation is assessed in two ways.
The first consists of comparing predictions obtained from the approximate
surface with those obtained through running the actual simulation model at
hold-out points in parameter space. The second consists of a measure based on
the relative variance of sample weights obtained from sampling the approximate
posterior probability distribution of the model parameters. The efficacy of
this new algorithm is tested on inferring reaction rate parameters in a 3-node
and 6-node network toy problems, which imitate idealized reaction networks in
combustion applications.
| stat |
Stress Testing Network Reconstruction via Graphical Causal Model | An resilience optimal evaluation of financial portfolios implies having
plausible hypotheses about the multiple interconnections between the
macroeconomic variables and the risk parameters. In this paper, we propose a
graphical model for the reconstruction of the causal structure that links the
multiple macroeconomic variables and the assessed risk parameters, it is this
structure that we call Stress Testing Network (STN). In this model, the
relationships between the macroeconomic variables and the risk parameter define
a "relational graph" among their time-series, where related time-series are
connected by an edge. Our proposal is based on the temporal causal models, but
unlike, we incorporate specific conditions in the structure which correspond to
intrinsic characteristics this type of networks. Using the proposed model and
given the high-dimensional nature of the problem, we used regularization
methods to efficiently detect causality in the time-series and reconstruct the
underlying causal structure. In addition, we illustrate the use of model in
credit risk data of a portfolio. Finally, we discuss its uses and practical
benefits in stress testing.
| stat |
Likelihood Ratios for Out-of-Distribution Detection | Discriminative neural networks offer little or no performance guarantees when
deployed on data not generated by the same process as the training
distribution. On such out-of-distribution (OOD) inputs, the prediction may not
only be erroneous, but confidently so, limiting the safe deployment of
classifiers in real-world applications. One such challenging application is
bacteria identification based on genomic sequences, which holds the promise of
early detection of diseases, but requires a model that can output low
confidence predictions on OOD genomic sequences from new bacteria that were not
present in the training data. We introduce a genomics dataset for OOD detection
that allows other researchers to benchmark progress on this important problem.
We investigate deep generative model based approaches for OOD detection and
observe that the likelihood score is heavily affected by population level
background statistics. We propose a likelihood ratio method for deep generative
models which effectively corrects for these confounding background statistics.
We benchmark the OOD detection performance of the proposed method against
existing approaches on the genomics dataset and show that our method achieves
state-of-the-art performance. We demonstrate the generality of the proposed
method by showing that it significantly improves OOD detection when applied to
deep generative models of images.
| stat |
A U-statistic-based test of treatment effect heterogeneity | Many studies include a goal of determining whether there is treatment effect
heterogeneity across different subpopulations. In this paper, we propose a
U-statistic-based non-parametric test of the null hypothesis that the treatment
effects are identical in different subgroups. The proposed test provides more
power than the standard parametric test when the underlying distribution
assumptions of the latter are violated. We apply the method to data from an
economic study of program effectiveness and find that there is treatment effect
heterogeneity in different subpopulations.
| stat |
On the Approximation Properties of Random ReLU Features | We study the approximation properties of random ReLU features through their
reproducing kernel Hilbert space (RKHS). We first prove a universality theorem
for the RKHS induced by random features whose feature maps are of the form of
nodes in neural networks. The universality result implies that the random ReLU
features method is a universally consistent learning algorithm. We prove that
despite the universality of the RKHS induced by the random ReLU features,
composition of functions in it generates substantially more complicated
functions that are harder to approximate than those functions simply in the
RKHS. We also prove that such composite functions can be efficiently
approximated by multi-layer ReLU networks with bounded weights. This depth
separation result shows that the random ReLU features models suffer from the
same weakness as that of shallow models. We show in experiments that the
performance of random ReLU features is comparable to that of random Fourier
features and, in general, has a lower computational cost. We also demonstrate
that when the target function is the composite function as described in the
depth separation theorem, 3-layer neural networks indeed outperform both random
ReLU features and 2-layer neural networks.
| stat |
Applying Bayesian Hierarchical Probit Model to Interview Grade
Evaluation | Job interviews are a fundamental activity for most corporations to acquire
potential candidates, and for job seekers to get well-rewarded and fulfilling
career opportunities. In many cases, interviews are conducted in multiple
processes such as telephone interviews and several face-to-face interviews. At
each stage, candidates are evaluated in various aspects. Among them, grade
evaluation, such as a rating on a 1-4 scale, might be used as a reasonable
method to evaluate candidates. However, because each evaluation is based on a
subjective judgment of interviewers, the aggregated evaluations can be biased
because the difference in toughness of interviewers is not examined.
Additionally, it is noteworthy that the toughness of interviewers might vary
depending on the interview round. As described herein, we propose an analytical
framework of simultaneous estimation for both the true potential of candidates
and toughness of interviewers' judgment considering job interview rounds, with
algorithms to extract unseen knowledge of the true potential of candidates and
toughness of interviewers as latent variables through analyzing grade data of
job interviews. We apply a Bayesian Hierarchical Ordered Probit Model to the
grade data from HRMOS, a cloud-based Applicant Tracking System (ATS) operated
by BizReach, Inc., an IT start-up particularly addressing human-resource needs
in Japan. Our model successfully quantifies the candidate potential and the
interviewers' toughness. An interpretation and applications of the model are
given along with a discussion of its place within hiring processes in
real-world settings. The parameters are estimated by Markov Chain Monte Carlo
(MCMC). A discussion of uncertainty, which is given by the posterior
distribution of the parameters, is also provided along with the analysis.
| stat |
Handling time-dependent exposures and confounders when estimating
attributable fractions -- bridging the gap between multistate and
counterfactual modeling | The population-attributable fraction (PAF) expresses the percentage of events
that could have been prevented by eradicating a certain exposure in a certain
population. It can be strongly time-dependent because either exposure incidence
or excess risk may change over time. Competing events may moreover hinder the
outcome of interest from being observed. Occurrence of either of these events
may, in turn, prevent the exposure of interest. Estimation approaches thus need
to carefully account for the timing of potential events in such highly dynamic
settings. The use of multistate models (MSMs) has been widely encouraged to
meet this need so as to eliminate preventable yet common types of bias, such as
immortal time bias. However, certain MSM based proposals for PAF estimation
fail to fully eliminate such biases. In addition, assessing whether patients
die from rather than with a certain exposure, not only requires adequate
modeling of the timing of events, but also of the confounding factors affecting
these events and their timing. While proposed MSM approaches for confounding
adjustment may be sufficient to accommodate imbalances between infected and
uninfected patients present at baseline, these proposals generally fail to
adequately tackle time-dependent confounding. For this, a class of generalized
methods (g-methods) which includes inverse probability (IP) weighting can be
used. Because the connection between MSMs and g-methods is not readily
apparent, we here provide a detailed mapping between MSM and IP of censoring
weighting approaches for estimating PAFs. In particular, we illustrate that the
connection between these two approaches can be made more apparent by means of a
weighting-based characterization of MSM approaches that aids to both pinpoint
current shortcomings of MSM based proposals and to enhance intuition into
simple modifications to overcome these limitations.
| stat |
The Gaussian Process Autoregressive Regression Model (GPAR) | Multi-output regression models must exploit dependencies between outputs to
maximise predictive performance. The application of Gaussian processes (GPs) to
this setting typically yields models that are computationally demanding and
have limited representational power. We present the Gaussian Process
Autoregressive Regression (GPAR) model, a scalable multi-output GP model that
is able to capture nonlinear, possibly input-varying, dependencies between
outputs in a simple and tractable way: the product rule is used to decompose
the joint distribution over the outputs into a set of conditionals, each of
which is modelled by a standard GP. GPAR's efficacy is demonstrated on a
variety of synthetic and real-world problems, outperforming existing GP models
and achieving state-of-the-art performance on established benchmarks.
| stat |
Beyond the Mean-Field: Structured Deep Gaussian Processes Improve the
Predictive Uncertainties | Deep Gaussian Processes learn probabilistic data representations for
supervised learning by cascading multiple Gaussian Processes. While this model
family promises flexible predictive distributions, exact inference is not
tractable. Approximate inference techniques trade off the ability to closely
resemble the posterior distribution against speed of convergence and
computational efficiency. We propose a novel Gaussian variational family that
allows for retaining covariances between latent processes while achieving fast
convergence by marginalising out all global latent variables. After providing a
proof of how this marginalisation can be done for general covariances, we
restrict them to the ones we empirically found to be most important in order to
also achieve computational efficiency. We provide an efficient implementation
of our new approach and apply it to several benchmark datasets. It yields
excellent results and strikes a better balance between accuracy and calibrated
uncertainty estimates than its state-of-the-art alternatives.
| stat |
Simple conditions for convergence of sequential Monte Carlo genealogies
with applications | We present simple conditions under which the limiting genealogical process
associated with a class of interacting particle systems with non-neutral
selection mechanisms, as the number of particles grows, is a time-rescaled
Kingman coalescent. Sequential Monte Carlo algorithms are popular methods for
approximating integrals in problems such as non-linear filtering and smoothing
which employ this type of particle system. Their performance depends strongly
on the properties of the induced genealogical process. We verify the conditions
of our main result for standard sequential Monte Carlo algorithms with a broad
class of low-variance resampling schemes, as well as for conditional sequential
Monte Carlo with multinomial resampling.
| stat |
Importance sampling type estimators based on approximate marginal MCMC | We consider importance sampling (IS) type weighted estimators based on Markov
chain Monte Carlo (MCMC) targeting an approximate marginal of the target
distribution. In the context of Bayesian latent variable models, the MCMC
typically operates on the hyperparameters, and the subsequent weighting may be
based on IS or sequential Monte Carlo (SMC), but allows for multilevel
techniques as well. The IS approach provides a natural alternative to delayed
acceptance (DA) pseudo-marginal/particle MCMC, and has many advantages over DA,
including a straightforward parallelisation and additional flexibility in MCMC
implementation. We detail minimal conditions which ensure strong consistency of
the suggested estimators, and provide central limit theorems with expressions
for asymptotic variances. We demonstrate how our method can make use of SMC in
the state space models context, using Laplace approximations and
time-discretised diffusions. Our experimental results are promising and show
that the IS type approach can provide substantial gains relative to an
analogous DA scheme, and is often competitive even without parallelisation.
| stat |
Can You Trust This Prediction? Auditing Pointwise Reliability After
Learning | To use machine learning in high stakes applications (e.g. medicine), we need
tools for building confidence in the system and evaluating whether it is
reliable. Methods to improve model reliability often require new learning
algorithms (e.g. using Bayesian inference to obtain uncertainty estimates). An
alternative is to audit a model after it is trained. In this paper, we describe
resampling uncertainty estimation (RUE), an algorithm to audit the pointwise
reliability of predictions. Intuitively, RUE estimates the amount that a
prediction would change if the model had been fit on different training data.
The algorithm uses the gradient and Hessian of the model's loss function to
create an ensemble of predictions. Experimentally, we show that RUE more
effectively detects inaccurate predictions than existing tools for auditing
reliability subsequent to training. We also show that RUE can create predictive
distributions that are competitive with state-of-the-art methods like Monte
Carlo dropout, probabilistic backpropagation, and deep ensembles, but does not
depend on specific algorithms at train-time like these methods do.
| stat |
Is Time to Intervention in the COVID-19 Outbreak Really Important? A
Global Sensitivity Analysis Approach | Italy has been one of the first countries timewise strongly impacted by the
COVID-19 pandemic. The adoption of social distancing and heavy lockdown
measures is posing a heavy burden on the population and the economy. The timing
of the measures has crucial policy-making implications. Using publicly
available data for the pandemic progression in Italy, we quantitatively assess
the effect of the intervention time on the pandemic expansion, with a
methodology that combines a generalized
susceptible-exposed-infectious-recovered (SEIR) model together with statistical
learning methods. The modeling shows that the lockdown has strongly deviated
the pandemic trajectory in Italy. However, the difference between the forecasts
and real data up to 20 April 2020 can be explained only by the existence of a
time lag between the actual issuance date and the full effect of the measures.
To understand the relative importance of intervention with respect to other
factors, a thorough uncertainty quantification of the model predictions is
performed. Global sensitivity indices show that the the time of intervention is
4 times more relevant than quarantine, and eight times more important than
intrinsic features of the pandemic such as protection and infection rates. The
relevance of their interactions is also quantified and studied.
| stat |
The perils of automated fitting of datasets: the case of a wind turbine
cost model | Rinne et al. conduct an interesting analysis of the impact of wind turbine
technology and land-use on wind power potentials, which allows profound
insights into each factors contribution to overall potentials. The paper
presents a detailed model of site-specific wind turbine investment cost (i.e.
road- and grid access costs) complemented by a model used to estimate
site-independent costs. We believe that propose a cutting edge model of
site-specific investment costs. However, the site-independent cost model is
flawed in our opinion. This flaw most likely does not impact the results
presented in the paper, although we expect a considerable generalization error.
Thus the application of the wind turbine cost model in other contexts may lead
to unreasonable results. More generally, the derivation of the wind turbine
cost model serves as an example of how applications of automated regression
analysis can go wrong.
| stat |
Nonpar MANOVA via Independence Testing | The $k$-sample testing problem tests whether or not $k$ groups of data points
are sampled from the same distribution. Multivariate analysis of variance
(MANOVA) is currently the gold standard for $k$-sample testing but makes
strong, often inappropriate, parametric assumptions. Moreover, independence
testing and $k$-sample testing are tightly related, and there are many
nonparametric multivariate independence tests with strong theoretical and
empirical properties, including distance correlation (Dcorr) and
Hilbert-Schmidt-Independence-Criterion (Hsic). We prove that universally
consistent independence tests achieve universally consistent $k$-sample testing
and that $k$-sample statistics like Energy and Maximum Mean Discrepancy (MMD)
are exactly equivalent to Dcorr. Empirically evaluating these tests for
$k$-sample scenarios demonstrates that these nonparametric independence tests
typically outperform MANOVA, even for Gaussian distributed settings. Finally,
we extend these non-parametric $k$-sample testing procedures to perform
multiway and multilevel tests. Thus, we illustrate the existence of many
theoretically motivated and empirically performant $k$-sample tests. A Python
package with all independence and k-sample tests called hyppo is available from
https://hyppo.neurodata.io/.
| stat |
A Wasserstein Minimum Velocity Approach to Learning Unnormalized Models | Score matching provides an effective approach to learning flexible
unnormalized models, but its scalability is limited by the need to evaluate a
second-order derivative. In this paper, we present a scalable approximation to
a general family of learning objectives including score matching, by observing
a new connection between these objectives and Wasserstein gradient flows. We
present applications with promise in learning neural density estimators on
manifolds, and training implicit variational and Wasserstein auto-encoders with
a manifold-valued prior.
| stat |
Path Dependent Structural Equation Models | Causal analyses of longitudinal data generally assume that the qualitative
causal structure relating variables remains invariant over time. In structured
systems that transition between qualitatively different states in discrete time
steps, such an approach is deficient on two fronts. First, time-varying
variables may have state-specific causal relationships that need to be
captured. Second, an intervention can result in state transitions downstream of
the intervention different from those actually observed in the data. In other
words, interventions may counterfactually alter the subsequent temporal
evolution of the system. We introduce a generalization of causal graphical
models, Path Dependent Structural Equation Models (PDSEMs), that can describe
such systems. We show how causal inference may be performed in such models and
illustrate its use in simulations and data obtained from a septoplasty surgical
procedure.
| stat |
Latent function-on-scalar regression models for observed sequences of
binary data: a restricted likelihood approach | In this paper, we study a functional regression setting where the random
response curve is unobserved, and only its dichotomized version observed at a
sequence of correlated binary data is available. We propose a practical
computational framework for maximum likelihood analysis via the parameter
expansion technique. Compared to existing methods, our proposal relies on the
use of a complete data likelihood, with the advantage of being able to handle
non-equally spaced and missing observations effectively. The proposed method is
used in the Function-on-Scalar regression setting, with the latent response
variable being a Gaussian random element taking values in a separable Hilbert
space. Smooth estimations of functional regression coefficients and principal
components are provided by introducing an adaptive MCEM algorithm that
circumvents selecting the smoothing parameters. Finally, the performance of our
novel method is demonstrated by various simulation studies and on a real case
study. The proposed method is implemented in the R package dfrr.
| stat |
Iterative regularization for convex regularizers | We study iterative regularization for linear models, when the bias is convex
but not necessarily strongly convex. We characterize the stability properties
of a primal-dual gradient based approach, analyzing its convergence in the
presence of worst case deterministic noise. As a main example, we specialize
and illustrate the results for the problem of robust sparse recovery. Key to
our analysis is a combination of ideas from regularization theory and
optimization in the presence of errors. Theoretical results are complemented by
experiments showing that state-of-the-art performances can be achieved with
considerable computational speed-ups.
| stat |
Active set algorithms for estimating shape-constrained density ratios | In many instances, imposing a constraint on the shape of a density is a
reasonable and flexible assumption. It offers an alternative to parametric
models which can be too rigid and to other nonparametric methods requiring the
choice of tuning parameters. This paper treats the nonparametric estimation of
log-concave or log-convex density ratios by means of active set algorithms in a
unified framework. In the setting of log-concave densities, the new algorithm
is similar to but substantially faster than previously considered active set
methods. Log-convexity is a less common shape constraint which is described by
some authors as "tail inflation". The active set method proposed here is novel
in this context. As a by-product, new goodness-of-fit tests of single
hypotheses are formulated and are shown to be more powerful than higher
criticism tests in a simulation study.
| stat |
Heterogeneous Multi-task Metric Learning across Multiple Domains | Distance metric learning (DML) plays a crucial role in diverse machine
learning algorithms and applications. When the labeled information in target
domain is limited, transfer metric learning (TML) helps to learn the metric by
leveraging the sufficient information from other related domains. Multi-task
metric learning (MTML), which can be regarded as a special case of TML,
performs transfer across all related domains. Current TML tools usually assume
that the same feature representation is exploited for different domains.
However, in real-world applications, data may be drawn from heterogeneous
domains. Heterogeneous transfer learning approaches can be adopted to remedy
this drawback by deriving a metric from the learned transformation across
different domains. But they are often limited in that only two domains can be
handled. To appropriately handle multiple domains, we develop a novel
heterogeneous multi-task metric learning (HMTML) framework. In HMTML, the
metrics of all different domains are learned together. The transformations
derived from the metrics are utilized to induce a common subspace, and the
high-order covariance among the predictive structures of these domains is
maximized in this subspace. There do exist a few heterogeneous transfer
learning approaches that deal with multiple domains, but the high-order
statistics (correlation information), which can only be exploited by
simultaneously examining all domains, is ignored in these approaches. Compared
with them, the proposed HMTML can effectively explore such high-order
information, thus obtaining more reliable feature transformations and metrics.
Effectiveness of our method is validated by the extensive and intensive
experiments on text categorization, scene classification, and social image
annotation.
| stat |
Analysis of high-dimensional Continuous Time Markov Chains using the
Local Bouncy Particle Sampler | Sampling the parameters of high-dimensional Continuous Time Markov Chains
(CTMC) is a challenging problem with important applications in many fields of
applied statistics. In this work a recently proposed type of non-reversible
rejection-free Markov Chain Monte Carlo (MCMC) sampler, the Bouncy Particle
Sampler (BPS), is brought to bear to this problem. BPS has demonstrated its
favorable computational efficiency compared with state-of-the-art MCMC
algorithms, however to date applications to real-data scenario were scarce. An
important aspect of the practical implementation of BPS is the simulation of
event times. Default implementations use conservative thinning bounds. Such
bounds can slow down the algorithm and limit the computational performance. Our
paper develops an algorithm with an exact analytical solution to the random
event times in the context of CTMCs. Our local version of BPS algorithm takes
advantage of the sparse structure in the target factor graph and we also
provide a framework for assessing the computational complexity of local BPS
algorithms.
| stat |
Generalised learning of time-series: Ornstein-Uhlenbeck processes | In machine learning, statistics, econometrics and statistical physics
cross-validation (CV) is used as a standard approach in quantifying the
generalization performance of a statistical model. In practice, direct usage of
CV is avoided for time-series due to several issues. A direct application of CV
in time-series leads to the loss of serial correlations, a requirement of
preserving any non-stationarity and the prediction of the past data using
future data. In this work, we propose a meta-algorithm called reconstructive
cross-validation (rCV ) that avoids all these issues. At first, k folds are
formed with non-overlapping randomly selected subsets of the original
time-series. Then, we generate k new partial time-series by removing data
points from a given fold: every new partial time-series have missing points at
random from a different entire fold. A suitable imputation or a smoothing
technique is used to reconstruct k time-series. We call these reconstructions
secondary models. Thereafter, we build the primary k time-series models using
new time-series coming from the secondary models. The performance of the
primary models is evaluated simultaneously by computing the deviations from the
originally removed data points and out-of-sample (OSS) data. These amounts to
reconstruction and prediction errors. If the secondary models use a technique
that retains the data points exactly, such as Gaussian process regression,
there will be no errors present on the data points that are not removed. By
this procedure serial correlations are retained, any non-stationarity is
preserved within models and there will be no prediction of past data using the
future data points. The cross-validation in time-series model can be practised
with rCV. Moreover, we can build time-series learning curves by repeating rCV
procedure with different k values.
| stat |
Deterministic bootstrapping for a class of bootstrap methods | An algorithm is described that enables efficient deterministic approximate
computation of the bootstrap distribution for any linear bootstrap method
$T_n^*$, alleviating the need for repeated resampling from observations (resp.
input-derived data). In essence, the algorithm computes the distribution
function from a linear mixture of independent random variables each having a
finite discrete distribution. The algorithm is applicable to elementary
bootstrap scenarios (targetting the mean as parameter of interest), for block
bootstrap, as well as for certain residual bootstrap scenarios. Moreover, the
algorithm promises a much broader applicability, in non-bootstrapped hypothesis
testing.
| stat |
Complex-Valued vs. Real-Valued Neural Networks for Classification
Perspectives: An Example on Non-Circular Data | The contributions of this paper are twofold. First, we show the potential
interest of Complex-Valued Neural Network (CVNN) on classification tasks for
complex-valued datasets. To highlight this assertion, we investigate an example
of complex-valued data in which the real and imaginary parts are statistically
dependent through the property of non-circularity. In this context, the
performance of fully connected feed-forward CVNNs is compared against a
real-valued equivalent model. The results show that CVNN performs better for a
wide variety of architectures and data structures. CVNN accuracy presents a
statistically higher mean and median and lower variance than Real-Valued Neural
Network (RVNN). Furthermore, if no regularization technique is used, CVNN
exhibits lower overfitting. The second contribution is the release of a Python
library (Barrachina 2019) using Tensorflow as back-end that enables the
implementation and training of CVNNs in the hopes of motivating further
research on this area.
| stat |
Evaluation of approaches for accommodating interactions and non-linear
terms in multiple imputation of incomplete three-level data | Three-level data structures arising from repeated measures on individuals
clustered within larger units are common in health research studies. Missing
data are prominent in such studies and are often handled via multiple
imputation (MI). Although several MI approaches can be used to account for the
three-level structure, including adaptations to single- and two-level
approaches, when the substantive analysis model includes interactions or
quadratic effects these too need to be accommodated in the imputation model. In
such analyses, substantive model compatible (SMC) MI has shown great promise in
the context of single-level data. While there have been recent developments in
multilevel SMC MI, to date only one approach that explicitly handles incomplete
three-level data is available. Alternatively, researchers can use pragmatic
adaptations to single- and two-level MI approaches, or two-level SMC-MI
approaches. We describe the available approaches and evaluate them via
simulation in the context of a three three-level random effects analysis models
involving an interaction between the incomplete time-varying exposure and time,
an interaction between the time-varying exposure and an incomplete time-fixed
confounder, or a quadratic effect of the exposure. Results showed that all
approaches considered performed well in terms of bias and precision when the
target analysis involved an interaction with time, but the three-level SMC MI
approach performed best when the target analysis involved an interaction
between the time-varying exposure and an incomplete time-fixed confounder, or a
quadratic effect of the exposure. We illustrate the methods using data from the
Childhood to Adolescence Transition Study.
| stat |
Computing the Dirichlet-Multinomial Log-Likelihood Function | Dirichlet-multinomial (DMN) distribution is commonly used to model
over-dispersion in count data. Precise and fast numerical computation of the
DMN log-likelihood function is important for performing statistical inference
using this distribution, and remains a challenge. To address this, we use
mathematical properties of the gamma function to derive a closed form
expression for the DMN log-likelihood function. Compared to existing methods,
calculation of the closed form has a lower computational complexity, hence is
much faster without comprimising computational accuracy.
| stat |
Bridging preference-based instrumental variable studies and
cluster-randomized encouragement experiments: study design, noncompliance,
and average cluster effect ratio | Instrumental variable methods are widely used in medical and social science
research to draw causal conclusions when the treatment and outcome are
confounded by unmeasured confounding variables. One important feature of such
studies is that the instrumental variable is often applied at the cluster
level, e.g., hospitals' or physicians' preference for a certain treatment where
each hospital or physician naturally defines a cluster. This paper proposes to
embed such observational instrumental variable data into a cluster-randomized
encouragement experiment using statistical matching. Potential outcomes and
causal assumptions underpinning the design are formalized and examined. Testing
procedures for two commonly-used estimands, Fisher's sharp null hypothesis and
the pooled effect ratio, are extended to the current setting. We then introduce
a novel cluster-heterogeneous proportional treatment effect model and the
relevant estimand: the average cluster effect ratio. This new estimand is
advantageous over the structural parameter in a constant proportional treatment
effect model in that it allows treatment heterogeneity, and is advantageous
over the pooled effect ratio estimand in that it is immune to Simpson's
paradox. We develop an asymptotically valid randomization-based testing
procedure for this new estimand based on solving a mixed integer
quadratically-constrained optimization problem. The proposed design and
inferential methods are applied to a study of the effect of using
transesophageal echocardiography during CABG surgery on patients' 30-day
mortality rate.
| stat |
Modularizing Deep Learning via Pairwise Learning With Kernels | By redefining the conventional notions of layers, we present an alternative
view on finitely wide, fully trainable deep neural networks as stacked linear
models in feature spaces, leading to a kernel machine interpretation. Based on
this construction, we then propose a provably optimal modular learning
framework for classification that does not require between-module
backpropagation. This modular approach brings new insights into the label
requirement of deep learning: It leverages only implicit pairwise labels (weak
supervision) when learning the hidden modules. When training the output module,
on the other hand, it requires full supervision but achieves high label
efficiency, needing as few as 10 randomly selected labeled examples (one from
each class) to achieve 94.88% accuracy on CIFAR-10 using a ResNet-18 backbone.
Moreover, modular training enables fully modularized deep learning workflows,
which then simplify the design and implementation of pipelines and improve the
maintainability and reusability of models. To showcase the advantages of such a
modularized workflow, we describe a simple yet reliable method for estimating
reusability of pre-trained modules as well as task transferability in a
transfer learning setting. At practically no computation overhead, it precisely
described the task space structure of 15 binary classification tasks from
CIFAR-10.
| stat |
Semiparametric Causal Sufficient Dimension Reduction Of High Dimensional
Treatments | Cause-effect relationships are typically evaluated by comparing the outcome
responses to binary treatment values, representing two arms of a hypothetical
randomized controlled trial. However, in certain applications, treatments of
interest are continuous and high dimensional. For example, understanding the
causal relationship between severity of radiation therapy, represented by a
high dimensional vector of radiation exposure values and post-treatment side
effects is a problem of clinical interest in radiation oncology. An appropriate
strategy for making interpretable causal conclusions is to reduce the dimension
of treatment. If individual elements of a high dimensional treatment vector
weakly affect the outcome, but the overall relationship between the treatment
variable and the outcome is strong, careless approaches to dimension reduction
may not preserve this relationship. Moreover, methods developed for regression
problems do not transfer in a straightforward way to causal inference due to
confounding complications between the treatment and outcome. In this paper, we
use semiparametric inference theory for structural models to give a general
approach to causal sufficient dimension reduction of a high dimensional
treatment such that the cause-effect relationship between the treatment and
outcome is preserved. We illustrate the utility of our proposal through
simulations and a real data application in radiation oncology.
| stat |
Natural Experiments | The term natural experiment is used inconsistently. In one interpretation, it
refers to an experiment where a treatment is randomly assigned by someone other
than the researcher. In another interpretation, it refers to a study in which
there is no controlled random assignment, but treatment is assigned by some
external factor in a way that loosely resembles a randomized experiment---often
described as an "as if random" assignment. In yet another interpretation, it
refers to any non-randomized study that compares a treatment to a control
group, without any specific requirements on how the treatment is assigned. I
introduce an alternative definition that seeks to clarify the integral features
of natural experiments and at the same time distinguish them from randomized
controlled experiments. I define a natural experiment as a research study where
the treatment assignment mechanism (i) is neither designed nor implemented by
the researcher, (ii) is unknown to the researcher, and (iii) is probabilistic
by virtue of depending on an external factor. The main message of this
definition is that the difference between a randomized controlled experiment
and a natural experiment is not a matter of degree, but of essence, and thus
conceptualizing a natural experiment as a research design akin to a randomized
experiment is neither rigorous nor a useful guide to empirical analysis. Using
my alternative definition, I discuss how a natural experiment differs from a
traditional observational study, and offer practical recommendations for
researchers who wish to use natural experiments to study causal effects.
| stat |
Identification and Inference of Network Formation Games with
Misclassified Links | This paper considers a network formation model when links are potentially
measured with error. We focus on a game-theoretical model of strategic network
formation with incomplete information, in which the linking decisions depend on
agents' exogenous attributes and endogenous network characteristics. In the
presence of link misclassification, we derive moment conditions that
characterize the identified set for the preference parameters associated with
homophily and network externalities. Based on the moment equality conditions,
we provide an inference method that is asymptotically valid when a single
network of many agents is observed. Finally, we apply our proposed method to
study trust networks in rural villages in southern India.
| stat |
Identifiability of Latent Class Models with Covariates | Latent class models with covariates are widely used for psychological,
social, and educational researches. Yet the fundamental identifiability issue
of these models has not been fully addressed. Among the previous researches on
the identifiability of latent class models containing covariates, Huang and
Bandeen-Roche (2004, Psychometrika, 69:5-32) studied the local identifiability
conditions. However, motivated by recent advances in the identifiability of
restricted latent class models, particularly the Cognitive Diagnosis Models
(CDMs), we show in this work that the conditions in Huang and Bandeen-Roche
(2004) are only necessary but not sufficient to determine the local
identifiability of the model parameters. To address the open identifiability
issue for latent class models with covariates, this work establishes conditions
to ensure the global identifiability of the model parameters in both strict and
generic sense. Moreover, our results extend to polytomous-response CDMs with
covariates, which generalizes the existing identifiability results for CDMs.
| stat |
A flexible Particle Markov chain Monte Carlo method | Particle Markov Chain Monte Carlo methods are used to carry out inference in
non-linear and non-Gaussian state space models, where the posterior density of
the states is approximated using particles. Current approaches usually perform
Bayesian inference using either a particle Marginal Metropolis-Hastings (PMMH)
algorithm or a particle Gibbs (PG) sampler. This paper shows how the two ways
of generating variables mentioned above can be combined in a flexible manner to
give sampling schemes that converge to a desired target distribution. The
advantage of our approach is that the sampling scheme can be tailored to obtain
good results for different applications. For example, when some parameters and
the states are highly correlated, such parameters can be generated using PMMH,
while all other parameters are generated using PG because it is easier to
obtain good proposals for the parameters within the PG framework. We derive
some convergence properties of our sampling scheme and also investigate its
performance empirically by applying it to univariate and multivariate
stochastic volatility models and comparing it to other PMCMC methods proposed
in the literature.
| stat |
Hijacking Malaria Simulators with Probabilistic Programming | Epidemiology simulations have become a fundamental tool in the fight against
the epidemics of various infectious diseases like AIDS and malaria. However,
the complicated and stochastic nature of these simulators can mean their output
is difficult to interpret, which reduces their usefulness to policymakers. In
this paper, we introduce an approach that allows one to treat a large class of
population-based epidemiology simulators as probabilistic generative models.
This is achieved by hijacking the internal random number generator calls,
through the use of a universal probabilistic programming system (PPS). In
contrast to other methods, our approach can be easily retrofitted to simulators
written in popular industrial programming frameworks. We demonstrate that our
method can be used for interpretable introspection and inference, thus shedding
light on black-box simulators. This reinstates much-needed trust between
policymakers and evidence-based methods.
| stat |
Decision Variance in Online Learning | Online learning has traditionally focused on the expected rewards. In this
paper, a risk-averse online learning problem under the performance measure of
the mean-variance of the rewards is studied. Both the bandit and full
information settings are considered. The performance of several existing
policies is analyzed, and new fundamental limitations on risk-averse learning
is established. In particular, it is shown that although a logarithmic
distribution-dependent regret in time $T$ is achievable (similar to the
risk-neutral problem), the worst-case (i.e. minimax) regret is lower bounded by
$\Omega(T)$ (in contrast to the $\Omega(\sqrt{T})$ lower bound in the
risk-neutral problem). This sharp difference from the risk-neutral counterpart
is caused by the the variance in the player's decisions, which, while absent in
the regret under the expected reward criterion, contributes to excess
mean-variance due to the non-linearity of this risk measure. The role of the
decision variance in regret performance reflects a risk-averse player's desire
for robust decisions and outcomes.
| stat |
Improve variational autoEncoder with auxiliary softmax multiclassifier | As a general-purpose generative model architecture, VAE has been widely used
in the field of image and natural language processing. VAE maps high
dimensional sample data into continuous latent variables with unsupervised
learning. Sampling in the latent variable space of the feature, VAE can
construct new image or text data. As a general-purpose generation model, the
vanilla VAE can not fit well with various data sets and neural networks with
different structures. Because of the need to balance the accuracy of
reconstruction and the convenience of latent variable sampling in the training
process, VAE often has problems known as "posterior collapse". images
reconstructed by VAE are also often blurred. In this paper, we analyze the main
cause of these problem, which is the lack of mutual information between the
sample variable and the latent feature variable during the training process. To
maintain mutual information in model training, we propose to use the auxiliary
softmax multi-classification network structure to improve the training effect
of VAE, named VAE-AS. We use MNIST and Omniglot data sets to test the VAE-AS
model. Based on the test results, It can be show that VAE-AS has obvious
effects on the mutual information adjusting and solving the posterior collapse
problem.
| stat |
Coupling physical understanding and statistical modeling to estimate ice
jam flood frequency in the northern Peace-Athabasca Delta under climate
change | The Peace-Athabasca Delta (PAD) of northwestern Alberta is one of the largest
inland freshwater deltas in the world, laying at the confluence of the Peace
and Athabasca Rivers. The PAD is recognized as a having unique ecological
significance and periodic ice jam flooding from both rivers is an important
feature of its current ecology. Past studies have debated whether a change in
ice jam flood (IJF) frequency on the Peace River has recently occurred, and
what factors might be driving any perceived changes. This study contributes to
this debate by addressing two questions: (1) what factors are most predictive
of Peace River IJFs, and (2) how might climate change impact IJF frequency?
This work starts with a physically-based conceptual model of the necessary
conditions for a large Peace River IJF, and the factors that indicate whether
those conditions are met. Logistic regression is applied to the historical
flood record to determine which combination of hydroclimatic and riverine
factors best predict IJFs and the uncertainty in those relationships given the
available data. Winter precipitation and temperature are most predictive of
Peace River IJFs, while freeze-up elevation contains little predictive power
and is not closely related to IJF occurrence. The best logistic regression
model is forced with downscaled climate change scenarios from multiple climate
models to project IJF frequency for a variety of plausible futures. Parametric
uncertainty in the best logistic regression model is propagated into the
projections using a parametric bootstrap to sample many plausible statistical
models. Although there is variability across emissions scenarios and climate
models, all projections indicate that the frequency of Peace River IJFs is
likely to decrease substantially in the coming decades, and that average
waiting times between future IJFs will likely surpass recent experience.
| stat |
Shrinkage with shrunken shoulders: inference via geometrically /
uniformly ergodic Gibbs sampler | Use of continuous shrinkage priors -- with a "spike" near zero and
heavy-tails towards infinity -- is an increasingly popular approach to induce
sparsity in parameter estimates. When the parameters are only weakly identified
by the likelihood, however, the posterior may end up with tails as heavy as the
prior, jeopardizing robustness of inference. A natural solution is to "shrink
the shoulders" of a shrinkage prior by lightening up its tails beyond a
reasonable parameter range, yielding the regularized version of the prior. We
develop a regularization approach which, unlike previous proposals, preserves
computationally attractive structures of original shrinkage priors. We study
theoretical properties of the Gibbs sampler on resulting posterior
distributions, with emphasis on convergence rates of the P{\'o}lya-Gamma Gibbs
sampler for sparse logistic regression. Our analysis shows that the proposed
regularization leads to geometric ergodicity under a broad range of
global-local shrinkage priors. Essentially, the only requirement is for the
prior $\pi_{\rm local}$ on the local scale $\lambda$ to satisfy $\pi_{\rm
local}(0) < \infty$. If $\pi_{\rm local}(\cdot)$ further satisfies
$\lim_{\lambda \to 0} \pi_{\rm local}(\lambda) / \lambda^a < \infty$ for $a >
0$, as in the case of Bayesian bridge priors, we show the sampler to be
uniformly ergodic.
| stat |
Tailored Bayes: a risk modelling framework under unequal
misclassification costs | Risk prediction models are a crucial tool in healthcare. Risk prediction
models with a binary outcome (i.e., binary classification models) are often
constructed using methodology which assumes the costs of different
classification errors are equal. In many healthcare applications this
assumption is not valid, and the differences between misclassification costs
can be quite large. For instance, in a diagnostic setting, the cost of
misdiagnosing a person with a life-threatening disease as healthy may be larger
than the cost of misdiagnosing a healthy person as a patient. In this work, we
present Tailored Bayes (TB), a novel Bayesian inference framework which
"tailors" model fitting to optimise predictive performance with respect to
unbalanced misclassification costs. We use simulation studies to showcase when
TB is expected to outperform standard Bayesian methods in the context of
logistic regression. We then apply TB to three real-world applications, a
cardiac surgery, a breast cancer prognostication task and a breast cancer
tumour classification task, and demonstrate the improvement in predictive
performance over standard methods.
| stat |
Theoretical Analysis of Adversarial Learning: A Minimax Approach | Here we propose a general theoretical method for analyzing the risk bound in
the presence of adversaries. Specifically, we try to fit the adversarial
learning problem into the minimax framework. We first show that the original
adversarial learning problem can be reduced to a minimax statistical learning
problem by introducing a transport map between distributions. Then, we prove a
new risk bound for this minimax problem in terms of covering numbers under a
weak version of Lipschitz condition. Our method can be applied to multi-class
classification problems and commonly used loss functions such as the hinge and
ramp losses. As some illustrative examples, we derive the adversarial risk
bounds for SVMs, deep neural networks, and PCA, and our bounds have two
data-dependent terms, which can be optimized for achieving adversarial
robustness.
| stat |
Statistical evaluation of in-vivo bioassays in regulatory toxicology
considering males and females | The separate evaluation for males and females is the recent standard in
in-vivo toxicology for dose or treatment effects using Dunnett tests. The
alternative pre-test for sex-by-treatment interaction is problematic. Here a
joint test is proposed considering the two sex-specific and the pooled
Dunnett-type comparisons. The calculation of either simultaneous confidence
intervals or adjusted p-values with the R-package multcomp is demonstrated
using a real data example.
| stat |
De-Biased Machine Learning of Global and Local Parameters Using
Regularized Riesz Representers | We provide adaptive inference methods, based on $\ell_1$ regularization, for
regular (semiparametric) and non-regular (nonparametric) linear functionals of
the conditional expectation function. Examples of regular functionals include
average treatment effects, policy effects, and derivatives. Examples of
non-regular functionals include average treatment effects, policy effects, and
derivatives conditional on a covariate subvector fixed at a point. We construct
a Neyman orthogonal equation for the target parameter that is approximately
invariant to small perturbations of the nuisance parameters. To achieve this
property, we include the Riesz representer for the functional as an additional
nuisance parameter. Our analysis yields weak "double sparsity robustness":
either the approximation to the regression or the approximation to the
representer can be "completely dense" as long as the other is sufficiently
"sparse". Our main results are non-asymptotic and imply asymptotic uniform
validity over large classes of models, translating into honest confidence bands
for both global and local parameters.
| stat |
Treatment Effect Bias from Sample Snooping: Blinding Outcomes is Neither
Necessary nor Sufficient | Popular guidance on observational data analysis states that outcomes should
be blinded when determining matching criteria or propensity scores. Such a
blinding is informally said to maintain the "objectivity" of the analysis, and
to prevent analysts from artificially amplifying the treatment effect by
exploiting chance imbalances. Contrary to this notion, we show that outcome
blinding is not a sufficient safeguard against fishing. Blinded and unblinded
analysts can produce bias of the same order of magnitude in cases where the
outcomes can be approximately predicted from baseline covariates. We illustrate
this vulnerability with a combination of analytical results and simulations.
Finally, to show that outcome blinding is not necessary to prevent bias, we
outline an alternative sample partitioning procedure for estimating the average
treatment effect on the controls, or the average treatment effect on the
treated. This procedure uses all of the the outcome data from all partitions in
the final analysis step, but does not require the analysis to not be fully
prespecified.
| stat |
Statistical Inference of the Value Function for Reinforcement Learning
in Infinite Horizon Settings | Reinforcement learning is a general technique that allows an agent to learn
an optimal policy and interact with an environment in sequential decision
making problems. The goodness of a policy is measured by its value function
starting from some initial state. The focus of this paper is to construct
confidence intervals (CIs) for a policy's value in infinite horizon settings
where the number of decision points diverges to infinity. We propose to model
the action-value state function (Q-function) associated with a policy based on
series/sieve method to derive its confidence interval. When the target policy
depends on the observed data as well, we propose a SequentiAl Value Evaluation
(SAVE) method to recursively update the estimated policy and its value
estimator. As long as either the number of trajectories or the number of
decision points diverges to infinity, we show that the proposed CI achieves
nominal coverage even in cases where the optimal policy is not unique.
Simulation studies are conducted to back up our theoretical findings. We apply
the proposed method to a dataset from mobile health studies and find that
reinforcement learning algorithms could help improve patient's health status.
| stat |
Network depth: identifying median and contours in complex networks | Centrality descriptors are widely used to rank nodes according to specific
concept(s) of importance. Despite the large number of centrality measures
available nowadays, it is still poorly understood how to identify the node
which can be considered as the `centre' of a complex network. In fact, this
problem corresponds to finding the median of a complex network. The median is a
non-parametric and robust estimator of the location parameter of a probability
distribution. In this work, we present the most natural generalisation of the
concept of median to the realm of complex networks, discussing its advantages
for defining the centre of the system and percentiles around that centre. To
this aim, we introduce a new statistical data depth and we apply it to networks
embedded in a geometric space induced by different metrics. The application of
our framework to empirical networks allows us to identify median nodes which
are socially or biologically relevant.
| stat |
A Bayesian - Deep Learning model for estimating Covid-19 evolution in
Spain | This work proposes a semi-parametric approach to estimate Covid-19
(SARS-CoV-2) evolution in Spain. Considering the sequences of 14 days
cumulative incidence of all Spanish regions, it combines modern Deep Learning
(DL) techniques for analyzing sequences with the usual Bayesian Poisson-Gamma
model for counts. DL model provides a suitable description of observed
sequences but no reliable uncertainty quantification around it can be obtained.
To overcome this we use the prediction from DL as an expert elicitation of the
expected number of counts along with their uncertainty and thus obtaining the
posterior predictive distribution of counts in an orthodox Bayesian analysis
using the well known Poisson-Gamma model. The overall resulting model allows us
to either predict the future evolution of the sequences on all regions, as well
as, estimating the consequences of eventual scenarios.
| stat |
Temporal orders and causal vector for physiological data analysis | In addition to the global parameter- and time-series-based approaches,
physiological analyses should constitute a local temporal one, particularly
when analyzing data within protocol segments. Hence, we introduce the R package
implementing the estimation of temporal orders with a causal vector (CV). It
may use linear modeling or time series distance. The algorithm was tested on
cardiorespiratory data comprising tidal volume and tachogram curves, obtained
from elite athletes (supine and standing, in static conditions) and a control
group (different rates and depths of breathing, while supine). We checked the
relation between CV and body position or breathing style. The rate of breathing
had a greater impact on the CV than does the depth. The tachogram curve
preceded the tidal volume relatively more when breathing was slower. Clinical
relevance - The method can assess (1) relationships between two signals, having
one of the two time-shifted in a given range, (2) curves of most optimal
inter-signal shift, and (c) their stability across time; also for other
physiological studies.
| stat |
Dirichlet Process Mixture Models for Regression Discontinuity Designs | The Regression Discontinuity Design (RDD) is a quasi-experimental design that
estimates the causal effect of a treatment when its assignment is defined by a
threshold value for a continuous assignment variable. The RDD assumes that
subjects with measurements within a bandwidth around the threshold belong to a
common population, so that the threshold can be seen as a randomising device
assigning treatment to those falling just above the threshold and withholding
it from those who fall just below.
Bandwidth selection represents a compelling decision for the RDD analysis as
the results may be highly sensitive to its choice. A number of methods to
select the optimal bandwidth, mainly originating from the econometric
literature, have been proposed. However, their use in practice is limited.
We propose a methodology that, tackling the problem from an applied point of
view, consider units' exchangeability, i.e., their similarity with respect to
measured covariates, as the main criteria to select subjects for the analysis,
irrespectively of their distance from the threshold. We carry out clustering on
the sample using a Dirichlet process mixture model to identify balanced and
homogeneous clusters. Our proposal exploits the posterior similarity matrix,
which contains the pairwise probabilities that two observations are allocated
to the same cluster in the MCMC sample. Thus we include in the RDD analysis
only those clusters for which we have stronger evidence of exchangeability.
We illustrate the validity of our methodology with both a simulated
experiment and a motivating example on the effect of statins to lower
cholesterol level, using UK primary care data.
| stat |
Reparameterizing Distributions on Lie Groups | Reparameterizable densities are an important way to learn probability
distributions in a deep learning setting. For many distributions it is possible
to create low-variance gradient estimators by utilizing a `reparameterization
trick'. Due to the absence of a general reparameterization trick, much research
has recently been devoted to extend the number of reparameterizable
distributional families. Unfortunately, this research has primarily focused on
distributions defined in Euclidean space, ruling out the usage of one of the
most influential class of spaces with non-trivial topologies: Lie groups. In
this work we define a general framework to create reparameterizable densities
on arbitrary Lie groups, and provide a detailed practitioners guide to further
the ease of usage. We demonstrate how to create complex and multimodal
distributions on the well known oriented group of 3D rotations,
$\operatorname{SO}(3)$, using normalizing flows. Our experiments on applying
such distributions in a Bayesian setting for pose estimation on objects with
discrete and continuous symmetries, showcase their necessity in achieving
realistic uncertainty estimates.
| stat |
Modeling Interference Via Symmetric Treatment Decomposition | Classical causal inference assumes treatments meant for a given unit do not
have an effect on other units. This assumption is violated in interference
problems, where new types of spillover causal effects arise, and causal
inference becomes much more difficult. In addition, interference introduces a
unique complication where variables may transmit treatment influences to each
other, which is a relationship that has some features of a causal one, but is
symmetric. In settings where a natural causal ordering on variables is not
available, addressing this complication using statistical inference methods
based on Directed Acyclic Graphs (DAGs) leads to conceptual difficulties. In
this paper, we develop a new approach to decomposing the spillover effect into
unit-specific components that extends the DAG based treatment decomposition
approach to mediation of Robins and Richardson. We give conditions for these
components of the spillover effect to be identified in a natural type of causal
model that permits stable symmetric relations among outcomes induced by a
process in equilibrium. We discuss statistical inference for identified
components of the spillover effect, including a maximum likelihood estimator,
and a doubly robust estimator for the special case of two interacting outcomes.
We verify consistency and robustness of our estimators via a simulation study,
and illustrate our method by assessing the causal effect of education
attainment on depressive symptoms using the data on households from the
Wisconsin Longitudinal Study.
| stat |
Communication-efficient distributed eigenspace estimation | Distributed computing is a standard way to scale up machine learning and data
science algorithms to process large amounts of data. In such settings, avoiding
communication amongst machines is paramount for achieving high performance.
Rather than distribute the computation of existing algorithms, a common
practice for avoiding communication is to compute local solutions or parameter
estimates on each machine and then combine the results; in many convex
optimization problems, even simple averaging of local solutions can work well.
However, these schemes do not work when the local solutions are not unique.
Spectral methods are a collection of such problems, where solutions are
orthonormal bases of the leading invariant subspace of an associated data
matrix, which are only unique up to rotation and reflections. Here, we develop
a communication-efficient distributed algorithm for computing the leading
invariant subspace of a data matrix. Our algorithm uses a novel alignment
scheme that minimizes the Procrustean distance between local solutions and a
reference solution, and only requires a single round of communication. For the
important case of principal component analysis (PCA), we show that our
algorithm achieves a similar error rate to that of a centralized estimator. We
present numerical experiments demonstrating the efficacy of our proposed
algorithm for distributed PCA, as well as other problems where solutions
exhibit rotational symmetry, such as node embeddings for graph data and
spectral initialization for quadratic sensing.
| stat |
Shapley Values of Reconstruction Errors of PCA for Explaining Anomaly
Detection | We present a method to compute the Shapley values of reconstruction errors of
principal component analysis (PCA), which is particularly useful in explaining
the results of anomaly detection based on PCA. Because features are usually
correlated when PCA-based anomaly detection is applied, care must be taken in
computing a value function for the Shapley values. We utilize the probabilistic
view of PCA, particularly its conditional distribution, to exactly compute a
value function for the Shapely values. We also present numerical examples,
which imply that the Shapley values are advantageous for explaining detected
anomalies than raw reconstruction errors of each feature.
| stat |
A Simultaneous Inference Procedure to Identify Subgroups from RCTs with
Survival Outcomes: Application to Analysis of AMD Progression Studies | With the uptake of targeted therapies, instead of the "one-fits-all"
approach, modern randomized clinical trials (RCTs) often aim to develop
treatments that target a subgroup of patients. Motivated by analyzing the
Age-Related Eye Disease Study (AREDS) data, a large RCT to study the efficacy
of nutritional supplements in delaying the progression of an eye disease,
age-related macular degeneration (AMD), we develop a simultaneous inference
procedure to identify and infer subgroups with differential treatment efficacy
in RCTs with survival outcome. Specifically, we formulate the multiple testing
problem through contrasts and construct their simultaneous confidence
intervals, which control both within- and across- marker multiplicity
appropriately. Realistic simulations are conducted using real genotype data to
evaluate the method performance under various scenarios. The method is then
applied to AREDS to assess the efficacy of antioxidants and zinc combination in
delaying AMD progression. Multiple gene regions including ESRRB-VASH1 on
chromosome 14 have been identified with subgroups showing differential
efficacy. We further validate our findings in an independent subsequent RCT,
AREDS2, by discovering consistent differential treatment responses in the
targeted and non-targeted subgroups been identified from AREDS. This
simultaneous inference approach provides a step forward to confidently identify
and infer subgroups in modern drug development.
| stat |
Quantifying uncertainty and dynamical changes in multi-species fishing
mortality rates, catches and biomass by combining state-space and mechanistic
multi-species models | In marine management, fish stocks are often managed on a stock-by-stock basis
using single-species models. Many of these models are based upon statistical
techniques and are good at assessing the current state and making short-term
predictions; however, as they do not model interactions between stocks, they
lack predictive power on longer timescales. Additionally, there are mechanistic
multi-species models that represent key biological processes and consider
interactions between stocks such as predation and competition for resources.
Due to the complexity of these models, they are difficult to fit to data, and
so many mechanistic multi-species models depend upon single-species models
where they exist, or ad hoc assumptions when they don't, for parameters such as
annual fishing mortality. In this paper we demonstrate that by taking a
state-space approach, many of the uncertain parameters can be treated
dynamically, allowing us to fit, with quantifiable uncertainty, mechanistic
multi-species models directly to data. We demonstrate this by fitting uncertain
parameters, including annual fishing mortality, of a size-based multi-species
model of the Celtic Sea, for species with and without single-species
stock-assessments. Consequently, errors in the single-species models no longer
propagate through the multi-species model and underlying assumptions are more
transparent. Building mechanistic multi-species models that are internally
consistent, with quantifiable uncertainty, will improve their credibility and
utility for management. This may lead to their uptake by being either used to
corroborate single-species models; directly in the advice process to make
predictions into the future; or used to provide a new way of managing
data-limited stocks.
| stat |
Neural Networks as Functional Classifiers | In recent years, there has been considerable innovation in the world of
predictive methodologies. This is evident by the relative domination of machine
learning approaches in various classification competitions. While these
algorithms have excelled at multivariate problems, they have remained dormant
in the realm of functional data analysis. We extend notable deep learning
methodologies to the domain of functional data for the purpose of
classification problems. We highlight the effectiveness of our method in a
number of classification applications such as classification of spectrographic
data. Moreover, we demonstrate the performance of our classifier through
simulation studies in which we compare our approach to the functional linear
model and other conventional classification methods.
| stat |
Targeting Learning: Robust Statistics for Reproducible Research | Targeted Learning is a subfield of statistics that unifies advances in causal
inference, machine learning and statistical theory to help answer
scientifically impactful questions with statistical confidence. Targeted
Learning is driven by complex problems in data science and has been implemented
in a diversity of real-world scenarios: observational studies with missing
treatments and outcomes, personalized interventions, longitudinal settings with
time-varying treatment regimes, survival analysis, adaptive randomized trials,
mediation analysis, and networks of connected subjects. In contrast to the
(mis)application of restrictive modeling strategies that dominate the current
practice of statistics, Targeted Learning establishes a principled standard for
statistical estimation and inference (i.e., confidence intervals and p-values).
This multiply robust approach is accompanied by a guiding roadmap and a
burgeoning software ecosystem, both of which provide guidance on the
construction of estimators optimized to best answer the motivating question.
The roadmap of Targeted Learning emphasizes tailoring statistical procedures so
as to minimize their assumptions, carefully grounding them only in the
scientific knowledge available. The end result is a framework that honestly
reflects the uncertainty in both the background knowledge and the available
data in order to draw reliable conclusions from statistical analyses -
ultimately enhancing the reproducibility and rigor of scientific findings.
| stat |
An Alternative Formulation of Coxian Phase-type Distributions with
Covariates: Application to Emergency Department Length of Stay | In this paper we present a new methodology to model patient transitions and
length of stay in the emergency department using a series of conditional Coxian
phase-type distributions, with covariates. We reformulate the Coxian models
(standard Coxian, Coxian with multiple absorbing states, joint Coxian, and
conditional Coxian) to take into account heterogeneity in patient
characteristics such as arrival mode, time of admission and age. The approach
differs from previous research in that it reduces the computational time, and
it allows the inclusion of patient covariate information directly into the
model. The model is applied to emergency department data from University
Hospital Limerick in Ireland.
| stat |
Enhanced Balancing of Bias-Variance Tradeoff in Stochastic Estimation: A
Minimax Perspective | Biased stochastic estimators, such as finite-differences for noisy gradient
estimation, often contain parameters that need to be properly chosen to balance
impacts from the bias and the variance. While the optimal order of these
parameters in terms of the simulation budget can be readily established, the
precise best values depend on model characteristics that are typically unknown
in advance. We introduce a framework to construct new classes of estimators,
based on judicious combinations of simulation runs on sequences of tuning
parameter values, such that the estimators consistently outperform a given
tuning parameter choice in the conventional approach, regardless of the unknown
model characteristics. We argue the outperformance via what we call the
asymptotic minimax risk ratio, obtained by minimizing the worst-case asymptotic
ratio between the mean square errors of our estimators and the conventional
one, where the worst case is over any possible values of the model unknowns. In
particular, when the minimax ratio is less than 1, the calibrated estimator is
guaranteed to perform better asymptotically. We identify this minimax ratio for
general classes of weighted estimators, and the regimes where this ratio is
less than 1. Moreover, we show that the best weighting scheme is characterized
by a sum of two components with distinct decay rates. We explain how this
arises from bias-variance balancing that combats the adversarial selection of
the model constants, which can be analyzed via a tractable reformulation of a
non-convex optimization problem.
| stat |
Validated Variational Inference via Practical Posterior Error Bounds | Variational inference has become an increasingly attractive fast alternative
to Markov chain Monte Carlo methods for approximate Bayesian inference.
However, a major obstacle to the widespread use of variational methods is the
lack of post-hoc accuracy measures that are both theoretically justified and
computationally efficient. In this paper, we provide rigorous bounds on the
error of posterior mean and uncertainty estimates that arise from
full-distribution approximations, as in variational inference. Our bounds are
widely applicable, as they require only that the approximating and exact
posteriors have polynomial moments. Our bounds are also computationally
efficient for variational inference because they require only standard values
from variational objectives, straightforward analytic calculations, and simple
Monte Carlo estimates. We show that our analysis naturally leads to a new and
improved workflow for validated variational inference. Finally, we demonstrate
the utility of our proposed workflow and error bounds on a robust regression
problem and on a real-data example with a widely used multilevel hierarchical
model.
| stat |
Panel data analysis via mechanistic models | Panel data, also known as longitudinal data, consist of a collection of time
series. Each time series, which could itself be multivariate, comprises a
sequence of measurements taken on a distinct unit. Mechanistic modeling
involves writing down scientifically motivated equations describing the
collection of dynamic systems giving rise to the observations on each unit. A
defining characteristic of panel systems is that the dynamic interaction
between units should be negligible. Panel models therefore consist of a
collection of independent stochastic processes, generally linked through shared
parameters while also having unit-specific parameters. To give the scientist
flexibility in model specification, we are motivated to develop a framework for
inference on panel data permitting the consideration of arbitrary nonlinear,
partially observed panel models. We build on iterated filtering techniques that
provide likelihood-based inference on nonlinear partially observed Markov
process models for time series data. Our methodology depends on the latent
Markov process only through simulation; this plug-and-play property ensures
applicability to a large class of models. We demonstrate our methodology on a
toy example and two epidemiological case studies. We address inferential and
computational issues arising due to the combination of model complexity and
dataset size.
| stat |
Almost Politically Acceptable Criminal Justice Risk Assessment | In criminal justice risk forecasting, one can prove that it is impossible to
optimize accuracy and fairness at the same time. One can also prove that it is
impossible optimize at once all of the usual group definitions of fairness. In
the policy arena, one is left with tradeoffs about which many stakeholders will
adamantly disagree. In this paper, we offer a different approach. We do not
seek perfectly accurate and fair risk assessments. We seek politically
acceptable risk assessments. We describe and apply to data on 300,000 offenders
a machine learning approach that responds to many of the most visible charges
of "racial bias." Regardless of whether such claims are true, we adjust our
procedures to compensate. We begin by training the algorithm on White offenders
only and computing risk with test data separately for White offenders and Black
offenders. Thus, the fitted algorithm structure is exactly the same for both
groups; the algorithm treats all offenders as if they are White. But because
White and Black offenders can bring different predictors distributions to the
white-trained algorithm, we provide additional adjustments if needed. Insofar
are conventional machine learning procedures do not produce accuracy and
fairness that some stakeholders require, it is possible to alter conventional
practice to respond explicitly to many salient stakeholder claims even if they
are unsupported by the facts. The results can be a politically acceptable risk
assessment tools.
| stat |
Fused Spatial Point Process IntensityEstimation on Complex Domains | The availability of large spatial data geocoded at accurate locations has
fueled a growing interest in spatial modeling and analysis of point processes.
The proposed research is motivated by the intensity estimation problem for
large spatial point patterns on complex domains, where many existing spatial
point process models suffer from the problems of "leakage" and computation. We
propose an efficient intensity estimation algorithm to estimate the spatially
varying intensity function and to study the varying relationship between
intensity and explanatory variables on complex domains. The method is built
upon a graph regularization technique and hence can be flexibly applied to
point patterns on complex domains such as regions with irregular boundaries and
holes, or linear networks. An efficient proximal gradient optimization
algorithm is proposed to handle large spatial point patterns. We also derive
the asymptotic error bound for the proposed estimator. Numerical studies are
conducted to illustrate the performance of the method. Finally, We apply the
method to study and visualize the intensity patterns of the accidents on the
Western Australia road network, and the spatial variations in the effects of
income, lights condition, and population density on the Toronto homicides
occurrences.
| stat |
Distributional Regression Forests for Probabilistic Precipitation
Forecasting in Complex Terrain | To obtain a probabilistic model for a dependent variable based on some set of
explanatory variables, a distributional approach is often adopted where the
parameters of the distribution are linked to regressors. In many classical
models this only captures the location of the distribution but over the last
decade there has been increasing interest in distributional regression
approaches modeling all parameters including location, scale, and shape.
Notably, so-called non-homogeneous Gaussian regression (NGR) models both mean
and variance of a Gaussian response and is particularly popular in weather
forecasting. Moreover, generalized additive models for location, scale, and
shape (GAMLSS) provide a framework where each distribution parameter is modeled
separately capturing smooth linear or nonlinear effects. However, when variable
selection is required and/or there are non-smooth dependencies or interactions
(especially unknown or of high-order), it is challenging to establish a good
GAMLSS. A natural alternative in these situations would be the application of
regression trees or random forests but, so far, no general distributional
framework is available for these. Therefore, a framework for distributional
regression trees and forests is proposed that blends regression trees and
random forests with classical distributions from the GAMLSS framework as well
as their censored or truncated counterparts. To illustrate these novel
approaches in practice, they are employed to obtain probabilistic precipitation
forecasts at numerous sites in a mountainous region based on a large number of
numerical weather prediction quantities. It is shown that the novel
distributional regression forests automatically select variables and
interactions, performing on par or often even better than GAMLSS specified
either through prior meteorological knowledge or a computationally more
demanding boosting approach.
| stat |
Polling bias and undecided voter allocations: US Presidential elections,
2004 - 2016 | Accounting for undecided and uncertain voters is a challenging issue for
predicting election results from public opinion polls. Undecided voters typify
the uncertainty of swing voters in polls but are often ignored or allocated to
each candidate in a simple, deterministic manner. Historically this may have
been adequate because the undecided were comparatively small enough to assume
that they do not affect the relative proportions of the decided voters.
However, in the presence of high numbers of undecided voters, these static
rules may in fact bias election predictions from election poll authors and
meta-poll analysts. In this paper, we examine the effect of undecided voters in
the 2016 US presidential election to the previous three presidential elections.
We show there were a relatively high number of undecided voters over the
campaign and on election day, and that the allocation of undecided voters in
this election was not consistent with two-party proportional (or even)
allocations. We find evidence that static allocation regimes are inadequate for
election prediction models and that probabilistic allocations may be superior.
We also estimate the bias attributable to polling agencies, often referred to
as "house effects".
| stat |
Database Alignment with Gaussian Features | We consider the problem of aligning a pair of databases with jointly Gaussian
features. We consider two algorithms, complete database alignment via MAP
estimation among all possible database alignments, and partial alignment via a
thresholding approach of log likelihood ratios. We derive conditions on mutual
information between feature pairs, identifying the regimes where the algorithms
are guaranteed to perform reliably and those where they cannot be expected to
succeed.
| stat |
On the Art and Science of Machine Learning Explanations | This text discusses several popular explanatory methods that go beyond the
error measurements and plots traditionally used to assess machine learning
models. Some of the explanatory methods are accepted tools of the trade while
others are rigorously derived and backed by long-standing theory. The methods,
decision tree surrogate models, individual conditional expectation (ICE) plots,
local interpretable model-agnostic explanations (LIME), partial dependence
plots, and Shapley explanations, vary in terms of scope, fidelity, and suitable
application domain. Along with descriptions of these methods, this text
presents real-world usage recommendations supported by a use case and public,
in-depth software examples for reproducibility.
| stat |
Simultaneous confidence bands for nonparametric regression with
partially missing covariates | In this paper, we consider a weighted local linear estimator based on the
inverse selection probability for nonparametric regression with missing
covariates at random. The asymptotic distribution of the maximal deviation
between the estimator and the true regression function is derived and an
asymptotically accurate simultaneous confidence band is constructed. The
estimator for the regression function is shown to be oracally efficient in the
sense that it is uniformly indistinguishable from that when the selection
probabilities are known. Finite sample performance is examined via simulation
studies which support our asymptotic theory. The proposed method is
demonstrated via an analysis of a data set from the Canada 2010/2011 Youth
Student Survey.
| stat |
Extreme Classification via Adversarial Softmax Approximation | Training a classifier over a large number of classes, known as 'extreme
classification', has become a topic of major interest with applications in
technology, science, and e-commerce. Traditional softmax regression induces a
gradient cost proportional to the number of classes $C$, which often is
prohibitively expensive. A popular scalable softmax approximation relies on
uniform negative sampling, which suffers from slow convergence due a poor
signal-to-noise ratio. In this paper, we propose a simple training method for
drastically enhancing the gradient signal by drawing negative samples from an
adversarial model that mimics the data distribution. Our contributions are
three-fold: (i) an adversarial sampling mechanism that produces negative
samples at a cost only logarithmic in $C$, thus still resulting in cheap
gradient updates; (ii) a mathematical proof that this adversarial sampling
minimizes the gradient variance while any bias due to non-uniform sampling can
be removed; (iii) experimental results on large scale data sets that show a
reduction of the training time by an order of magnitude relative to several
competitive baselines.
| stat |
An efficient surrogate-aided importance sampling framework for
reliability analysis | Surrogates in lieu of expensive-to-evaluate performance functions can
accelerate the reliability analysis greatly. This paper proposes a new
two-stage framework for surrogate-aided reliability analysis named Surrogates
for Importance Sampling (S4IS). In the first stage, a coarse surrogate is built
to gain the information about failure regions; the second stage zooms into the
important regions and improves the accuracy of the failure probability
estimator by adaptively selecting support points therein. The learning
functions are proposed to guide the selection of support points such that the
exploration and exploitation can be dynamically balanced. As a generic
framework, S4IS has the potential to incorporate different types of surrogates
(Gaussian Processes, Support Vector Machines, Neural Network, etc.). The
effectiveness and efficiency of S4IS is validated by five illustrative
examples, which involve system reliability, highly nonlinear limit-state
function, small failure probability and moderately high dimensionality. The
implementation of S4IS is made available to download at
https://github.com/RobinSeaside/S4IS.
| stat |
Robust Identification of Gene-Environment Interactions under
High-Dimensional Accelerated Failure Time Models | For complex diseases, beyond the main effects of genetic (G) and
environmental (E) factors, gene-environment (G-E) interactions also play an
important role. Many of the existing G-E interaction methods conduct marginal
analysis, which may not appropriately describe disease biology. Joint analysis
methods have been developed, with most of the existing loss functions
constructed based on likelihood. In practice, data contamination is not
uncommon. Development of robust methods for interaction analysis that can
accommodate data contamination is very limited. In this study, we consider
censored survival data and adopt an accelerated failure time (AFT) model. An
exponential squared loss is adopted to achieve robustness. A sparse group
penalization approach, which respects the "main effects, interactions"
hierarchy, is adopted for estimation and identification. Consistency properties
are rigorously established. Simulation shows that the proposed method
outperforms direct competitors. In data analysis, the proposed method makes
biologically sensible findings.
| stat |
Profile regression for subgrouping patients with early stage Parkinson's
disease | Falls are detrimental to people with Parkinson's Disease (PD) because of the
potentially severe consequences to the patients' quality of life. While many
studies have attempted to predict falls/non-falls, this study aimed to
determine factors related to falls frequency in people with early PD. Ninety
nine participants with early stage PD were assessed based on two types of
tests. The first type of tests is disease-specific tests, comprised of the
Unified Parkinson's Disease Rating Scale (UPDRS) and the Schwab and England
activities of daily living scale (SEADL). A measure of postural instability and
gait disorder (PIGD) and subtotal scores for subscales I, II, and III were
derived from the UPDRS. The second type of tests is functional tests, including
Tinetti gait and balance, Berg Balance Scale (BBS), Timed-Up and Go (TUG),
Functional Reach (FR), Freezing of Gait (FOG), Mini Mental State Examination
(MMSE), and Melbourne Edge Test (MET). Falls were recorded each month for 6
months. Clustering of patients via Finite Mixture Model (FMM) was conducted.
Three clusters of patients were found: non-or single-fallers, low frequency
fallers, and high frequency fallers. Several factors that are important to
clustering PD patients were identified: UPDRS subscales II and III subtotals,
PIGD and SE ADL. However these factors could not differentiate PD patients with
low frequency fallers from high frequency fallers. While Tinetti,TUG, and BBS
turned to be important factors in clustering PD patients, and could
differentiate the three clusters. FMM is able to cluster people with PD into
three groups. We obtain several factors important to explaining the clusters
and also found different role of disease specific measures and functional tests
to clustering PD patients. Upon examining these measures, it might be possible
to develop new disease treatment to prevent, or to delay, the occurrence of
falls.
| stat |
Adaptive Variable Selection for Sequential Prediction in Multivariate
Dynamic Models | We discuss Bayesian model uncertainty analysis and forecasting in sequential
dynamic modeling of multivariate time series. The perspective is that of a
decision-maker with a specific forecasting objective that guides thinking about
relevant models. Based on formal Bayesian decision-theoretic reasoning, we
develop a time-adaptive approach to exploring, weighting, combining and
selecting models that differ in terms of predictive variables included. The
adaptivity allows for changes in the sets of favored models over time, and is
guided by the specific forecasting goals. A synthetic example illustrates how
decision-guided variable selection differs from traditional Bayesian model
uncertainty analysis and standard model averaging. An applied study in one
motivating application of long-term macroeconomic forecasting highlights the
utility of the new approach in terms of improving predictions as well as its
ability to identify and interpret different sets of relevant models over time
with respect to specific, defined forecasting goals.
| stat |
Fairness Evaluation in Presence of Biased Noisy Labels | Risk assessment tools are widely used around the country to inform decision
making within the criminal justice system. Recently, considerable attention has
been devoted to the question of whether such tools may suffer from racial bias.
In this type of assessment, a fundamental issue is that the training and
evaluation of the model is based on a variable (arrest) that may represent a
noisy version of an unobserved outcome of more central interest (offense). We
propose a sensitivity analysis framework for assessing how assumptions on the
noise across groups affect the predictive bias properties of the risk
assessment model as a predictor of reoffense. Our experimental results on two
real world criminal justice data sets demonstrate how even small biases in the
observed labels may call into question the conclusions of an analysis based on
the noisy outcome.
| stat |
Consistent detection and optimal localization of all detectable change
points in piecewise stationary arbitrarily sparse network-sequences | We consider the offline change point detection and localization problem in
the context of piecewise stationary networks, where the observable is a finite
sequence of networks. We develop algorithms involving some suitably modified
CUSUM statistics based on adaptively trimmed adjacency matrices of the observed
networks for both detection and localization of single or multiple change
points present in the input data. We provide rigorous theoretical analysis and
finite sample estimates evaluating the performance of the proposed methods when
the input (finite sequence of networks) is generated from an inhomogeneous
random graph model, where the change points are characterized by the change in
the mean adjacency matrix. We show that the proposed algorithms can detect
(resp. localize) all change points, where the change in the expected adjacency
matrix is above the minimax detectability (resp. localizability) threshold,
consistently without any a priori assumption about (a) a lower bound for the
sparsity of the underlying networks, (b) an upper bound for the number of
change points, and (c) a lower bound for the separation between successive
change points, provided either the minimum separation between successive pairs
of change points or the average degree of the underlying networks goes to
infinity arbitrarily slowly. We also prove that the above condition is
necessary to have consistency.
| stat |
Monte Carlo Sampling Bias in the Microwave Uncertainty Framework | Uncertainty propagation software can have unknown, inadvertent biases
introduced by various means. This work is a case study in bias identification
and reduction in one such software package, the Microwave Uncertainty Framework
(MUF). The general purpose of the MUF is to provide automated multivariate
statistical uncertainty propagation and analysis on a Monte Carlo (MC) basis.
Combine is a key module in the MUF, responsible for merging data, raw or
transformed, to accurately reflect the variability in the data and in its
central tendency. In this work the performance of Combine's MC replicates is
analytically compared against its stated design goals. An alternative
construction is proposed for Combine's MC replicates and its performance is
compared, too, against Combine's design goals. These comparisons are made
within an archetypal two-stage scenario in which received data are first
transformed in conjunction with shared systematic error and then combined to
produce summary information. These comparisons reveal the limited conditions
under which Combine's uncertainty results are unbiased and the extent of these
biases when these conditions are not met. For small MC sample sizes neither
construction, current or alternative, fully meets Combine's design goals, nor
does either construction consistently outperform the other. However, for large
MC sample sizes the bias in the proposed alternative construction is
asymptotically zero, and this construction is recommended.
| stat |
Latent Multivariate Log-Gamma Models for High-Dimensional Multi-Type
Responses with Application to Daily Fine Particulate Matter and Mortality
Counts | Tracking and estimating Daily Fine Particulate Matter (PM2.5) is very
important as it has been shown that PM2.5 is directly related to mortality
related to lungs, cardiovascular system, and stroke. That is, high values of
PM2.5 constitute a public health problem in the US, and it is important that we
precisely estimate PM2.5 to aid in public policy decisions. Thus, we propose a
Bayesian hierarchical model for high-dimensional "multi-type" responses. By
"multi-type" responses we mean a collection of correlated responses that have
different distributional assumptions (e.g., continuous skewed observations, and
count-valued observations). The Centers for Disease Control and Prevention
(CDC) database provides counts of mortalities related to PM2.5 and daily
averaged PM2.5 which are both treated as responses in our analysis. Our model
capitalizes on the shared conjugate structure between the Weibull (to model
PM2.5), Poisson (to model diseases mortalities), and multivariate log-gamma
distributions, and we use dimension reduction to aid with computation. Our
model can also be used to improve the precision of estimates and estimate
values at undisclosed/missing counties. We provide a simulation study to
illustrate the performance of the model, and give an in-depth analysis of the
CDC dataset.
| stat |
Estimation of marriage incidence rates by combining two cross-sectional
retrospective designs: Event history analysis of two dependent processes | The aim of this work is to develop methods for studying the determinants of
marriage incidence using marriage histories collected under two different types
of retrospective cross-sectional study designs. These designs are: sampling of
ever married women before the cross-section, a prevalent cohort, and sampling
of women irrespective of marital status, a general cross-sectional cohort.
While retrospective histories from a prevalent cohort do not identify incidence
rates without parametric modelling assumptions, the rates can be identified
when combined with data from a general cohort. Moreover, education, a strong
endogenous covariate, and marriage processes are correlated. Hence, they need
to be modelled jointly in order to estimate the marriage incidence. For this
purpose, we specify a multi-state model and propose a likelihood-based
estimation method. We outline the assumptions under which a likelihood
expression involving only marriage incidence parameters can be derived. This is
of particular interest when either retrospective education histories are not
available or related parameters are not of interest. Our simulation results
confirm the gain in efficiency by combining data from the two designs, while
demonstrating how the parameter estimates are affected by violations of the
assumptions used in deriving the simplified likelihood expressions. Two Indian
National Family Health Surveys are used as motivation for the methodological
development and to demonstrate the application of the methods.
| stat |
Discussion of 'Detecting possibly frequent change-points: Wild Binary
Segmentation 2 and steepest-drop model selection' | We discuss the theoretical guarantee provided by the WBS2.SDLL proposed in
Fryzlewicz (2020) and explore an alternative, MOSUM-based candidate generation
method for the SDLL.
| stat |
Stein Variational Gradient Descent With Matrix-Valued Kernels | Stein variational gradient descent (SVGD) is a particle-based inference
algorithm that leverages gradient information for efficient approximate
inference. In this work, we enhance SVGD by leveraging preconditioning
matrices, such as the Hessian and Fisher information matrix, to incorporate
geometric information into SVGD updates. We achieve this by presenting a
generalization of SVGD that replaces the scalar-valued kernels in vanilla SVGD
with more general matrix-valued kernels. This yields a significant extension of
SVGD, and more importantly, allows us to flexibly incorporate various
preconditioning matrices to accelerate the exploration in the probability
landscape. Empirical results show that our method outperforms vanilla SVGD and
a variety of baseline approaches over a range of real-world Bayesian inference
tasks.
| stat |
Finite mixture modeling of censored and missing data using the
multivariate skew-normal distribution | Finite mixture models have been widely used to model and analyze data from a
heterogeneous populations. Moreover, data of this kind can be missing or
subject to some upper and/or lower detection limits because of the restriction
of experimental apparatuses. Another complication arises when measures of each
population depart significantly from normality, for instance, asymmetric
behavior. For such data structures, we propose a robust model for censored
and/or missing data based on finite mixtures of multivariate skew-normal
distributions. This approach allows us to model data with great flexibility,
accommodating multimodality and skewness, simultaneously, depending on the
structure of the mixture components. We develop an analytically simple, yet
efficient, EM- type algorithm for conducting maximum likelihood estimation of
the parameters. The algorithm has closed-form expressions at the E-step that
rely on formulas for the mean and variance of the truncated multivariate
skew-normal distributions. Furthermore, a general information-based method for
approximating the asymptotic covariance matrix of the estimators is also
presented. Results obtained from the analysis of both simulated and real
datasets are reported to demonstrate the effectiveness of the proposed method.
The proposed algorithm and method are implemented in the new R package CensMFM.
| stat |
Learning binary undirected graph in low dimensional regime | Given a random sample extracted from a Multivariate Bernoulli Variable (MBV),
we consider the problem of estimating the structure of the undirected graph for
which the distribution is pairwise Markov and the parameters' vector of its
exponential form. We propose a simple method that provides a closed form
estimator of the parameters' vector and through its support also provides an
estimate of the undirected graph associated to the MBV distribution. The
estimator is proved to be consistent but it is feasible only in low-dimensional
regimes. Synthetic examples illustrates its performance compared with another
method that represents the state of the art in literature. Finally, the
proposed procedure is used for the analysis of a real data set in the pediatric
allergology area showing its practical efficiency.
| stat |
Statistical methods for linking geostatistical maps and transmission
models: Application to lymphatic filariasis in East Africa | Infectious diseases remain one of the major causes of human mortality and
suffering. Mathematical models have been established as an important tool for
capturing the features that drive the spread of the disease, predicting the
progression of an epidemic and hence guiding the development of strategies to
control it. Another important area of epidemiological interest is the
development of geostatistical methods for the analysis of data from spatially
referenced prevalence surveys. Maps of prevalence are useful, not only for
enabling a more precise disease risk stratification, but also for guiding the
planning of more reliable spatial control programmes by identifying affected
areas. Despite the methodological advances that have been made in each area
independently, efforts to link transmission models and geostatistical maps have
been limited. Motivated by this fact, we developed a Bayesian approach that
combines fine-scale geostatistical maps of disease prevalence with transmission
models to provide quantitative, spatially explicit projections of the current
and future impact of control programs against a disease. These estimates can
then be used at a local level to identify the effectiveness of suggested
intervention schemes and allow investigation of alternative strategies. The
methodology has been applied to lymphatic filariasis in East Africa to provide
estimates of the impact of different intervention strategies against the
disease.
| stat |
Estimation of Causal Effects in the Presence of Unobserved Confounding
in the Alzheimer's Continuum | Studying the relationship between neuroanatomy and cognitive decline due to
Alzheimer's has been a major research focus in the last decade. However, to
infer cause-effect relationships rather than simple associations from
observational data, we need to (i) express the causal relationships leading to
cognitive decline in a graphical model, and (ii) ensure the causal effect of
interest is identifiable from the collected data. We derive a causal graph from
the current clinical knowledge on cause and effect in the Alzheimer's disease
continuum, and show that identifiability of the causal effect requires all
confounders to be known and measured. However, in complex neuroimaging studies,
we neither know all potential confounders nor do we have data on them. To
alleviate this requirement, we leverage the dependencies among multiple causes
by deriving a substitute confounder via a probabilistic latent factor model. In
our theoretical analysis, we prove that using the substitute confounder enables
identifiability of the causal effect of neuroanatomy on cognition. We
quantitatively evaluate the effectiveness of our approach on semi-synthetic
data, where we know the true causal effects, and illustrate its use on real
data on the Alzheimer's disease continuum, where it reveals important causes
that otherwise would have been missed.
| stat |
Robust Classification via Support Vector Machines | The loss function choice for any Support Vector Machine classifier has raised
great interest in the literature due to the lack of robustness of the Hinge
loss, which is the standard loss choice. In this paper, we plan to robustify
the binary classifier by maintaining the overall advantages of the Hinge loss,
rather than modifying this standard choice. We propose two robust classifiers
under data uncertainty. The first is called Single Perturbation SVM (SP-SVM)
and provides a constructive method by allowing a controlled perturbation to one
feature of the data. The second method is called Extreme Empirical Loss SVM
(EEL-SVM) and is based on a new empirical loss estimate, namely, the Extreme
Empirical Loss (EEL), that puts more emphasis on extreme violations of the
classification hyper-plane, rather than taking the usual sample average with
equal importance for all hyper-plane violations. Extensive numerical
investigation reveals the advantages of the two robust classifiers on simulated
data and well-known real datasets.
| stat |
Evaluation of a meta-analysis of ambient air quality as a risk factor
for asthma exacerbation | False-positive results and bias may be common features of the biomedical
literature today, including risk factor-chronic disease research. A study was
undertaken to assess the reliability of base studies used in a meta-analysis
examining whether carbon monoxide, particulate matter 10 and 2.5 micro molar,
sulfur dioxide, nitrogen dioxide and ozone are risk factors for asthma
exacerbation (hospital admission and emergency room visits for asthma attack).
The number of statistical tests and models were counted in 17 randomly selected
base papers from 87 used in the meta-analysis. P-value plots for each air
component were constructed to evaluate the effect heterogeneity of p-values
used from all 87 base papers The number of statistical tests possible in the 17
selected base papers was large, median=15,360 (interquartile range=1,536 to
40,960), in comparison to results presented. Each p-value plot showed a
two-component mixture with small p-values less than .001 while other p-values
appeared random (p-values greater than .05). Given potentially large numbers of
statistical tests conducted in the 17 selected base papers, p-hacking cannot be
ruled out as explanations for small p-values. Our interpretation of the
meta-analysis is that the random p-values indicating null associations are more
plausible and that the meta-analysis will not likely replicate in the absence
of bias. We conclude the meta-analysis and base papers used are unreliable and
do not offer evidence of value to inform public health practitioners about air
quality as a risk factor for asthma exacerbation. The following areas are
crucial for enabling improvements in risk factor chronic disease observational
studies at the funding agency and journal level: preregistration, changes in
funding agency and journal editor (and reviewer) practices, open sharing of
data and facilitation of reproducibility research.
| stat |
Composite mixture of log-linear models for categorical data | Multivariate categorical data are routinely collected in many application
areas. As the number of cells in the table grows exponentially with the number
of variables, many or even most cells will contain zero observations. This
severe sparsity motivates appropriate statistical methodologies that
effectively reduce the number of free parameters, with penalized log-linear
models and latent structure analysis being popular options. This article
proposes a fundamentally new class of methods, which we refer to as Mixture of
Log Linear models (mills). Combining latent class analysis and log-linear
models, mills defines a novel Bayesian methodology to model complex
multivariate categorical with flexibility and interpretability. Mills is shown
to have key advantages over alternative methods for contingency tables in
simulations and an application investigating the relation among suicide
attempts and empathy.
| stat |
Understanding Black-box Predictions via Influence Functions | How can we explain the predictions of a black-box model? In this paper, we
use influence functions -- a classic technique from robust statistics -- to
trace a model's prediction through the learning algorithm and back to its
training data, thereby identifying training points most responsible for a given
prediction. To scale up influence functions to modern machine learning
settings, we develop a simple, efficient implementation that requires only
oracle access to gradients and Hessian-vector products. We show that even on
non-convex and non-differentiable models where the theory breaks down,
approximations to influence functions can still provide valuable information.
On linear models and convolutional neural networks, we demonstrate that
influence functions are useful for multiple purposes: understanding model
behavior, debugging models, detecting dataset errors, and even creating
visually-indistinguishable training-set attacks.
| stat |
A comparative study of counterfactual estimators | We provide a comparative study of several widely used off-policy estimators
(Empirical Average, Basic Importance Sampling and Normalized Importance
Sampling), detailing the different regimes where they are individually
suboptimal. We then exhibit properties optimal estimators should possess. In
the case where examples have been gathered using multiple policies, we show
that fused estimators dominate basic ones but can still be improved.
| stat |
Spatial Weibull Regression with Multivariate Log Gamma Process and Its
Applications to China Earthquake Economic Loss | Bayesian spatial modeling of heavy-tailed distributions has become
increasingly popular in various areas of science in recent decades. We propose
a Weibull regression model with spatial random effects for analyzing extreme
economic loss. Model estimation is facilitated by a computationally efficient
Bayesian sampling algorithm utilizing the multivariate Log-Gamma distribution.
Simulation studies are carried out to demonstrate better empirical performances
of the proposed model than the generalized linear mixed effects model. An
earthquake data obtained from Yunnan Seismological Bureau, China is analyzed.
Logarithm of the Pseudo-marginal likelihood values are obtained to select the
optimal model, and Value-at-risk, expected shortfall, and tail-value-at-risk
based on posterior predictive distribution of the optimal model are calculated
under different confidence levels.
| stat |
Detection of Change Points in Piecewise Polynomial Signals Using Trend
Filtering | While many approaches have been proposed for discovering abrupt changes in
piecewise constant signals, few methods are available to capture these changes
in piecewise polynomial signals. In this paper, we propose a change point
detection method, PRUTF, based on trend filtering. By providing a comprehensive
dual solution path for trend filtering, PRUTF allows us to discover change
points of the underlying signal for either a given value of the regularization
parameter or a specific number of steps of the algorithm. We demonstrate that
the dual solution path constitutes a Gaussian bridge process that enables us to
derive an exact and efficient stopping rule for terminating the search
algorithm. We also prove that the estimates produced by this algorithm are
asymptotically consistent in pattern recovery. This result holds even in the
case of staircases (consecutive change points of the same sign) in the signal.
Finally, we investigate the performance of our proposed method for various
signals and then compare its performance against some state-of-the-art methods
in the context of change point detection. We apply our method to three
real-world datasets including the UK House Price Index (HPI), the GISS surface
Temperature Analysis (GISTEMP) and the Coronavirus disease (COVID-19) pandemic.
| stat |
Deep Geometric Functional Maps: Robust Feature Learning for Shape
Correspondence | We present a novel learning-based approach for computing correspondences
between non-rigid 3D shapes. Unlike previous methods that either require
extensive training data or operate on handcrafted input descriptors and thus
generalize poorly across diverse datasets, our approach is both accurate and
robust to changes in shape structure. Key to our method is a feature-extraction
network that learns directly from raw shape geometry, combined with a novel
regularized map extraction layer and loss, based on the functional map
representation. We demonstrate through extensive experiments in challenging
shape matching scenarios that our method can learn from less training data than
existing supervised approaches and generalizes significantly better than
current descriptor-based learning methods. Our source code is available at:
https://github.com/LIX-shape-analysis/GeomFmaps.
| stat |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.