title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Unbiased estimators for random design regression | In linear regression we wish to estimate the optimum linear least squares
predictor for a distribution over d-dimensional input points and real-valued
responses, based on a small sample. Under standard random design analysis,
where the sample is drawn i.i.d. from the input distribution, the least squares
solution for that sample can be viewed as the natural estimator of the optimum.
Unfortunately, this estimator almost always incurs an undesirable bias coming
from the randomness of the input points. In this paper we show that it is
possible to draw a non-i.i.d. sample of input points such that, regardless of
the response model, the least squares solution is an unbiased estimator of the
optimum. Moreover, this sample can be produced efficiently by augmenting a
previously drawn i.i.d. sample with an additional set of d points drawn jointly
from the input distribution rescaled by the squared volume spanned by the
points. Motivated by this, we develop a theoretical framework for studying
volume-rescaled sampling, and in the process prove a number of new matrix
expectation identities. We use them to show that for any input distribution and
$\epsilon>0$ there is a random design consisting of $O(d\log d+ d/\epsilon)$
points from which an unbiased estimator can be constructed whose square loss
over the entire distribution is with high probability bounded by $1+\epsilon$
times the loss of the optimum. We provide efficient algorithms for generating
such unbiased estimators in a number of practical settings and support our
claims experimentally.
| stat |
Regularized Estimation and Feature Selection in Mixtures of
Gaussian-Gated Experts Models | Mixtures-of-Experts models and their maximum likelihood estimation (MLE) via
the EM algorithm have been thoroughly studied in the statistics and machine
learning literature. They are subject of a growing investigation in the context
of modeling with high-dimensional predictors with regularized MLE. We examine
MoE with Gaussian gating network, for clustering and regression, and propose an
$\ell_1$-regularized MLE to encourage sparse models and deal with the
high-dimensional setting. We develop an EM-Lasso algorithm to perform parameter
estimation and utilize a BIC-like criterion to select the model parameters,
including the sparsity tuning hyperparameters. Experiments conducted on
simulated data show the good performance of the proposed regularized MLE
compared to the standard MLE with the EM algorithm.
| stat |
A Decentralized Approach to Bayesian Learning | Motivated by decentralized approaches to machine learning, we propose a
collaborative Bayesian learning algorithm taking the form of decentralized
Langevin dynamics in a non-convex setting. Our analysis show that the initial
KL-divergence between the Markov Chain and the target posterior distribution is
exponentially decreasing while the error contributions to the overall
KL-divergence from the additive noise is decreasing in polynomial time. We
further show that the polynomial-term experiences speed-up with number of
agents and provide sufficient conditions on the time-varying step-sizes to
guarantee convergence to the desired distribution. The performance of the
proposed algorithm is evaluated on a wide variety of machine learning tasks.
The empirical results show that the performance of individual agents with
locally available data is on par with the centralized setting with considerable
improvement in the convergence rate.
| stat |
Joint Stochastic Approximation and Its Application to Learning Discrete
Latent Variable Models | Although with progress in introducing auxiliary amortized inference models,
learning discrete latent variable models is still challenging. In this paper,
we show that the annoying difficulty of obtaining reliable stochastic gradients
for the inference model and the drawback of indirectly optimizing the target
log-likelihood can be gracefully addressed in a new method based on stochastic
approximation (SA) theory of the Robbins-Monro type. Specifically, we propose
to directly maximize the target log-likelihood and simultaneously minimize the
inclusive divergence between the posterior and the inference model. The
resulting learning algorithm is called joint SA (JSA). To the best of our
knowledge, JSA represents the first method that couples an SA version of the EM
(expectation-maximization) algorithm (SAEM) with an adaptive MCMC procedure.
Experiments on several benchmark generative modeling and structured prediction
tasks show that JSA consistently outperforms recent competitive algorithms,
with faster convergence, better final likelihoods, and lower variance of
gradient estimates.
| stat |
Rejoinder: "Gene Hunting with Hidden Markov Model Knockoffs" | In this paper we deepen and enlarge the reflection on the possible advantages
of a knockoff approach to genome wide association studies (Sesia et al., 2018),
starting from the discussions in Bottolo & Richardson (2019); Jewell & Witten
(2019); Rosenblatt et al. (2019) and Marchini (2019). The discussants bring up
a number of important points, either related to the knockoffs methodology in
general, or to its specific application to genetic studies. In the following we
offer some clarifications, mention relevant recent developments and highlight
some of the still open problems.
| stat |
Functional clustering methods for resistance spot welding process data
in the automotive industry | Quality assessment of resistance spot welding (RSW) joints of metal sheets in
the automotive industry is typically based on costly and lengthy off-line tests
that are unfeasible on the full production, especially on large scale. However,
the massive industrial digitalization triggered by the industry 4.0 framework
makes available, for every produced joint, on-line RSW process parameters, such
as, in particular, the so-called dynamic resistance curve (DRC), which is
recognized as the full technological signature of the spot welds. Motivated by
this context, the present paper means to show the potentiality and the
practical applicability to clustering methods of the functional data approach
that avoids the need for arbitrary and often controversial feature extraction
to find out homogeneous groups of DRCs, which likely pertain to spot welds
sharing common mechanical and metallurgical properties. We intend is to provide
an essential hands-on overview of the most promising functional clustering
methods, and to apply the latter to the DRCs collected from the RSW process at
hand, even if they could go far beyond the specific application hereby
investigated. The methods analyzed are demonstrated to possibly support
practitioners along the identification of the mapping relationship between
process parameters and the final quality of RSW joints as well as, more
specifically, along the priority assignment for off-line testing of welded
spots and the welding tool wear analysis. The analysis code, that has been
developed through the software environment R, and the DRC data set are made
openly available online at https://github.com/unina-sfere/funclustRSW/
| stat |
Robust Bayesian Inference for Big Data: Combining Sensor-based Records
with Traditional Survey Data | Big Data often presents as massive non-probability samples. Not only is the
selection mechanism often unknown, but larger data volume amplifies the
relative contribution of selection bias to total error. Existing bias
adjustment approaches assume that the conditional mean structures have been
correctly specified for the selection indicator or key substantive measures. In
the presence of a reference probability sample, these methods rely on a
pseudo-likelihood method to account for the sampling weights of the reference
sample, which is parametric in nature. Under a Bayesian framework, handling the
sampling weights is an even bigger hurdle. To further protect against model
misspecification, we expand the idea of double robustness such that more
flexible non-parametric methods as well as Bayesian models can be used for
prediction. In particular, we employ Bayesian additive regression trees, which
not only capture non-linear associations automatically but permit direct
quantification of the uncertainty of point estimates through its posterior
predictive draws. We apply our method to sensor-based naturalistic driving data
from the second Strategic Highway Research Program using the 2017 National
Household Travel Survey as a benchmark.
| stat |
Fitting Laplacian Regularized Stratified Gaussian Models | We consider the problem of jointly estimating multiple related zero-mean
Gaussian distributions from data. We propose to jointly estimate these
covariance matrices using Laplacian regularized stratified model fitting, which
includes loss and regularization terms for each covariance matrix, and also a
term that encourages the different covariances matrices to be close. This
method `borrows strength' from the neighboring covariances, to improve its
estimate. With well chosen hyper-parameters, such models can perform very well,
especially in the low data regime. We propose a distributed method that scales
to large problems, and illustrate the efficacy of the method with examples in
finance, radar signal processing, and weather forecasting.
| stat |
Generalized Bayesian Filtering via Sequential Monte Carlo | We introduce a framework for inference in general state-space hidden Markov
models (HMMs) under likelihood misspecification. In particular, we leverage the
loss-theoretic perspective of Generalized Bayesian Inference (GBI) to define
generalised filtering recursions in HMMs, that can tackle the problem of
inference under model misspecification. In doing so, we arrive at principled
procedures for robust inference against observation contamination by utilising
the $\beta$-divergence. Operationalising the proposed framework is made
possible via sequential Monte Carlo methods (SMC), where most standard particle
methods, and their associated convergence results, are readily adapted to the
new setting. We apply our approach to object tracking and Gaussian process
regression problems, and observe improved performance over both standard
filtering algorithms and other robust filters.
| stat |
Smaller $p$-values in genomics studies using distilled historical
information | Medical research institutions have generated massive amounts of biological
data by genetically profiling hundreds of cancer cell lines. In parallel,
academic biology labs have conducted genetic screens on small numbers of cancer
cell lines under custom experimental conditions. In order to share information
between these two approaches to scientific discovery, this article proposes a
"frequentist assisted by Bayes" (FAB) procedure for hypothesis testing that
allows historical information from massive genomics datasets to increase the
power of hypothesis tests in specialized studies. The exchange of information
takes place through a novel probability model for multimodal genomics data,
which distills historical information pertaining to cancer cell lines and genes
across a wide variety of experimental contexts. If the relevance of the
historical information for a given study is high, then the resulting FAB tests
can be more powerful than the corresponding classical tests. If the relevance
is low, then the FAB tests yield as many discoveries as the classical tests.
Simulations and practical investigations demonstrate that the FAB testing
procedure can increase the number of effects discovered in genomics studies
while still maintaining strict control of type I error and false discovery
rates.
| stat |
ABC Learning of Hawkes Processes with Missing or Noisy Event Times | The self-exciting Hawkes process is widely used to model events which occur
in bursts. However, many real world data sets contain missing events and/or
noisily observed event times, which we refer to as data distortion. The
presence of such distortion can severely bias the learning of the Hawkes
process parameters. To circumvent this, we propose modeling the distortion
function explicitly. This leads to a model with an intractable likelihood
function which makes it difficult to deploy standard parameter estimation
techniques. As such, we develop the ABC-Hawkes algorithm which is a novel
approach to estimation based on Approximate Bayesian Computation (ABC) and
Markov Chain Monte Carlo. This allows the parameters of the Hawkes process to
be learned in settings where conventional methods induce substantial bias or
are inapplicable. The proposed approach is shown to perform well on both real
and simulated data.
| stat |
Detecting spatial clusters in functional data: new scan statistic
approaches | We have developed two scan statistics for detecting clusters of functional
data indexed in space. The first method is based on an adaptation of a
functional analysis of variance and the second one is based on a
distribution-free spatial scan statistic for univariate data. In a simulation
study, the distribution-free method always performed better than a
nonparametric functional scan statistic, and the adaptation of the anova also
performed better for data with a normal or a quasi-normal distribution. Our
methods can detect smaller spatial clusters than the nonparametric method.
Lastly, we used our scan statistics for functional data to search for spatial
clusters of abnormal unemployment rates in France over the period 1998-2013
(divided into quarters).
| stat |
GP-VAE: Deep Probabilistic Time Series Imputation | Multivariate time series with missing values are common in areas such as
healthcare and finance, and have grown in number and complexity over the years.
This raises the question whether deep learning methodologies can outperform
classical data imputation methods in this domain. However, naive applications
of deep learning fall short in giving reliable confidence estimates and lack
interpretability. We propose a new deep sequential latent variable model for
dimensionality reduction and data imputation. Our modeling assumption is simple
and interpretable: the high dimensional time series has a lower-dimensional
representation which evolves smoothly in time according to a Gaussian process.
The non-linear dimensionality reduction in the presence of missing data is
achieved using a VAE approach with a novel structured variational
approximation. We demonstrate that our approach outperforms several classical
and deep learning-based data imputation methods on high-dimensional data from
the domains of computer vision and healthcare, while additionally improving the
smoothness of the imputations and providing interpretable uncertainty
estimates.
| stat |
On the Suboptimality of Thompson Sampling in High Dimensions | In this paper we consider Thompson Sampling for combinatorial semi-bandits.
We demonstrate that, perhaps surprisingly, Thompson Sampling is sub-optimal for
this problem in the sense that its regret scales exponentially in the ambient
dimension, and its minimax regret scales almost linearly. This phenomenon
occurs under a wide variety of assumptions including both non-linear and linear
reward functions. We also show that including a fixed amount of forced
exploration to Thompson Sampling does not alleviate the problem. We complement
our theoretical results with numerical results and show that in practice
Thompson Sampling indeed can perform very poorly in high dimensions.
| stat |
Learning When-to-Treat Policies | Many applied decision-making problems have a dynamic component: The
policymaker needs not only to choose whom to treat, but also when to start
which treatment. For example, a medical doctor may choose between postponing
treatment (watchful waiting) and prescribing one of several available
treatments during the many visits from a patient. We develop an "advantage
doubly robust" estimator for learning such dynamic treatment rules using
observational data under the assumption of sequential ignorability. We prove
welfare regret bounds that generalize results for doubly robust learning in the
single-step setting, and show promising empirical performance in several
different contexts. Our approach is practical for policy optimization, and does
not need any structural (e.g., Markovian) assumptions.
| stat |
Spatiotemporal Dynamics, Nowcasting and Forecasting of COVID-19 in the
United States | Epidemic modeling is an essential tool to understand the spread of the novel
coronavirus and ultimately assist in disease prevention, policymaking, and
resource allocation. In this article, we establish a state of the art interface
between classic mathematical and statistical models and propose a novel
space-time epidemic modeling framework to study the spatial-temporal pattern in
the spread of infectious disease. We propose a quasi-likelihood approach via
the penalized spline approximation and alternatively reweighted least-squares
technique to estimate the model. Furthermore, we provide a short-term and
long-term county-level prediction of the infected/death count for the U.S. by
accounting for the control measures, health service resources, and other local
features. Utilizing spatiotemporal analysis, our proposed model enhances the
dynamics of the epidemiological mechanism and dissects the spatiotemporal
structure of the spreading disease. To assess the uncertainty associated with
the prediction, we develop a projection band based on the envelope of the
bootstrap forecast paths. The performance of the proposed method is evaluated
by a simulation study. We apply the proposed method to model and forecast the
spread of COVID-19 at both county and state levels in the United States.
| stat |
Bayesian inference for a single factor copula stochastic volatility
model using Hamiltonian Monte Carlo | For modeling multivariate financial time series we propose a single factor
copula model together with stochastic volatility margins. This model
generalizes single factor models relying on the multivariate normal
distribution and allows for symmetric and asymmetric tail dependence. We
develop joint Bayesian inference using Hamiltonian Monte Carlo (HMC) within
Gibbs sampling. Thus we avoid information loss caused by the two-step approach
for margins and dependence in copula models as followed by Schamberger et
al(2017). Further, the Bayesian approach allows for high dimensional parameter
spaces as they are present here in addition to uncertainty quantification
through credible intervals. By allowing for indicators for different copula
families the copula families are selected automatically in the Bayesian
framework. In a first simulation study the performance of HMC is compared to
the Markov Chain Monte Carlo (MCMC) approach developed by Schamberger et
al(2017) for the copula part. It is shown that HMC considerably outperforms
this approach in terms of effective sample size, MSE and observed coverage
probabilities. In a second simulation study satisfactory performance is seen
for the full HMC within Gibbs procedure. The approach is illustrated for a
portfolio of financial assets with respect to one-day ahead value at risk
forecasts. We provide comparison to a two-step estimation procedure of the
proposed model and to relevant benchmark models: a model with dynamic linear
models for the margins and a single factor copula for the dependence proposed
by Schamberger et al(2017) and a multivariate factor stochastic volatility
model proposed by Kastner et al(2017). Our proposed approach shows superior
performance.
| stat |
Maximum Likelihood for Gaussian Process Classification and Generalized
Linear Mixed Models under Case-Control Sampling | Modern data sets in various domains often include units that were sampled
non-randomly from the population and have a latent correlation structure. Here
we investigate a common form of this setting, where every unit is associated
with a latent variable, all latent variables are correlated, and the
probability of sampling a unit depends on its response. Such settings often
arise in case-control studies, where the sampled units are correlated due to
spatial proximity, family relations, or other sources of relatedness. Maximum
likelihood estimation in such settings is challenging from both a computational
and statistical perspective, necessitating approximations that take the
sampling scheme into account. We propose a family of approximate likelihood
approaches which combine composite likelihood and expectation propagation. We
demonstrate the efficacy of our solutions via extensive simulations. We utilize
them to investigate the genetic architecture of several complex disorders
collected in case-control genetic association studies, where hundreds of
thousands of genetic variants are measured for every individual, and the
underlying disease liabilities of individuals are correlated due to genetic
similarity. Our work is the first to provide a tractable likelihood-based
solution for case-control data with complex dependency structures.
| stat |
Ordinal Regression with Fenton-Wilkinson Order Statistics: A Case Study
of an Orienteering Race | In sports, individuals and teams are typically interested in final rankings.
Final results, such as times or distances, dictate these rankings, also known
as places. Places can be further associated with ordered random variables,
commonly referred to as order statistics. In this work, we introduce a simple,
yet accurate order statistical ordinal regression function that predicts relay
race places with changeover-times. We call this function the Fenton-Wilkinson
Order Statistics model. This model is built on the following educated
assumption: individual leg-times follow log-normal distributions. Moreover, our
key idea is to utilize Fenton-Wilkinson approximations of changeover-times
alongside an estimator for the total number of teams as in the notorious German
tank problem. This original place regression function is sigmoidal and thus
correctly predicts the existence of a small number of elite teams that
significantly outperform the rest of the teams. Our model also describes how
place increases linearly with changeover-time at the inflection point of the
log-normal distribution function. With real-world data from Jukola 2019, a
massive orienteering relay race, the model is shown to be highly accurate even
when the size of the training set is only 5% of the whole data set. Numerical
results also show that our model exhibits smaller place prediction
root-mean-square-errors than linear regression, mord regression and Gaussian
process regression.
| stat |
Proximal Methods for Sparse Optimal Scoring and Discriminant Analysis | Linear discriminant analysis (LDA) is a classical method for dimensionality
reduction, where discriminant vectors are sought to project data to a lower
dimensional space for optimal separability of classes. Several recent papers
have outlined strategies for exploiting sparsity for using LDA with
high-dimensional data. However, many lack scalable methods for solution of the
underlying optimization problems. We propose three new numerical optimization
schemes for solving the sparse optimal scoring formulation of LDA based on
block coordinate descent, the proximal gradient method, and the alternating
direction method of multipliers. We show that the per-iteration cost of these
methods scales linearly in the dimension of the data provided restricted
regularization terms are employed, and cubically in the dimension of the data
in the worst case. Furthermore, we establish that if our block coordinate
descent framework generates convergent subsequences of iterates, then these
subsequences converge to the stationary points of the sparse optimal scoring
problem. We demonstrate the effectiveness of our new methods with empirical
results for classification of Gaussian data and data sets drawn from
benchmarking repositories, including time-series and multispectral X-ray data,
and provide Matlab and R implementations of our optimization schemes.
| stat |
A faster and more accurate algorithm for calculating population genetics
statistics requiring sums of Stirling numbers of the first kind | Stirling numbers of the first kind are used in the derivation of several
population genetics statistics, which in turn are useful for testing
evolutionary hypotheses directly from DNA sequences. Here, we explore the
cumulative distribution function of these Stirling numbers, which enables a
single direct estimate of the sum, using representations in terms of the
incomplete beta function. This estimator enables an improved method for
calculating an asymptotic estimate for one useful statistic, Fu's $F_s$. By
reducing the calculation from a sum of terms involving Stirling numbers to a
single estimate, we simultaneously improve accuracy and dramatically increase
speed.
| stat |
Causal Effects of Prenatal Drug Exposure on Birth Defects with Missing
by Terathanasia | We investigate the causal effects of drug exposure on birth defects,
motivated by a recent cohort study of birth outcomes in pregnancies of women
treated with a given medication, that revealed a higher rate of major
structural birth defects in infants born to exposed versus unexposed women. An
outstanding problem in this study was the missing birth defect outcomes among
pregnancy losses resulting from spontaneous abortion. This led to missing not
at random because, according to the theory of "terathanasia", a defected fetus
is more likely to be spontaneously aborted. In addition, the previous analysis
stratified on live birth against spontaneous abortion, which was itself a
post-exposure variable and hence did not lead to causal interpretation of the
stratified results. In this paper we aimed to estimate and provide inference
for the causal parameters of scientific interest, including the principal
effects, making use of the missing data mechanism informed by terathanasia.
During this process we also dealt with complications in the data including left
truncation, observational nature, and rare events. We report our findings which
shed light on how studies on causal effects of medication or other exposures
during pregnancy may be analyzed.
| stat |
Comparisons of proportions in k dose groups against a negative control
assuming order restriction: Williams-type test vs. closed test procedures | The comparison of proportions is considered in the asymptotic generalized
linear model with the odds ratio as effect size. When several doses are
compared with a control assuming an order restriction, a Williams-type trend
test can be used. As an alternative, two variants of the closed testing
approach are considered, one using global Williams-tests in the partition
hypotheses, one with pairwise contrasts. Their advantages in terms of power and
simplicity are demonstrated. Related R-code is provided.
| stat |
Trees-Based Models for Correlated Data | This paper presents a new approach for trees-based regression, such as simple
regression tree, random forest and gradient boosting, in settings involving
correlated data. We show the problems that arise when implementing standard
trees-based regression models, which ignore the correlation structure. Our new
approach explicitly takes the correlation structure into account in the
splitting criterion, stopping rules and fitted values in the leaves, which
induces some major modifications of standard methodology. The superiority of
our new approach over trees-based models that do not account for the
correlation is supported by simulation experiments and real data analyses.
| stat |
A spatial functional count model for heterogeneity analysis in time | A spatial curve dynamical model framework is adopted for functional
prediction of counts in a spatiotemporal log-Gaussian Cox process model. Our
spatial functional estimation approach handles both wavelet-based heterogeneity
analysis in time, and spectral analysis in space. Specifically, model fitting
is achieved by minimising the information divergence or relative entropy
between the multiscale model underlying the data and the corresponding
candidates in the spatial spectral domain. A simulation study is carried out
within the family of log-Gaussian Spatial Autoregressive l2-valued processes
(SARl2 processes) to illustrate the asymptotic properties of the proposed
spatial functional estimators. We apply our modelling strategy to
spatiotemporal prediction of respiratory disease mortality.
| stat |
A New Perspective on Debiasing Linear Regressions | In this paper, we propose an abstract procedure for debiasing constrained or
regularized potentially high-dimensional linear models. It is elementary to
show that the proposed procedure can produce $\frac{1}{\sqrt{n}}$-confidence
intervals for individual coordinates (or even bounded contrasts) in models with
unknown covariance, provided that the covariance has bounded spectrum. While
the proof of the statistical guarantees of our procedure is simple, its
implementation requires more care due to the complexity of the optimization
programs we need to solve. We spend the bulk of this paper giving examples in
which the proposed algorithm can be implemented in practice. One fairly general
class of instances which are amenable to applications of our procedure include
convex constrained least squares. We are able to translate the procedure to an
abstract algorithm over this class of models, and we give concrete examples
where efficient polynomial time methods for debiasing exist. Those include the
constrained version of LASSO, regression under monotone constraints, regression
with positive monotone constraints and non-negative least squares. In addition,
we show that our abstract procedure can be applied to efficiently debias SLOPE
and square-root SLOPE, among other popular regularized procedures under certain
assumptions. We provide thorough simulation results in support of our
theoretical findings.
| stat |
Asian Handicap football betting with Rating-based Hybrid Bayesian
Networks | Despite the massive popularity of the Asian Handicap (AH) football betting
market, it has not been adequately studied by the relevant literature. This
paper combines rating systems with hybrid Bayesian networks and presents the
first published model specifically developed for prediction and assessment of
the AH betting market. The results are based on 13 English Premier League
seasons and are compared to the traditional 1X2 market. Different betting
situations have been examined including a) both average and maximum (best
available) market odds, b) all possible betting decision thresholds between
predicted and published odds, c) optimisations for both return-on-investment
and profit, and d) simple stake adjustments to investigate how the variance of
returns changes when targeting equivalent profit in both 1X2 and AH markets.
While the AH market is found to share the inefficiencies of the traditional 1X2
market, the findings reveal both interesting differences as well as
similarities between the two.
| stat |
The GWR route map: a guide to the informed application of Geographically
Weighted Regression | Geographically Weighted Regression (GWR) is increasingly used in spatial
analyses of social and environmental data. It allows spatial heterogeneities in
processes and relationships to be investigated through a series of local
regression models rather than a global one. Standard GWR assumes that the
relationships between the response and predictor variables operate at the same
spatial scale, which is frequently not the case. To address this, several GWR
variants have been proposed. This paper describes a route map to inform the
choice of whether to use a GWR model or not, and if so which of three core
variants to apply: a standard GWR, a mixed GWR or a multiscale GWR (MS-GWR).
The route map comprises primary steps: a basic linear regression, a MS-GWR, and
investigations of the results of these. The paper provides guidance for
deciding whether to use a GWR approach, and if so for determining the
appropriate GWR variant. It describes the importance of investigating a number
of secondary issues at global and local scales including collinearity, the
influence of outliers, and dependent error terms. Code and data for the case
study used to illustrate the route map are provided, and further considerations
are described in an extensive Appendix.
| stat |
HMC: avoiding rejections by not using leapfrog and some results on the
acceptance rate | The leapfrog integrator is routinely used within the Hamiltonian Monte Carlo
method and its variants. We give strong numerical evidence that alternative,
easy to implement algorithms yield fewer rejections with a given computational
effort. When the dimensionality of the target distribution is high, the number
of accepted proposals may be multiplied by a factor of three or more. This
increase in the number of accepted proposals is not achieved by impairing any
positive features of the sampling. We also establish new non-asymptotic and
asymptotic results on the monotonic relationship between the expected
acceptance rate and the expected energy error. These results further validate
the derivation of one of the integrators we consider and are of independent
interest.
| stat |
Persistence Codebooks for Topological Data Analysis | Persistent homology (PH) is a rigorous mathematical theory that provides a
robust descriptor of data in the form of persistence diagrams (PDs) which are
2D multisets of points. Their variable size makes them, however, difficult to
combine with typical machine learning workflows. In this paper we introduce
persistence codebooks, a novel expressive and discriminative fixed-size
vectorized representation of PDs. To this end, we adapt bag-of-words (BoW),
vectors of locally aggregated descriptors (VLAD) and Fischer vectors (FV) for
the quantization of PDs. Persistence codebooks represent PDs in a convenient
way for machine learning and statistical analysis and have a number of
favorable practical and theoretical properties including 1-Wasserstein
stability. We evaluate the presented representations on several heterogeneous
datasets and show their (high) discriminative power. Our approach achieves
state-of-the-art performance and beyond in much less time than alternative
approaches.
| stat |
A Theory of Hyperbolic Prototype Learning | We introduce Hyperbolic Prototype Learning, a type of supervised learning,
where class labels are represented by ideal points (points at infinity) in
hyperbolic space. Learning is achieved by minimizing the 'penalized Busemann
loss', a new loss function based on the Busemann function of hyperbolic
geometry. We discuss several theoretical features of this setup. In particular,
Hyperbolic Prototype Learning becomes equivalent to logistic regression in the
one-dimensional case.
| stat |
Sequential Design of Mixture Experiments with an Empirically Determined
Input Domain and an Application to Burn-up Credit Penalization of Nuclear
Fuel Rods | This paper proposes a sequential design for maximizing a stochastic computer
simulator output, y(x), over an unknown optimization domain. The training data
used to estimate the optimization domain are a set of (historical) inputs,
often from a physical system modeled by the simulator. Two methods are provided
for estimating the simulator input domain. An extension of the well-known
efficient global optimization algorithm is presented to maximize y(x). The
domain estimation/maximization procedure is applied to two readily understood
analytic examples. It is also used to solve a problem in nuclear safety by
maximizing the k-effective "criticality coefficient" of spent fuel rods,
considered as one-dimensional heterogeneous fissile media. One of the two
domain estimation methods relies on expertise-type constraints. We show that
these constraints, initially chosen to address the spent fuel rod example, are
robust in that they also lead to good results in the second analytic
optimization example. Of course, in other applications, it could be necessary
to design alternative constraints that are more suitable for these
applications.
| stat |
Random forests for high-dimensional longitudinal data | Random forests is a state-of-the-art supervised machine learning method which
behaves well in high-dimensional settings although some limitations may happen
when $p$, the number of predictors, is much larger than the number of
observations $n$. Repeated measurements can help by offering additional
information but no approach has been proposed for high-dimensional longitudinal
data. Random forests have been adapted to standard (i.e., $n > p$) longitudinal
data by using a semi-parametric mixed-effects model, in which the
non-parametric part is estimated using random forests. We first propose a
stochastic extension of the model which allows the covariance structure to vary
over time. Furthermore, we develop a new method which takes intra-individual
covariance into consideration to build the forest. Simulations reveal the
superiority of our approach compared to existing ones. The method has been
applied to an HIV vaccine trial including 17 HIV infected patients with 10
repeated measurements of 20000 gene transcripts and the blood concentration of
human immunodeficiency virus RNA at the time of antiretroviral interruption.
The approach selected 21 gene transcripts for which the association with HIV
viral load was fully relevant and consistent with results observed during
primary infection.
| stat |
An Interventionist Approach to Mediation Analysis | Judea Pearl's insight that, when errors are assumed independent, the Pure
(aka Natural) Direct Effect (PDE) is non-parametrically identified via the
Mediation Formula was `path-breaking' in more than one sense! In the same paper
Pearl described a thought-experiment as a way to motivate the PDE. Analysis of
this experiment led Robins \& Richardson to a novel way of conceptualizing
direct effects in terms of interventions on an expanded graph in which
treatment is decomposed into multiple separable components. We further develop
this novel theory here, showing that it provides a self-contained framework for
discussing mediation without reference to cross-world (nested) counterfactuals
or interventions on the mediator. The theory preserves the dictum `no causation
without manipulation' and makes questions of mediation empirically testable in
future Randomized Controlled Trials. Even so, we prove the interventionist and
nested counterfactual approaches remain tightly coupled under a Non-Parametric
Structural Equation Model except in the presence of a `recanting witness.' In
fact, our analysis also leads to a simple sound and complete algorithm for
determining identification in the (non-interventionist) theory of path-specific
counterfactuals.
| stat |
Stochastic geometry to generalize the Mondrian Process | The Mondrian process is a stochastic process that produces a recursive
partition of space with random axis-aligned cuts. Random forests and Laplace
kernel approximations built from the Mondrian process have led to efficient
online learning methods and Bayesian optimization. By viewing the Mondrian
process as a special case of the stable under iterated tessellation (STIT)
process, we utilize tools from stochastic geometry to resolve some fundamental
questions concerning the Mondrian process in machine learning. First, we show
that the Mondrian process with general cut directions can be efficiently
simulated by lifting to a higher dimensional axis-aligned Mondrian process.
Second, we characterize all possible kernels that generalizations of the
Mondrian process can approximate with fixed parameters as well as additional
kernels obtained from mixtures of STIT processes. This includes, for instance,
various forms of the weighted Laplace kernel and the exponential kernel.
Lastly, we give an explicit formula for the density estimator arising from a
Mondrian forest. This allows for precise comparisons between the Mondrian
forest, the Mondrian kernel and the Laplace kernel in density estimation. Our
paper calls for further developments at the novel intersection of stochastic
geometry and machine learning.
| stat |
Dirichlet process mixtures under affine transformations of the data | Location-scale Dirichlet process mixtures of Gaussians (DPM-G) have proved
extremely useful in dealing with density estimation and clustering problems in
a wide range of domains. Motivated by an astronomical application, in this work
we address the robustness of DPM-G models to affine transformations of the
data, a natural requirement for any sensible statistical method for density
estimation and clustering. First, we devise a coherent prior specification of
the model which makes posterior inference invariant with respect to affine
transformations of the data. Second, we formalise the notion of asymptotic
robustness under data transformation and show that mild assumptions on the true
data generating process are sufficient to ensure that DPM-G models feature such
a property. Our investigation is supported by an extensive simulation study and
illustrated by the analysis of an astronomical dataset consisting of physical
measurements of stars in the field of the globular cluster NGC 2419.
| stat |
Estimating Treatment Effects with Causal Forests: An Application | We apply causal forests to a dataset derived from the National Study of
Learning Mindsets, and consider resulting practical and conceptual challenges.
In particular, we discuss how causal forests use estimated propensity scores to
be more robust to confounding, and how they handle data with clustered errors.
| stat |
Fully Bayesian inference for spatiotemporal data with the
multi-resolution approximation | Large spatiotemporal datasets are a challenge for conventional Bayesian
models because of the cubic computational complexity of the algorithms for
obtaining the Cholesky decomposition of the covariance matrix in the
multivariate normal density. Moreover, standard numerical algorithms for
posterior estimation, such as Markov Chain Monte Carlo (MCMC), are intractable
in this context, as they require thousands, if not millions, of costly
likelihood evaluations. To overcome those limitations, we propose IS-MRA
(Importance sampling - Multi-Resolution Approximation), which takes advantage
of the sparse inverse covariance structure produced by the Multi-Resolution
Approximation (MRA) approach. IS-MRA is fully Bayesian and facilitates the
approximation of the hyperparameter marginal posterior distributions. We apply
IS-MRA to large MODIS Level 3 Land Surface Temperature (LST) datasets, sampled
between May 18 and May 31, 2012 in the western part of the state of
Maharashtra, India. We find that IS-MRA can produce realistic prediction
surfaces over regions where concentrated missingness, caused by sizable cloud
cover, is observed. Through a validation analysis and simulation study, we also
find that predictions tend to be very accurate.
| stat |
Nonparametric fusion learning: synthesize inferences from diverse
sources using depth confidence distribution | Fusion learning refers to synthesizing inferences from multiple sources or
studies to provide more effective inference and prediction than from any
individual source or study alone. Most existing methods for synthesizing
inferences rely on parametric model assumptions, such as normality, which often
do not hold in practice. In this paper, we propose a general nonparametric
fusion learning framework for synthesizing inferences of the target parameter
from multiple sources. The main tool underlying the proposed framework is the
notion of depth confidence distribution (depth-CD), which is also developed in
this paper. Broadly speaking, a depth-CD is a data-driven nonparametric summary
distribution of inferential information for the target parameter. We show that
a depth-CD is a useful inferential tool and, moreover, is an omnibus form of
confidence regions (or p-values), whose contours of level sets shrink toward
the true parameter value. The proposed fusion learning approach combines
depth-CDs from the individual studies, with each depth-CD constructed by
nonparametric bootstrap and data depth. This approach is shown to be efficient,
general and robust. Specifically, it achieves high-order accuracy and Bahadur
efficiency under suitably chosen combining elements. It allows the model or
inference structure to be different among individual studies. And it readily
adapts to heterogeneous studies with a broad range of complex and irregular
settings. This property enables it to utilize indirect evidence from incomplete
studies to gain efficiency in the overall inference. The advantages of the
proposed approach are demonstrated simulations and in a Federal Aviation
Administration (FAA) study of aircraft landing performance.
| stat |
Integrating Latent Classes in the Bayesian Shared Parameter Joint Model
of Longitudinal and Survival Outcomes | Cystic fibrosis is a chronic lung disease which requires frequent patient
monitoring to maintain lung function over time and minimize onset of acute
respiratory events known as pulmonary exacerbations. From the clinical point of
view it is important to characterize the association between key biomarkers
such as $FEV_1$ and time-to first exacerbation. Progression of the disease is
heterogeneous, yielding different sub-groups in the population exhibiting
distinct longitudinal profiles. It is desirable to categorize these unobserved
sub-groups (latent classes) according to their distinctive trajectories.
Accounting for these latent classes, in other words heterogeneity, will lead to
improved estimates of association arising from the joint longitudinal-survival
model.
The joint model of longitudinal and survival data constitutes a popular
framework to analyze such data arising from heterogeneous cohorts. In
particular, two paradigms within this framework are the shared parameter joint
models and the joint latent class models. The former paradigm allows one to
quantify the strength of the association between the longitudinal and survival
outcomes but does not allow for latent sub-populations. The latter paradigm
explicitly postulates the existence of sub-populations but does not directly
quantify the strength of the association.
We propose to integrate latent classes in the shared parameter joint model in
a fully Bayesian approach, which allows us to investigate the association
between $FEV_1$ and time-to first exacerbation within each latent class. We,
furthermore, focus on the selection of the optimal number of latent classes.
| stat |
Development of modeling and control strategies for an approximated
Gaussian process | The Gaussian process (GP) model, which has been extensively applied as priors
of functions, has demonstrated excellent performance. The specification of a
large number of parameters affects the computational efficiency and the
feasibility of implementation of a control strategy. We propose a linear model
to approximate GPs; this model expands the GP model by a series of basis
functions. Several examples and simulation studies are presented to demonstrate
the advantages of the proposed method. A control strategy is provided with the
proposed linear model.
| stat |
Approximate Bayesian computation with the Wasserstein distance | A growing number of generative statistical models do not permit the numerical
evaluation of their likelihood functions. Approximate Bayesian computation
(ABC) has become a popular approach to overcome this issue, in which one
simulates synthetic data sets given parameters and compares summaries of these
data sets with the corresponding observed values. We propose to avoid the use
of summaries and the ensuing loss of information by instead using the
Wasserstein distance between the empirical distributions of the observed and
synthetic data. This generalizes the well-known approach of using order
statistics within ABC to arbitrary dimensions. We describe how recently
developed approximations of the Wasserstein distance allow the method to scale
to realistic data sizes, and propose a new distance based on the Hilbert
space-filling curve. We provide a theoretical study of the proposed method,
describing consistency as the threshold goes to zero while the observations are
kept fixed, and concentration properties as the number of observations grows.
Various extensions to time series data are discussed. The approach is
illustrated on various examples, including univariate and multivariate g-and-k
distributions, a toggle switch model from systems biology, a queueing model,
and a L\'evy-driven stochastic volatility model.
| stat |
The magnitude and direction of collider bias for binary variables | Suppose we are interested in the effect of variable $X$ on variable $Y$. If
$X$ and $Y$ both influence, or are associated with variables that influence, a
common outcome, called a collider, then conditioning on the collider (or on a
variable influenced by the collider -- its "child") induces a spurious
association between $X$ and $Y$, which is known as collider bias.
Characterizing the magnitude and direction of collider bias is crucial for
understanding the implications of selection bias and for adjudicating decisions
about whether to control for variables that are known to be associated with
both exposure and outcome but could be either confounders or colliders.
Considering a class of situations where all variables are binary, and where $X$
and $Y$ either are, or are respectively influenced by, two marginally
independent causes of a collider, we derive collider bias that results from (i)
conditioning on specific levels of, or (ii) linear regression adjustment for,
the collider (or its child). We also derive simple conditions that determine
the sign of such bias.
| stat |
Robustness Against Outliers For Deep Neural Networks By Gradient
Conjugate Priors | We analyze a new robust method for the reconstruction of probability
distributions of observed data in the presence of output outliers. It is based
on a so-called gradient conjugate prior (GCP) network which outputs the
parameters of a prior. By rigorously studying the dynamics of the GCP learning
process, we derive an explicit formula for correcting the obtained variance of
the marginal distribution and removing the bias caused by outliers in the
training set. Assuming a Gaussian (input-dependent) ground truth distribution
contaminated with a proportion $\varepsilon$ of outliers, we show that the
fitted mean is in a $c e^{-1/\varepsilon}$-neighborhood of the ground truth
mean and the corrected variance is in a $b\varepsilon$-neighborhood of the
ground truth variance, whereas the uncorrected variance of the marginal
distribution can even be infinite. We explicitly find $b$ as a function of the
output of the GCP network, without a priori knowledge of the outliers (possibly
input-dependent) distribution. Experiments with synthetic and real-world data
sets indicate that the GCP network fitted with a standard optimizer outperforms
other robust methods for regression.
| stat |
Nonlinear Discovery of Slow Molecular Modes using State-Free Reversible
VAMPnets | The success of enhanced sampling molecular simulations that accelerate along
collective variables (CVs) is predicated on the availability of variables
coincident with the slow collective motions governing the long-time
conformational dynamics of a system. It is challenging to intuit these slow CVs
for all but the simplest molecular systems, and their data-driven discovery
directly from molecular simulation trajectories has been a central focus of the
molecular simulation community to both unveil the important physical mechanisms
and to drive enhanced sampling. In this work, we introduce state-free
reversible VAMPnets (SRV) as a deep learning architecture that learns nonlinear
CV approximants to the leading slow eigenfunctions of the spectral
decomposition of the transfer operator that evolves equilibrium-scaled
probability distributions through time. Orthogonality of the learned CVs is
naturally imposed within network training without added regularization. The CVs
are inherently explicit and differentiable functions of the input coordinates
making them well-suited to use in enhanced sampling calculations. We
demonstrate the utility of SRVs in capturing parsimonious nonlinear
representations of complex system dynamics in applications to 1D and 2D toy
systems where the true eigenfunctions are exactly calculable and to molecular
dynamics simulations of alanine dipeptide and the WW domain protein.
| stat |
Semiparametric proximal causal inference | Skepticism about the assumption of no unmeasured confounding, also known as
exchangeability, is often warranted in making causal inferences from
observational data; because exchangeability hinges on an investigator's ability
to accurately measure covariates that capture all potential sources of
confounding. In practice, the most one can hope for is that covariate
measurements are at best proxies of the true underlying confounding mechanism
operating in a given observational study. In this paper, we consider the
framework of proximal causal inference introduced by Tchetgen Tchetgen et al.
(2020), which while explicitly acknowledging covariate measurements as
imperfect proxies of confounding mechanisms, offers an opportunity to learn
about causal effects in settings where exchangeability on the basis of measured
covariates fails. We make a number of contributions to proximal inference
including (i) an alternative set of conditions for nonparametric proximal
identification of the average treatment effect; (ii) general semiparametric
theory for proximal estimation of the average treatment effect including
efficiency bounds for key semiparametric models of interest; (iii) a
characterization of proximal doubly robust and locally efficient estimators of
the average treatment effect. Moreover, we provide analogous identification and
efficiency results for the average treatment effect on the treated. Our
approach is illustrated via simulation studies and a data application on
evaluating the effectiveness of right heart catheterization in the intensive
care unit of critically ill patients.
| stat |
PF: A C++ Library for Fast Particle Filtering | Particle filters are a class of algorithms that are used for "tracking" or
"filtering" in real-time for a wide array of time series models. Despite their
comprehensive applicability, particle filters are not always the tool of choice
for many practitioners, due to how difficult they are to implement. This short
article presents PF, a C++ header-only template library that provides fast
implementations of many different particle filters. A tutorial along with an
extensive fully-worked example is provided.
| stat |
Statistical Analysis of Dynamic Functional Brain Networks in Twins | Recent studies have shown that functional brain brainwork is dynamic even
during rest. A common approach to modeling the brain network in whole brain
resting-state fMRI is to compute the correlation between anatomical regions via
sliding windows. However, the direct use of the sample correlation matrices is
not reliable due to the image acquisition, processing noises and the use of
discrete windows that often introduce spurious high-frequency fluctuations and
the zig-zag pattern in the estimated time-varying correlation measures. To
address the problem and obtain more robust correlation estimates, we propose
the heat kernel based dynamic correlations. We demonstrate that the proposed
heat kernel method can smooth out the unwanted high-frequency fluctuations in
correlation estimations and achieve higher accuracy in identifying dynamically
changing distinct states. The method is further used in determining if such
dynamic state change is genetically heritable using a large-scale twin study.
Various methodological challenges for analyzing paired twin dynamic networks
are addressed.
| stat |
On a Variational Approximation based Empirical Likelihood ABC Method | Many scientifically well-motivated statistical models in natural,
engineering, and environmental sciences are specified through a generative
process. However, in some cases, it may not be possible to write down the
likelihood for these models analytically. Approximate Bayesian computation
(ABC) methods allow Bayesian inference in such situations. The procedures are
nonetheless typically computationally intensive. Recently, computationally
attractive empirical likelihood-based ABC methods have been suggested in the
literature. All of these methods rely on the availability of several suitable
analytically tractable estimating equations, and this is sometimes problematic.
We propose an easy-to-use empirical likelihood ABC method in this article.
First, by using a variational approximation argument as a motivation, we show
that the target log-posterior can be approximated as a sum of an expected joint
log-likelihood and the differential entropy of the data generating density. The
expected log-likelihood is then estimated by an empirical likelihood where the
only inputs required are a choice of summary statistic, it's observed value,
and the ability to simulate the chosen summary statistics for any parameter
value under the model. The differential entropy is estimated from the simulated
summaries using traditional methods. Posterior consistency is established for
the method, and we discuss the bounds for the required number of simulated
summaries in detail. The performance of the proposed method is explored in
various examples.
| stat |
A data-driven prospective study of incident dementia among older adults
in the United States | We conducted a prospective analysis of incident dementia and its association
with 65 sociodemographic, early-life, economic, health and behavioral, social,
and genetic risk factors in a sample of 7,908 adults over the age of 50 from
the nationally representative US-based Health and Retirement Study. We used
traditional survival analysis methods (Fine-Gray models) and a data-driven
approach (random survival forests for competing risks) which allowed us to
account for the competing risk of death with up to 14 years of follow-up.
Overall, the top five predictors across all groups were lower education,
loneliness, lower wealth and income, and lower self-reported health. However,
we observed variation in the leading predictors of dementia across
racial/ethnic and gender groups. Our ranked lists may be useful for guiding
future observational and quasi-experimental research that investigates
understudied domains of risk and emphasizes life course economic and health
conditions as well as disparities therein.
| stat |
The Broad Optimality of Profile Maximum Likelihood | We study three fundamental statistical-learning problems: distribution
estimation, property estimation, and property testing. We establish the profile
maximum likelihood (PML) estimator as the first unified sample-optimal approach
to a wide range of learning tasks. In particular, for every alphabet size $k$
and desired accuracy $\varepsilon$:
$\textbf{Distribution estimation}$ Under $\ell_1$ distance, PML yields
optimal $\Theta(k/(\varepsilon^2\log k))$ sample complexity for
sorted-distribution estimation, and a PML-based estimator empirically
outperforms the Good-Turing estimator on the actual distribution;
$\textbf{Additive property estimation}$ For a broad class of additive
properties, the PML plug-in estimator uses just four times the sample size
required by the best estimator to achieve roughly twice its error, with
exponentially higher confidence;
$\boldsymbol{\alpha}\textbf{-R\'enyi entropy estimation}$ For integer
$\alpha>1$, the PML plug-in estimator has optimal $k^{1-1/\alpha}$ sample
complexity; for non-integer $\alpha>3/4$, the PML plug-in estimator has sample
complexity lower than the state of the art;
$\textbf{Identity testing}$ In testing whether an unknown distribution is
equal to or at least $\varepsilon$ far from a given distribution in $\ell_1$
distance, a PML-based tester achieves the optimal sample complexity up to
logarithmic factors of $k$.
Most of these results also hold for a near-linear-time computable variant of
PML. Stronger results hold for a different and novel variant called truncated
PML (TPML).
| stat |
Hierarchical Total Variations and Doubly Penalized ANOVA Modeling for
Multivariate Nonparametric Regression | For multivariate nonparametric regression, functional analysis-of-variance
(ANOVA) modeling aims to capture the relationship between a response and
covariates by decomposing the unknown function into various components,
representing main effects, two-way interactions, etc. Such an approach has been
pursued explicitly in smoothing spline ANOVA modeling and implicitly in various
greedy methods such as MARS. We develop a new method for functional ANOVA
modeling, based on doubly penalized estimation using total-variation and
empirical-norm penalties, to achieve sparse selection of component functions
and their knots. For this purpose, we formulate a new class of hierarchical
total variations, which measures total variations at different levels including
main effects and multi-way interactions, possibly after some order of
differentiation. Furthermore, we derive suitable basis functions for
multivariate splines such that the hierarchical total variation can be
represented as a regular Lasso penalty, and hence we extend a previous
backfitting algorithm to handle doubly penalized estimation for ANOVA modeling.
We present extensive numerical experiments on simulations and real data to
compare our method with existing methods including MARS, tree boosting, and
random forest. The results are very encouraging and demonstrate considerable
gains from our method in both prediction or classification accuracy and
simplicity of the fitted functions.
| stat |
Likelihood-free inference by ratio estimation | We consider the problem of parametric statistical inference when likelihood
computations are prohibitively expensive but sampling from the model is
possible. Several so-called likelihood-free methods have been developed to
perform inference in the absence of a likelihood function. The popular
synthetic likelihood approach infers the parameters by modelling summary
statistics of the data by a Gaussian probability distribution. In another
popular approach called approximate Bayesian computation, the inference is
performed by identifying parameter values for which the summary statistics of
the simulated data are close to those of the observed data. Synthetic
likelihood is easier to use as no measure of `closeness' is required but the
Gaussianity assumption is often limiting. Moreover, both approaches require
judiciously chosen summary statistics. We here present an alternative inference
approach that is as easy to use as synthetic likelihood but not as restricted
in its assumptions, and that, in a natural way, enables automatic selection of
relevant summary statistic from a large set of candidates. The basic idea is to
frame the problem of estimating the posterior as a problem of estimating the
ratio between the data generating distribution and the marginal distribution.
This problem can be solved by logistic regression, and including regularising
penalty terms enables automatic selection of the summary statistics relevant to
the inference task. We illustrate the general theory on canonical examples and
employ it to perform inference for challenging stochastic nonlinear dynamical
systems and high-dimensional summary statistics.
| stat |
Replicated Vector Approximate Message Passing For Resampling Problem | Resampling techniques are widely used in statistical inference and ensemble
learning, in which estimators' statistical properties are essential. However,
existing methods are computationally demanding, because repetitions of
estimation/learning via numerical optimization/integral for each resampled data
are required. In this study, we introduce a computationally efficient method to
resolve such problem: replicated vector approximate message passing. This is
based on a combination of the replica method of statistical physics and an
accurate approximate inference algorithm, namely the vector approximate message
passing of information theory. The method provides tractable densities without
repeating estimation/learning, and the densities approximately offer an
arbitrary degree of the estimators' moment in practical time. In the
experiment, we apply the proposed method to the stability selection method,
which is commonly used in variable selection problems. The numerical results
show its fast convergence and high approximation accuracy for problems
involving both synthetic and real-world datasets.
| stat |
Normal variance mixtures: Distribution, density and parameter estimation | Normal variance mixtures are a class of multivariate distributions that
generalize the multivariate normal by randomizing (or mixing) the covariance
matrix via multiplication by a non-negative random variable W. The multivariate
t distribution is an example of such mixture, where W has an inverse-gamma
distribution. Algorithms to compute the joint distribution function and perform
parameter estimation for the multivariate normal and t (with integer degrees of
freedom) can be found in the literature and are implemented in, e.g., the R
package mvtnorm. In this paper, efficient algorithms to perform these tasks in
the general case of a normal variance mixture are proposed. In addition to the
above two tasks, the evaluation of the joint (logarithmic) density function of
a general normal variance mixture is tackled as well, as it is needed for
parameter estimation and does not always exist in closed form in this more
general setup. For the evaluation of the joint distribution function, the
proposed algorithms apply randomized quasi-Monte Carlo (RQMC) methods in a way
that improves upon existing methods proposed for the multivariate normal and t
distributions. An adaptive RQMC algorithm that similarly exploits the superior
convergence properties of RQMC methods is presented for the task of evaluating
the joint log-density function. This allows the parameter estimation task to be
accomplished via an EM-like algorithm where all weights and log-densities are
numerically estimated. It is demonstrated through numerical examples that the
suggested algorithms are quite fast; even for high dimensions around 1000 the
distribution function can be estimated with moderate accuracy using only a few
seconds of run time. Even log-densities around -100 can be estimated accurately
and quickly. An implementation of all algorithms presented in this work is
available in the R package nvmix (version >= 0.0.4).
| stat |
BNAS:An Efficient Neural Architecture Search Approach Using Broad
Scalable Architecture | In this paper, we propose Broad Neural Architecture Search (BNAS) where we
elaborately design broad scalable architecture dubbed Broad Convolutional
Neural Network (BCNN) to solve the above issue. On one hand, the proposed broad
scalable architecture has fast training speed due to its shallow topology.
Moreover, we also adopt reinforcement learning and parameter sharing used in
ENAS as the optimization strategy of BNAS. Hence, the proposed approach can
achieve higher search efficiency. On the other hand, the broad scalable
architecture extracts multi-scale features and enhancement representations, and
feeds them into global average pooling layer to yield more reasonable and
comprehensive representations. Therefore, the performance of broad scalable
architecture can be promised. In particular, we also develop two variants for
BNAS who modify the topology of BCNN. In order to verify the effectiveness of
BNAS, several experiments are performed and experimental results show that 1)
BNAS delivers 0.19 days which is 2.37x less expensive than ENAS who ranks the
best in reinforcement learning-based NAS approaches, 2) compared with
small-size (0.5 millions parameters) and medium-size (1.1 millions parameters)
models, the architecture learned by BNAS obtains state-of-the-art performance
(3.58% and 3.24% test error) on CIFAR-10, 3) the learned architecture achieves
25.3% top-1 error on ImageNet just using 3.9 millions parameters.
| stat |
Stagewise Training Accelerates Convergence of Testing Error Over SGD | Stagewise training strategy is widely used for learning neural networks,
which runs a stochastic algorithm (e.g., SGD) starting with a relatively large
step size (aka learning rate) and geometrically decreasing the step size after
a number of iterations. It has been observed that the stagewise SGD has much
faster convergence than the vanilla SGD with a polynomially decaying step size
in terms of both training error and testing error. {\it But how to explain this
phenomenon has been largely ignored by existing studies.} This paper provides
some theoretical evidence for explaining this faster convergence. In
particular, we consider a stagewise training strategy for minimizing empirical
risk that satisfies the Polyak-\L ojasiewicz (PL) condition, which has been
observed/proved for neural networks and also holds for a broad family of convex
functions. For convex loss functions and two classes of "nice-behaviored"
non-convex objectives that are close to a convex function, we establish faster
convergence of stagewise training than the vanilla SGD under the PL condition
on both training error and testing error. Experiments on stagewise learning of
deep residual networks exhibits that it satisfies one type of non-convexity
assumption and therefore can be explained by our theory. Of independent
interest, the testing error bounds for the considered non-convex loss functions
are dimensionality and norm independent.
| stat |
Non-reversibly updating a uniform [0,1] value for Metropolis
accept/reject decisions | I show how it can be beneficial to express Metropolis accept/reject decisions
in terms of comparison with a uniform [0,1] value, u, and to then update u
non-reversibly, as part of the Markov chain state, rather than sampling it
independently each iteration. This provides a small improvement for random walk
Metropolis and Langevin updates in high dimensions. It produces a larger
improvement when using Langevin updates with persistent momentum, giving
performance comparable to that of Hamiltonian Monte Carlo (HMC) with long
trajectories. This is of significance when some variables are updated by other
methods, since if HMC is used, these updates can be done only between
trajectories, whereas they can be done more often with Langevin updates. I
demonstrate that for a problem with some continuous variables, updated by HMC
or Langevin updates, and also discrete variables, updated by Gibbs sampling
between updates of the continuous variables, Langevin with persistent momentum
and non-reversible updates to u samples nearly a factor of two more efficiently
than HMC. Benefits are also seen for a Bayesian neural network model in which
hyperparameters are updated by Gibbs sampling.
| stat |
Statistical Analysis of Modern Reliability Data | Traditional reliability analysis has been using time to event data,
degradation data, and recurrent event data, while the associated covariates
tend to be simple and constant over time. Over the past years, we have
witnessed the rapid development of sensor and wireless technology, which
enables us to track how the product has been used and under which environmental
conditions it has been used. Nowadays, we are able to collect richer
information on covariates which provides opportunities for better reliability
predictions. In this chapter, we first review recent development on statistical
methods for reliability analysis. We then focus on introducing several specific
methods that were developed for different types of reliability data with
covariate information. Illustrations of those methods are also provided using
examples from industry. Test planning is also an important part of reliability
analysis. In addition to data analysis, we also provide a briefly review on
recent developments of test planning and then focus on illustrating the
sequential Bayesian design with an example of fatigue testing for polymer
composites. The paper is concluded with some discussions and remarks.
| stat |
Testing for Stochastic Order in Interval-Valued Data | We construct a procedure to test the stochastic order of two samples of
interval-valued data. We propose a test statistic which belongs to U-statistic
and derive its asymptotic distribution under the null hypothesis. We compare
the performance of the newly proposed method with the existing one-sided
bivariate Kolmogorov-Smirnov test using real data and simulated data.
| stat |
Kernel machines that adapt to GPUs for effective large batch training | Modern machine learning models are typically trained using Stochastic
Gradient Descent (SGD) on massively parallel computing resources such as GPUs.
Increasing mini-batch size is a simple and direct way to utilize the parallel
computing capacity. For small batch an increase in batch size results in the
proportional reduction in the training time, a phenomenon known as linear
scaling. However, increasing batch size beyond a certain value leads to no
further improvement in training time. In this paper we develop the first
analytical framework that extends linear scaling to match the parallel
computing capacity of a resource. The framework is designed for a class of
classical kernel machines. It automatically modifies a standard kernel machine
to output a mathematically equivalent prediction function, yet allowing for
extended linear scaling, i.e., higher effective parallelization and faster
training time on given hardware.
The resulting algorithms are accurate, principled and very fast. For example,
using a single Titan Xp GPU, training on ImageNet with $1.3\times 10^6$ data
points and $1000$ labels takes under an hour, while smaller datasets, such as
MNIST, take seconds. As the parameters are chosen analytically, based on the
theoretical bounds, little tuning beyond selecting the kernel and the kernel
parameter is needed, further facilitating the practical use of these methods.
| stat |
Know Your Boundaries: Constraining Gaussian Processes by Variational
Harmonic Features | Gaussian processes (GPs) provide a powerful framework for extrapolation,
interpolation, and noise removal in regression and classification. This paper
considers constraining GPs to arbitrarily-shaped domains with boundary
conditions. We solve a Fourier-like generalised harmonic feature representation
of the GP prior in the domain of interest, which both constrains the GP and
attains a low-rank representation that is used for speeding up inference. The
method scales as $\mathcal{O}(nm^2)$ in prediction and $\mathcal{O}(m^3)$ in
hyperparameter learning for regression, where $n$ is the number of data points
and $m$ the number of features. Furthermore, we make use of the variational
approach to allow the method to deal with non-Gaussian likelihoods. The
experiments cover both simulated and empirical data in which the boundary
conditions allow for inclusion of additional physical information.
| stat |
Missing Value Imputation for Mixed Data via Gaussian Copula | Missing data imputation forms the first critical step of many data analysis
pipelines. The challenge is greatest for mixed data sets, including real,
Boolean, and ordinal data, where standard techniques for imputation fail basic
sanity checks: for example, the imputed values may not follow the same
distributions as the data. This paper proposes a new semiparametric algorithm
to impute missing values, with no tuning parameters. The algorithm models mixed
data as a Gaussian copula. This model can fit arbitrary marginals for
continuous variables and can handle ordinal variables with many levels,
including Boolean variables as a special case. We develop an efficient
approximate EM algorithm to estimate copula parameters from incomplete mixed
data. The resulting model reveals the statistical associations among variables.
Experimental results on several synthetic and real datasets show superiority of
our proposed algorithm to state-of-the-art imputation algorithms for mixed
data.
| stat |
A Stable Cardinality Distance for Topological Classification | This work incorporates topological features via persistence diagrams to
classify point cloud data arising from materials science. Persistence diagrams
are multisets summarizing the connectedness and holes of given data. A new
distance on the space of persistence diagrams generates relevant input features
for a classification algorithm for materials science data. This distance
measures the similarity of persistence diagrams using the cost of matching
points and a regularization term corresponding to cardinality differences
between diagrams. Establishing stability properties of this distance provides
theoretical justification for the use of the distance in comparisons of such
diagrams. The classification scheme succeeds in determining the crystal
structure of materials on noisy and sparse data retrieved from synthetic atom
probe tomography experiments.
| stat |
Identification of Model Uncertainty via Optimal Design of Experiments
Applied to a Mechanical Press | In engineering applications almost all processes are described with the help
of models. Especially forming machines heavily rely on mathematical models for
control and condition monitoring. Inaccuracies during the modeling,
manufacturing and assembly of these machines induce model uncertainty which
impairs the controller's performance. In this paper we propose an approach to
identify model uncertainty using parameter identification, optimal design of
experiments and hypothesis testing. The experimental setup is characterized by
optimal sensor positions such that specific model parameters can be determined
with minimal variance. This allows for the computation of confidence regions in
which the real parameters or the parameter estimates from different test sets
have to lie. We claim that inconsistencies in the estimated parameter values,
considering their approximated confidence ellipsoids as well, cannot be
explained by data uncertainty but are indicators of model uncertainty. The
proposed method is demonstrated using a component of the 3D Servo Press, a
multi-technology forming machine that combines spindles with eccentric servo
drives.
| stat |
DERGMs: Degeneracy-restricted exponential random graph models | Exponential random graph models, or ERGMs, are a flexible class of models for
networks. Recent work highlights difficulties related to the models' ill
behavior, dubbed `degeneracy', such as most of the probability mass being
concentrated on a very small subset of the parameter space. This behavior
limits both the applicability of an ERGM as a model for real data and parameter
estimation via the usual MCMC algorithms.
To address this problem, we propose a new exponential family of models for
random graphs that build on the standard ERGM framework. We resolve the
degenerate model behavior by an interpretable support restriction. Namely, we
introduce a new parameter based on the graph-theoretic notion of degeneracy, a
measure of sparsity whose value is low in real-worlds networks. We prove this
support restriction does not eliminate too many graphs from the support of an
ERGM, and we also prove that degeneracy of a model is captured precisely by
stability of its sufficient statistics. We show examples of ERGMs that are
degenerate whose counterpart DERGMs are not, both theoretically and by
simulations, and we test our model class on a set of real world networks.
| stat |
A Neural Network Based on First Principles | In this paper, a Neural network is derived from first principles, assuming
only that each layer begins with a linear dimension-reducing transformation.
The approach appeals to the principle of Maximum Entropy (MaxEnt) to find the
posterior distribution of the input data of each layer, conditioned on the
layer output variables. This posterior has a well-defined mean, the conditional
mean estimator, that is calculated using a type of neural network with
theoretically-derived activation functions similar to sigmoid, softplus, and
relu. This implicitly provides a theoretical justification for their use. A
theorem that finds the conditional distribution and conditional mean estimator
under the MaxEnt prior is proposed, unifying results for special cases.
Combining layers results in an auto-encoder with conventional feed-forward
analysis network and a type of linear Bayesian belief network in the
reconstruction path.
| stat |
TreeGrad: Transferring Tree Ensembles to Neural Networks | Gradient Boosting Decision Tree (GBDT) are popular machine learning
algorithms with implementations such as LightGBM and in popular machine
learning toolkits like Scikit-Learn. Many implementations can only produce
trees in an offline manner and in a greedy manner. We explore ways to convert
existing GBDT implementations to known neural network architectures with
minimal performance loss in order to allow decision splits to be updated in an
online manner and provide extensions to allow splits points to be altered as a
neural architecture search problem. We provide learning bounds for our neural
network.
| stat |
A Versatile Estimation Procedure without Estimating the Nonignorable
Missingness Mechanism | We consider the estimation problem in a regression setting where the outcome
variable is subject to nonignorable missingness and identifiability is ensured
by the shadow variable approach. We propose a versatile estimation procedure
where modeling of missingness mechanism is completely bypassed. We show that
our estimator is easy to implement and we derive the asymptotic theory of the
proposed estimator. We also investigate some alternative estimators under
different scenarios. Comprehensive simulation studies are conducted to
demonstrate the finite sample performance of the method. We apply the estimator
to a children's mental health study to illustrate its usefulness.
| stat |
Class-Weighted Classification: Trade-offs and Robust Approaches | We address imbalanced classification, the problem in which a label may have
low marginal probability relative to other labels, by weighting losses
according to the correct class. First, we examine the convergence rates of the
expected excess weighted risk of plug-in classifiers where the weighting for
the plug-in classifier and the risk may be different. This leads to irreducible
errors that do not converge to the weighted Bayes risk, which motivates our
consideration of robust risks. We define a robust risk that minimizes risk over
a set of weightings and show excess risk bounds for this problem. Finally, we
show that particular choices of the weighting set leads to a special instance
of conditional value at risk (CVaR) from stochastic programming, which we call
label conditional value at risk (LCVaR). Additionally, we generalize this
weighting to derive a new robust risk problem that we call label heterogeneous
conditional value at risk (LHCVaR). Finally, we empirically demonstrate the
efficacy of LCVaR and LHCVaR on improving class conditional risks.
| stat |
How to Initialize your Network? Robust Initialization for WeightNorm &
ResNets | Residual networks (ResNet) and weight normalization play an important role in
various deep learning applications. However, parameter initialization
strategies have not been studied previously for weight normalized networks and,
in practice, initialization methods designed for un-normalized networks are
used as a proxy. Similarly, initialization for ResNets have also been studied
for un-normalized networks and often under simplified settings ignoring the
shortcut connection. To address these issues, we propose a novel parameter
initialization strategy that avoids explosion/vanishment of information across
layers for weight normalized networks with and without residual connections.
The proposed strategy is based on a theoretical analysis using mean field
approximation. We run over 2,500 experiments and evaluate our proposal on image
datasets showing that the proposed initialization outperforms existing
initialization methods in terms of generalization performance, robustness to
hyper-parameter values and variance between seeds, especially when networks get
deeper in which case existing methods fail to even start training. Finally, we
show that using our initialization in conjunction with learning rate warmup is
able to reduce the gap between the performance of weight normalized and batch
normalized networks.
| stat |
Versatile linkage: a family of space-conserving strategies for
agglomerative hierarchical clustering | Agglomerative hierarchical clustering can be implemented with several
strategies that differ in the way elements of a collection are grouped together
to build a hierarchy of clusters. Here we introduce versatile linkage, a new
infinite system of agglomerative hierarchical clustering strategies based on
generalized means, which go from single linkage to complete linkage, passing
through arithmetic average linkage and other clustering methods yet unexplored
such as geometric linkage and harmonic linkage. We compare the different
clustering strategies in terms of cophenetic correlation, mean absolute error,
and also tree balance and space distortion, two new measures proposed to
describe hierarchical trees. Unlike the $\beta$-flexible clustering system, we
show that the versatile linkage family is space-conserving.
| stat |
Individual Claims Forecasting with Bayesian Mixture Density Networks | We introduce an individual claims forecasting framework utilizing Bayesian
mixture density networks that can be used for claims analytics tasks such as
case reserving and triaging. The proposed approach enables incorporating claims
information from both structured and unstructured data sources, producing
multi-period cash flow forecasts, and generating different scenarios of future
payment patterns. We implement and evaluate the modeling framework using
publicly available data.
| stat |
Empirical Likelihood Inference With Public-Use Survey Data | Public-use survey data are an important source of information for researchers
in social science and health studies to build statistical models and make
inferences on the target finite population. This paper presents two general
inferential tools through the pseudo empirical likelihood and the sample
empirical likelihood methods. Theoretical results on point estimation and
linear or nonlinear hypothesis tests involving parameters defined through
estimating equations are established, and practical issues with the
implementation of the proposed methods are discussed. Results from simulation
studies and an application to the 2016 General Social Survey dataset of
Statistics Canada show that the proposed methods work well under different
scenarios. The inferential procedures and theoretical results presented in the
paper make the empirical likelihood a practically useful tool for users of
complex survey data.
| stat |
Particle-Gibbs Sampling For Bayesian Feature Allocation Models | Bayesian feature allocation models are a popular tool for modelling data with
a combinatorial latent structure. Exact inference in these models is generally
intractable and so practitioners typically apply Markov Chain Monte Carlo
(MCMC) methods for posterior inference. The most widely used MCMC strategies
rely on an element wise Gibbs update of the feature allocation matrix. These
element wise updates can be inefficient as features are typically strongly
correlated. To overcome this problem we have developed a Gibbs sampler that can
update an entire row of the feature allocation matrix in a single move.
However, this sampler is impractical for models with a large number of features
as the computational complexity scales exponentially in the number of features.
We develop a Particle Gibbs sampler that targets the same distribution as the
row wise Gibbs updates, but has computational complexity that only grows
linearly in the number of features. We compare the performance of our proposed
methods to the standard Gibbs sampler using synthetic data from a range of
feature allocation models. Our results suggest that row wise updates using the
PG methodology can significantly improve the performance of samplers for
feature allocation models.
| stat |
M-estimation in GARCH models without higher order moments | We consider a class of M-estimators of the parameters of the GARCH models
which are asymptotically normal under mild assumptions on the moments of the
underlying error distribution. Since heavy-tailed error distributions without
higher order moments are common in the GARCH modeling of many real financial
data, it becomes worthwhile to use such estimators for the time series
inference instead of the quasi maximum likelihood estimator. We discuss the
weighted bootstrap approximations of the distributions of M-estimators. Through
extensive simulations and data analysis, we demonstrate the robustness of the
M-estimators under heavy-tailed error distributions and the accuracy of the
bootstrap approximation. In addition to the GARCH~(1, 1) model, we obtain
extensive computation and simulation results which are useful in the context of
higher order models such as GARCH~(2, 1) and GARCH~(1, 2) but have not yet
received sufficient attention in the literature. Finally, we use M-estimators
for the analysis of three real financial time series fitted with GARCH~(1, 1)
or GARCH~(2, 1) models.
| stat |
Adaptive Nonparametric Psychophysics | We introduce a new set of models and adaptive psychometric testing methods
for multidimensional psychophysics. In contrast to traditional adaptive
staircase methods like PEST and QUEST, the method is multi-dimensional and does
not require a grid over contextual dimensions, retaining sub-exponential
scaling in the number of stimulus dimensions. In contrast to more recent
multi-dimensional adaptive methods, our underlying model does not require a
parametric assumption about the interaction between intensity and the
additional dimensions. In addition, we introduce a new active sampling policy
that explicitly targets psychometric detection threshold estimation and does so
substantially faster than policies that attempt to estimate the full
psychometric function (though it still provides estimates of the function,
albeit with lower accuracy). Finally, we introduce AEPsych, a user-friendly
open-source package for nonparametric psychophysics that makes these
technically-challenging methods accessible to the broader community.
| stat |
Accounting for spatial varying sampling effort due to accessibility in
Citizen Science data: A case study of moose in Norway | Citizen Scientists together with an increasing access to technology provide
large datasets that can be used to study e.g. ecology and biodiversity. Unknown
and varying sampling effort is a major issue when making inference based on
citizen science data. In this paper we propose a modeling approach for
accounting for variation in sampling effort due to accessibility. The paper is
based on a illustrative case study using citizen science data of moose
occurrence in Hedmark, Norway. The aim is to make inference about the
importance of two geographical properties known to influence moose occurrence;
terrain ruggedness index and solar radiation. Explanatory analysis show that
moose occurrences are overrepresented close to roads, and we use distance to
roads as a proxy for accessibility. We propose a model based on a Bayesian
Log-Gaussian Cox Process specification for occurrence. The model accounts for
accessibility through a distance sampling approach. This approach can be seen
as a thinning process where probability of thinning, i.e. not observing,
increases with increasing distances. For the moose case study distance to roads
are used. Computationally efficient full Bayesian inference is performed using
the Integrated Nested Laplace Approximation and the Stochastic Partial
Differential Equation approach for spatial modeling. The proposed model as well
as the consequences of not accounting for varying sampling effort due to
accessibility are studied through a simulation study based on the case study.
Considerable biases are found in estimates for the effect of radiation on moose
occurrence when accessibility is not considered in the model.
| stat |
Deep orthogonal linear networks are shallow | We consider the problem of training a deep orthogonal linear network, which
consists of a product of orthogonal matrices, with no non-linearity in-between.
We show that training the weights with Riemannian gradient descent is
equivalent to training the whole factorization by gradient descent. This means
that there is no effect of overparametrization and implicit bias at all in this
setting: training such a deep, overparametrized, network is perfectly
equivalent to training a one-layer shallow network.
| stat |
Dimension-Robust MCMC in Bayesian Inverse Problems | The methodology developed in this article is motivated by a wide range of
prediction and uncertainty quantification problems that arise in Statistics,
Machine Learning and Applied Mathematics, such as non-parametric regression,
multi-class classification and inversion of partial differential equations. One
popular formulation of such problems is as Bayesian inverse problems, where a
prior distribution is used to regularize inference on a high-dimensional latent
state, typically a function or a field. It is common that such priors are
non-Gaussian, for example piecewise-constant or heavy-tailed, and/or
hierarchical, in the sense of involving a further set of low-dimensional
parameters, which, for example, control the scale or smoothness of the latent
state. In this formulation prediction and uncertainty quantification relies on
efficient exploration of the posterior distribution of latent states and
parameters. This article introduces a framework for efficient MCMC sampling in
Bayesian inverse problems that capitalizes upon two fundamental ideas in MCMC,
non-centred parameterisations of hierarchical models and dimension-robust
samplers for latent Gaussian processes. Using a range of diverse applications
we showcase that the proposed framework is dimension-robust, that is, the
efficiency of the MCMC sampling does not deteriorate as the dimension of the
latent state gets higher. We showcase the full potential of the machinery we
develop in the article in semi-supervised multi-class classification, where our
sampling algorithm is used within an active learning framework to guide the
selection of input data to manually label in order to achieve high predictive
accuracy with a minimal number of labelled data.
| stat |
Compressed sensing reconstruction using Expectation Propagation | Many interesting problems in fields ranging from telecommunications to
computational biology can be formalized in terms of large underdetermined
systems of linear equations with additional constraints or regularizers. One of
the most studied ones, the Compressed Sensing problem (CS), consists in finding
the solution with the smallest number of non-zero components of a given system
of linear equations $\boldsymbol y = \mathbf{F} \boldsymbol{w}$ for known
measurement vector $\boldsymbol{y}$ and sensing matrix $\mathbf{F}$. Here, we
will address the compressed sensing problem within a Bayesian inference
framework where the sparsity constraint is remapped into a singular prior
distribution (called Spike-and-Slab or Bernoulli-Gauss). Solution to the
problem is attempted through the computation of marginal distributions via
Expectation Propagation (EP), an iterative computational scheme originally
developed in Statistical Physics. We will show that this strategy is
comparatively more accurate than the alternatives in solving instances of CS
generated from statistically correlated measurement matrices. For computational
strategies based on the Bayesian framework such as variants of Belief
Propagation, this is to be expected, as they implicitly rely on the hypothesis
of statistical independence among the entries of the sensing matrix. Perhaps
surprisingly, the method outperforms uniformly also all the other
state-of-the-art methods in our tests.
| stat |
Modeling short-ranged dependence in block extrema with application to
polar temperature data | The block maxima approach is an important method in univariate extreme value
analysis. While assuming that block maxima are independent results in
straightforward analysis, the resulting inferences maybe invalid when a series
of block maxima exhibits dependence. We propose a model, based on a first-order
Markov assumption, that incorporates dependence between successive block maxima
through the use of a bivariate logistic dependence structure while maintaining
generalized extreme value (GEV) marginal distributions. Modeling dependence in
this manner allows us to better estimate extreme quantiles when block maxima
exhibit short-ranged dependence. We demonstrate via a simulation study that our
first-order Markov GEV model performs well when successive block maxima are
dependent, while still being reasonably robust when maxima are independent. We
apply our method to two polar annual minimum air temperature data sets that
exhibit short-ranged dependence structures, and find that the proposed model
yields modified estimates of high quantiles.
| stat |
Regression on imperfect class labels derived by unsupervised clustering | Outcome regressed on class labels identified by unsupervised clustering is
custom in many applications. However, it is common to ignore the
misclassification of class labels caused by the learning algorithm, which
potentially leads to serious bias of the estimated effect parameters. Due to
its generality we suggest to redress the situation by use of the simulation and
extrapolation method. Performance is illustrated by simulated data from
Gaussian mixture models. Finally, we apply our method to a study which
regressed overall survival on class labels derived from unsupervised clustering
of gene expression data from bone marrow samples of multiple myeloma patients.
| stat |
Adaptive Schemes for Piecewise Deterministic Monte Carlo Algorithms | The Bouncy Particle sampler (BPS) and the Zig-Zag sampler (ZZS) are
continuous time, non-reversible Monte Carlo methods based on piecewise
deterministic Markov processes. Experiments show that the speed of convergence
of these samplers can be affected by the shape of the target distribution, as
for instance in the case of anisotropic targets. We propose an adaptive scheme
that iteratively learns all or part of the covariance matrix of the target and
takes advantage of the obtained information to modify the underlying process
with the aim of increasing the speed of convergence. Moreover, we define an
adaptive scheme that automatically tunes the refreshment rate of the BPS or
ZZS. We prove ergodicity and a law of large numbers for all the proposed
adaptive algorithms. Finally, we show the benefits of the adaptive samplers
with several numerical simulations.
| stat |
Nested Gaussian filters for recursive Bayesian inference and nonlinear
tracking in state space models | We introduce a new sequential methodology to calibrate the fixed parameters
and track the stochastic dynamical variables of a state-space system. The
proposed method is based on the nested hybrid filtering (NHF) framework of [1],
that combines two layers of filters, one inside the other, to compute the joint
posterior probability distribution of the static parameters and the state
variables. In particular, we explore the use of deterministic sampling
techniques for Gaussian approximation in the first layer of the algorithm,
instead of the Monte Carlo methods employed in the original procedure. The
resulting scheme reduces the computational cost and so makes the algorithms
potentially better-suited for high-dimensional state and parameter spaces. We
describe a specific instance of the new method and then study its performance
and efficiency of the resulting algorithms for a stochastic Lorenz 63 model
with uncertain parameters.
| stat |
Extrapolation Towards Imaginary $0$-Nearest Neighbour and Its Improved
Convergence Rate | $k$-nearest neighbour ($k$-NN) is one of the simplest and most widely-used
methods for supervised classification, that predicts a query's label by taking
weighted ratio of observed labels of $k$ objects nearest to the query. The
weights and the parameter $k \in \mathbb{N}$ regulate its bias-variance
trade-off, and the trade-off implicitly affects the convergence rate of the
excess risk for the $k$-NN classifier; several existing studies considered
selecting optimal $k$ and weights to obtain faster convergence rate. Whereas
$k$-NN with non-negative weights has been developed widely, it was also proved
that negative weights are essential for eradicating the bias terms and
attaining optimal convergence rate. In this paper, we propose a novel
multiscale $k$-NN (MS-$k$-NN), that extrapolates unweighted $k$-NN estimators
from several $k \ge 1$ values to $k=0$, thus giving an imaginary 0-NN
estimator. Our method implicitly computes optimal real-valued weights that are
adaptive to the query and its neighbour points. We theoretically prove that the
MS-$k$-NN attains the improved rate, which coincides with the existing optimal
rate under some conditions.
| stat |
Minimax optimal rates for Mondrian trees and forests | Introduced by Breiman, Random Forests are widely used classification and
regression algorithms. While being initially designed as batch algorithms,
several variants have been proposed to handle online learning. One particular
instance of such forests is the \emph{Mondrian Forest}, whose trees are built
using the so-called Mondrian process, therefore allowing to easily update their
construction in a streaming fashion. In this paper, we provide a thorough
theoretical study of Mondrian Forests in a batch learning setting, based on new
results about Mondrian partitions. Our results include consistency and
convergence rates for Mondrian Trees and Forests, that turn out to be minimax
optimal on the set of $s$-H\"older function with $s \in (0,1]$ (for trees and
forests) and $s \in (1,2]$ (for forests only), assuming a proper tuning of
their complexity parameter in both cases. Furthermore, we prove that an
adaptive procedure (to the unknown $s \in (0, 2]$) can be constructed by
combining Mondrian Forests with a standard model aggregation algorithm. These
results are the first demonstrating that some particular random forests achieve
minimax rates \textit{in arbitrary dimension}. Owing to their remarkably simple
distributional properties, which lead to minimax rates, Mondrian trees are a
promising basis for more sophisticated yet theoretically sound random forests
variants.
| stat |
Power laws in the Roman Empire: a survival analysis | The Roman Empire shaped Western civilization, and many Roman principles are
embodied in modern institutions. Although its political institutions proved
both resilient and adaptable, allowing it to incorporate diverse populations,
the Empire suffered from many internal conflicts. Indeed, most emperors died
violently, from assassination, suicide, or in battle. These internal conflicts
produced patterns in the length of time that can be identified by statistical
analysis. In this paper, we study the underlying patterns associated with the
reign of the Roman emperors by using statistical tools of survival data
analysis. We consider all the 175 Roman emperors and propose a new power-law
model with change points to predict the time-to-violent-death of the Roman
emperors. This model encompasses data in the presence of censoring and
long-term survivors, providing more accurate predictions than previous models.
Our results show that power-law distributions can also occur in survival data,
as verified in other data types from natural and artificial systems,
reinforcing the ubiquity of power law distributions. The generality of our
approach paves the way to further related investigations not only in other
ancient civilizations but also in applications in engineering and medicine.
| stat |
Bayesian nonparametric modelling of sequential discoveries | We aim at modelling the appearance of distinct tags in a sequence of labelled
objects. Common examples of this type of data include words in a corpus or
distinct species in a sample. These sequential discoveries are often summarised
via accumulation curves, which count the number of distinct entities observed
in an increasingly large set of objects. We propose a novel Bayesian
nonparametric method for species sampling modelling by directly specifying the
probability of a new discovery, therefore allowing for flexible specifications.
The asymptotic behavior and finite sample properties of such an approach are
extensively studied. Interestingly, our enlarged class of sequential processes
includes highly tractable special cases. We present a subclass of models
characterized by appealing theoretical and computational properties. Moreover,
due to strong connections with logistic regression models, the latter subclass
can naturally account for covariates. We finally test our proposal on both
synthetic and real data, with special emphasis on a large fungal biodiversity
study in Finland.
| stat |
Assessing the Significance of Model Selection in Ecology | Model Selection is a key part of many ecological studies, with Akaike's
Information Criterion the most commonly used technique. Typically, a number of
candidate models are defined a priori and ranked according to their expected
out-of-sample performance. Model selection, however, only assesses the relative
performance of the models and, as pointed out in a recent paper, a large
proportion of ecology papers that use model selection do not assess the
absolute fit of the `best' model. In this paper, it is argued that assessing
the absolute fit of the `best' model alone does not go far enough. This is
because a model that appears to perform well under model selection is also
likely to appear to perform well under measures of absolute fit, even when
there is no predictive value. A model selection permutation test is proposed
that assesses the probability that the model selection statistic of the `best'
model could have occurred by chance alone, whilst taking account of
dependencies between the models. It is argued that this test should always be
performed before formal model selection takes place. The test is demonstrated
on two real population modelling examples of ibex in northern Italy and wild
reindeer in Norway.
| stat |
Inference on weighted average value function in high-dimensional state
space | This paper gives a consistent, asymptotically normal estimator of the
expected value function when the state space is high-dimensional and the
first-stage nuisance functions are estimated by modern machine learning tools.
First, we show that value function is orthogonal to the conditional choice
probability, therefore, this nuisance function needs to be estimated only at
$n^{-1/4}$ rate. Second, we give a correction term for the transition density
of the state variable. The resulting orthogonal moment is robust to
misspecification of the transition density and does not require this nuisance
function to be consistently estimated. Third, we generalize this result by
considering the weighted expected value. In this case, the orthogonal moment is
doubly robust in the transition density and additional second-stage nuisance
functions entering the correction term. We complete the asymptotic theory by
providing bounds on second-order asymptotic terms.
| stat |
Online Gaussian LDA for Unsupervised Pattern Mining from Utility Usage
Data | Non-intrusive load monitoring (NILM) aims at separating a whole-home energy
signal into its appliance components. Such method can be harnessed to provide
various services to better manage and control energy consumption (optimal
planning and saving). NILM has been traditionally approached from signal
processing and electrical engineering perspectives. Recently, machine learning
has started to play an important role in NILM. While most work has focused on
supervised algorithms, unsupervised approaches can be more interesting and of
practical use in real case scenarios. Specifically, they do not require
labelled training data to be acquired from individual appliances and the
algorithm can be deployed to operate on the measured aggregate data directly.
In this paper, we propose a fully unsupervised NILM framework based on Bayesian
hierarchical mixture models. In particular, we develop a new method based on
Gaussian Latent Dirichlet Allocation (GLDA) in order to extract global
components that summarise the energy signal. These components provide a
representation of the consumption patterns. Designed to cope with big data, our
algorithm, unlike existing NILM ones, does not focus on appliance recognition.
To handle this massive data, GLDA works online. Another novelty of this work
compared to the existing NILM is that the data involves different utilities
(e.g, electricity, water and gas) as well as some sensors measurements.
Finally, we propose different evaluation methods to analyse the results which
show that our algorithm finds useful patterns.
| stat |
Combining Smoothing Spline with Conditional Gaussian Graphical Model for
Density and Graph Estimation | Multivariate density estimation and graphical models play important roles in
statistical learning. The estimated density can be used to construct a
graphical model that reveals conditional relationships whereas a graphical
structure can be used to build models for density estimation. Our goal is to
construct a consolidated framework that can perform both density and graph
estimation. Denote $\bm{Z}$ as the random vector of interest with density
function $f(\bz)$. Splitting $\bm{Z}$ into two parts,
$\bm{Z}=(\bm{X}^T,\bm{Y}^T)^T$ and writing $f(\bz)=f(\bx)f(\by|\bx)$ where
$f(\bx)$ is the density function of $\bm{X}$ and $f(\by|\bx)$ is the
conditional density of $\bm{Y}|\bm{X}=\bx$. We propose a semiparametric
framework that models $f(\bx)$ nonparametrically using a smoothing spline ANOVA
(SS ANOVA) model and $f(\by|\bx)$ parametrically using a conditional Gaussian
graphical model (cGGM). Combining flexibility of the SS ANOVA model with
succinctness of the cGGM, this framework allows us to deal with
high-dimensional data without assuming a joint Gaussian distribution. We
propose a backfitting estimation procedure for the cGGM with a computationally
efficient approach for selection of tuning parameters. We also develop a
geometric inference approach for edge selection. We establish asymptotic
convergence properties for both the parameter and density estimation. The
performance of the proposed method is evaluated through extensive simulation
studies and two real data applications.
| stat |
Fenton-Wilkinson Order Statistics and German Tanks: A Case Study of an
Orienteering Relay Race | Ordinal regression falls between discrete-valued classification and
continuous-valued regression. Ordinal target variables can be associated with
ranked random variables. These random variables are known as order statistics
and they are closely related to ordinal regression. However, the challenge of
using order statistics for ordinal regression prediction is finding a suitable
parent distribution. In this work, we provide a case study of a real-world
orienteering relay race by viewing it as a random process. For this process, we
show that accurate order statistical ordinal regression predictions of final
team rankings, or places, can be obtained by assuming a lognormal distribution
of individual leg times. Moreover, we apply Fenton-Wilkinson approximations to
intermediate changeover times alongside an estimator for the total number of
teams as in the notorious German tank problem. The purpose of this work is, in
part, to spark interest in studying the applicability of order statistics in
ordinal regression problems.
| stat |
Safe Policy Search with Gaussian Process Models | We propose a method to optimise the parameters of a policy which will be used
to safely perform a given task in a data-efficient manner. We train a Gaussian
process model to capture the system dynamics, based on the PILCO framework. Our
model has useful analytic properties, which allow closed form computation of
error gradients and estimating the probability of violating given state space
constraints. During training, as well as operation, only policies that are
deemed safe are implemented on the real system, minimising the risk of failure.
| stat |
Autoregressive flow-based causal discovery and inference | We posit that autoregressive flow models are well-suited to performing a
range of causal inference tasks - ranging from causal discovery to making
interventional and counterfactual predictions. In particular, we exploit the
fact that autoregressive architectures define an ordering over variables,
analogous to a causal ordering, in order to propose a single flow architecture
to perform all three aforementioned tasks. We first leverage the fact that flow
models estimate normalized log-densities of data to derive a bivariate measure
of causal direction based on likelihood ratios. Whilst traditional measures of
causal direction often require restrictive assumptions on the nature of causal
relationships (e.g., linearity),the flexibility of flow models allows for
arbitrary causal dependencies. Our approach compares favourably against
alternative methods on synthetic data as well as on the Cause-Effect Pairs
bench-mark dataset. Subsequently, we demonstrate that the invertible nature of
flows naturally allows for direct evaluation of both interventional and
counterfactual predictions, which require marginalization and conditioning over
latent variables respectively. We present examples over synthetic data where
autoregressive flows, when trained under the correct causal ordering, are able
to make accurate interventional and counterfactual predictions
| stat |
Lipschitz regularity of deep neural networks: analysis and efficient
estimation | Deep neural networks are notorious for being sensitive to small well-chosen
perturbations, and estimating the regularity of such architectures is of utmost
importance for safe and robust practical applications. In this paper, we
investigate one of the key characteristics to assess the regularity of such
methods: the Lipschitz constant of deep learning architectures. First, we show
that, even for two layer neural networks, the exact computation of this
quantity is NP-hard and state-of-art methods may significantly overestimate it.
Then, we both extend and improve previous estimation methods by providing
AutoLip, the first generic algorithm for upper bounding the Lipschitz constant
of any automatically differentiable function. We provide a power method
algorithm working with automatic differentiation, allowing efficient
computations even on large convolutions. Second, for sequential neural
networks, we propose an improved algorithm named SeqLip that takes advantage of
the linear computation graph to split the computation per pair of consecutive
layers. Third we propose heuristics on SeqLip in order to tackle very large
networks. Our experiments show that SeqLip can significantly improve on the
existing upper bounds. Finally, we provide an implementation of AutoLip in the
PyTorch environment that may be used to better estimate the robustness of a
given neural network to small perturbations or regularize it using more precise
Lipschitz estimations.
| stat |
Sufficient Dimension Reduction for Interactions | Dimension reduction lies at the heart of many statistical methods. In
regression, dimension reduction has been linked to the notion of sufficiency
whereby the relation of the response to a set of predictors is explained by a
lower dimensional subspace in the predictor space. In this paper, we consider
the notion of a dimension reduction in regression on subspaces that are
sufficient to explain interaction effects between predictors and another
variable of interest. The motivation for this work is from precision medicine
where the performance of an individualized treatment rule, given a set of
pretreatment predictors, is determined by interaction effects.
| stat |
The non-tightness of the reconstruction threshold of a 4 states
symmetric model with different in-block and out-block mutations | The tree reconstruction problem is to collect and analyze massive data at the
$n$th level of the tree, to identify whether there is non-vanishing information
of the root, as $n$ goes to infinity. Its connection to the clustering problem
in the setting of the stochastic block model, which has wide applications in
machine learning and data mining, has been well established. For the stochastic
block model, an "information-theoretically-solvable-but-computationally-hard"
region, or say "hybrid-hard phase", appears whenever the reconstruction bound
is not tight of the corresponding reconstruction on the tree problem. Although
it has been studied in numerous contexts, the existing literature with rigorous
reconstruction thresholds established are very limited, and it becomes
extremely challenging when the model under investigation has $4$ states (the
stochastic block model with $4$ communities). In this paper, inspired by the
newly proposed $q_1+q_2$ stochastic block model, we study a $4$ states
symmetric model with different in-block and out-block transition probabilities,
and rigorously give the conditions for the non-tightness of the reconstruction
threshold.
| stat |
Fisher Efficient Inference of Intractable Models | Maximum Likelihood Estimators (MLE) has many good properties. For example,
the asymptotic variance of MLE solution attains equality of the asymptotic
Cram{\'e}r-Rao lower bound (efficiency bound), which is the minimum possible
variance for an unbiased estimator. However, obtaining such MLE solution
requires calculating the likelihood function which may not be tractable due to
the normalization term of the density model. In this paper, we derive a
Discriminative Likelihood Estimator (DLE) from the Kullback-Leibler divergence
minimization criterion implemented via density ratio estimation and a Stein
operator. We study the problem of model inference using DLE. We prove its
consistency and show that the asymptotic variance of its solution can attain
the equality of the efficiency bound under mild regularity conditions. We also
propose a dual formulation of DLE which can be easily optimized. Numerical
studies validate our asymptotic theorems and we give an example where DLE
successfully estimates an intractable model constructed using a pre-trained
deep neural network.
| 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.