title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Modeling and Optimization with Gaussian Processes in Reduced Eigenbases
-- Extended Version | Parametric shape optimization aims at minimizing an objective function f(x)
where x are CAD parameters. This task is difficult when f is the output of an
expensive-to-evaluate numerical simulator and the number of CAD parameters is
large. Most often, the set of all considered CAD shapes resides in a manifold
of lower effective dimension in which it is preferable to build the surrogate
model and perform the optimization. In this work, we uncover the manifold
through a high-dimensional shape mapping and build a new coordinate system made
of eigenshapes. The surrogate model is learned in the space of eigenshapes: a
regularized likelihood maximization provides the most relevant dimensions for
the output. The final surrogate model is detailed (anisotropic) with respect to
the most sensitive eigenshapes and rough (isotropic) in the remaining
dimensions. Last, the optimization is carried out with a focus on the critical
dimensions, the remaining ones being coarsely optimized through a random
embedding and the manifold being accounted for through a replication strategy.
At low budgets, the methodology leads to a more accurate model and a faster
optimization than the classical approach of directly working with the CAD
parameters.
| stat |
Prediction Using a Bayesian Heteroscedastic Composite Gaussian Process | This research proposes a flexible Bayesian extension of the composite
Gaussian process (CGP) model of Ba and Joseph (2012) for predicting (stationary
or) non-stationary $y(\mathbf{x})$. The CGP generalizes the regression plus
stationary Gaussian process (GP) model by replacing the regression term with a
GP. The new model, $Y(\mathbf{x})$, can accommodate large-scale trends
estimated by a global GP, local trends estimated by an independent local GP,
and a third process to describe heteroscedastic data in which
$Var(Y(\mathbf{x}))$ can depend on the inputs. This paper proposes a prior
which ensures that the fitted global mean is smoother than the local
deviations, and extends the covariance structure of the CGP to allow for
differentially-weighted global and local components. A Markov chain Monte Carlo
algorithm is proposed to provide posterior estimates of the parameters,
including the values of the heteroscedastic variance at the training and test
data locations. The posterior distribution is used to make predictions and to
quantify the uncertainty of the predictions using prediction intervals. The
method is illustrated using both stationary and non-stationary $y(\mathbf{x})$.
| stat |
Learning Dynamics of Attention: Human Prior for Interpretable Machine
Reasoning | Without relevant human priors, neural networks may learn uninterpretable
features. We propose Dynamics of Attention for Focus Transition (DAFT) as a
human prior for machine reasoning. DAFT is a novel method that regularizes
attention-based reasoning by modelling it as a continuous dynamical system
using neural ordinary differential equations. As a proof of concept, we augment
a state-of-the-art visual reasoning model with DAFT. Our experiments reveal
that applying DAFT yields similar performance to the original model while using
fewer reasoning steps, showing that it implicitly learns to skip unnecessary
steps. We also propose a new metric, Total Length of Transition (TLT), which
represents the effective reasoning step size by quantifying how much a given
model's focus drifts while reasoning about a question. We show that adding DAFT
results in lower TLT, demonstrating that our method indeed obeys the human
prior towards shorter reasoning paths in addition to producing more
interpretable attention maps. Our code is available at
https://github.com/kakao/DAFT.
| stat |
Vector Autoregressive Models with Spatially Structured Coefficients for
Time Series on a Spatial Grid | We propose a parsimonious spatiotemporal model for time series data on a
spatial grid. Our model is capable of dealing with high-dimensional time series
data that may be collected at hundreds of locations and capturing the spatial
non-stationarity. In essence, our model is a vector autoregressive model that
utilizes the spatial structure to achieve parsimony of autoregressive matrices
at two levels. The first level ensures the sparsity of the autoregressive
matrices using a lagged-neighborhood scheme. The second level performs a
spatial clustering of the non-zero autoregressive coefficients such that nearby
locations share similar coefficients. This model is interpretable and can be
used to identify geographical subregions, within each of which, the time series
share similar dynamical behavior with homogeneous autoregressive coefficients.
The model parameters are obtained using the penalized maximum likelihood with
an adaptive fused Lasso penalty. The estimation procedure is easy to implement
and can be tailored to the need of a modeler. We illustrate the performance of
the proposed estimation algorithm in a simulation study. We apply our model to
a wind speed time series dataset generated from a climate model over Saudi
Arabia to illustrate its usefulness. Limitations and possible extensions of our
method are also discussed.
| stat |
Non-asymptotic Performances of Robust Markov Decision Processes | In this paper, we study the non-asymptotic performance of optimal policy on
robust value function with true transition dynamics. The optimal robust policy
is solved from a generative model or offline dataset without access to true
transition dynamics. In particular, we consider three different uncertainty
sets including the $L_1$, $\chi^2$ and KL balls in both $(s,a)$-rectangular and
$s$-rectangular assumptions. Our results show that when we assume
$(s,a)$-rectangular on uncertainty sets, the sample complexity is about
$\widetilde{O}\left(\frac{|\mathcal{S}|^2|\mathcal{A}|}{\varepsilon^2\rho^2(1-\gamma)^4}\right)$
in the generative model setting and
$\widetilde{O}\left(\frac{|\mathcal{S}|}{\nu_{\min}\varepsilon^2\rho^2(1-\gamma)^4}\right)$
in the offline dataset setting. While prior works on non-asymptotic
performances are restricted with the KL ball and $(s,a)$-rectangular
assumption, we also extend our results to a more general $s$-rectangular
assumption, which leads to a larger sample complexity than the
$(s,a)$-rectangular assumption.
| stat |
Analytic expressions for the output evolution of a deep neural network | We present a novel methodology based on a Taylor expansion of the network
output for obtaining analytical expressions for the expected value of the
network weights and output under stochastic training. Using these analytical
expressions the effects of the hyperparameters and the noise variance of the
optimization algorithm on the performance of the deep neural network are
studied. In the early phases of training with a small noise coefficient, the
output is equivalent to a linear model. In this case the network can generalize
better due to the noise preventing the output from fully converging on the
train data, however the noise does not result in any explicit regularization.
In the later training stages, when higher order approximations are required,
the impact of the noise becomes more significant, i.e. in a model which is
non-linear in the weights noise can regularize the output function resulting in
better generalization as witnessed by its influence on the weight Hessian, a
commonly used metric for generalization capabilities.
| stat |
Causal inference, social networks, and chain graphs | Traditionally, statistical and causal inference on human subjects rely on the
assumption that individuals are independently affected by treatments or
exposures. However, recently there has been increasing interest in settings,
such as social networks, where individuals may interact with one another such
that treatments may spill over from the treated individual to their social
contacts and outcomes may be contagious. Existing models proposed for causal
inference using observational data from networks of interacting individuals
have two major shortcomings. First, they often require a level of granularity
in the data that is practically infeasible to collect in most settings, and
second, the models are high-dimensional and often too big to fit to the
available data. In this paper we illustrate and justify a parsimonious
parameterization for network data with interference and contagion. Our
parameterization corresponds to a particular family of graphical models known
as chain graphs. We argue that, in some settings, chain graph models
approximate the marginal distribution of a snapshot of a longitudinal data
generating process on interacting units. We illustrate the use of chain graphs
for causal inference about collective decision making in social networks using
data from U.S. Supreme Court decisions between 1994 and 2004 and in
simulations.
| stat |
Mode Hunting Using Pettiest Components Analysis | Principal component analysis has been used to reduce dimensionality of
datasets for a long time. In this paper, we will demonstrate that in mode
detection the components of smallest variance, the pettiest components, are
more important. We prove that when the data follows a multivariate normal
distribution, by implementing "pettiest component analysis" when the data is
normally distributed, we obtain boxes of optimal size in the sense that their
size is minimal over all possible boxes with the same number of dimensions and
given probability. We illustrate our result with a simulation revealing that
pettiest component analysis works better than its competitors.
| stat |
Structure learning of undirected graphical models for count data | Biological processes underlying the basic functions of a cell involve complex
interactions between genes. From a technical point of view, these interactions
can be represented through a graph where genes and their connections are,
respectively, nodes and edges. The main objective of this paper is to develop a
statistical framework for modelling the interactions between genes when the
activity of genes is measured on a discrete scale. In detail, we define a new
algorithm for learning the structure of undirected graphs, PC-LPGM, proving its
theoretical consistence in the limit of infinite observations. The proposed
algorithm shows promising results when applied to simulated data as well as to
real data.
| stat |
Bayesian Causal Inference with Bipartite Record Linkage | In many scenarios, the observational data needed for causal inferences are
spread over two data files. In particular, we consider scenarios where one file
includes covariates and the treatment measured on one set of individuals, and a
second file includes responses measured on another, partially overlapping set
of individuals. In the absence of error free direct identifiers like social
security numbers, straightforward merging of separate files is not feasible, so
that records must be linked using error-prone variables such as names, birth
dates, and demographic characteristics. Typical practice in such situations
generally follows a two-stage procedure: first link the two files using a
probabilistic linkage technique, then make causal inferences with the linked
dataset. This does not propagate uncertainty due to imperfect linkages to the
causal inference, nor does it leverage relationships among the study variables
to improve the quality of the linkages. We propose a hierarchical model for
simultaneous Bayesian inference on probabilistic linkage and causal effects
that addresses these deficiencies. Using simulation studies and theoretical
arguments, we show the hierarchical model can improve the accuracy of estimated
treatment effects, as well as the record linkages, compared to the two-stage
modeling option. We illustrate the hierarchical model using a causal study of
the effects of debit card possession on household spending.
| stat |
A Pseudo-Marginal Metropolis-Hastings Algorithm for Estimating
Generalized Linear Models in the Presence of Missing Data | The missing data issue often complicates the task of estimating generalized
linear models (GLMs). We describe why the pseudo-marginal Metropolis-Hastings
algorithm, used in this setting, is an effective strategy for parameter
estimation. This approach requires fewer assumptions, it provides joint
inferences on the parameters in the likelihood, the covariate model, and the
parameters of the missingness-mechanism, and there is no logical inconsistency
of assuming that there are multiple posterior distributions. Moreover, this
approach is asymptotically exact, just like most other Markov chain Monte Carlo
techniques. We discuss computing strategies, conduct a simulation study
demonstrating how standard errors change as a function of percent missingness,
and we use our approach on a "real-world" data set to describe how a collection
of variables influences the car crash outcomes.
| stat |
Bayesian Neural Networks: An Introduction and Survey | Neural Networks (NNs) have provided state-of-the-art results for many
challenging machine learning tasks such as detection, regression and
classification across the domains of computer vision, speech recognition and
natural language processing. Despite their success, they are often implemented
in a frequentist scheme, meaning they are unable to reason about uncertainty in
their predictions. This article introduces Bayesian Neural Networks (BNNs) and
the seminal research regarding their implementation. Different approximate
inference methods are compared, and used to highlight where future research can
improve on current methods.
| stat |
Extension to mixed models of the Supervised Component-based Generalised
Linear Regression | We address the component-based regularisation of a multivariate Generalized
Linear Mixed Model (GLMM). A set of random responses Y is modelled by a GLMM,
using a set X of explanatory variables, a set T of additional covariates, and
random effects used to introduce the dependence between statistical units.
Variables in X are assumed many and redundant, so that regression demands
regularisation. By contrast, variables in T are assumed few and selected so as
to require no regularisation. Regularisation is performed building an
appropriate number of orthogonal components that both contribute to model Y and
capture relevant structural information in X. To estimate the model, we propose
to maximise a criterion specific to the Supervised Component-based Generalised
Linear Regression (SCGLR) within an adaptation of Schall's algorithm. This
extension of SCGLR is tested on both simulated and real data, and compared to
Ridge- and Lasso-based regularisations.
| stat |
Uncertainty Quantification for Inferring Hawkes Networks | Multivariate Hawkes processes are commonly used to model streaming networked
event data in a wide variety of applications. However, it remains a challenge
to extract reliable inference from complex datasets with uncertainty
quantification. Aiming towards this, we develop a statistical inference
framework to learn causal relationships between nodes from networked data,
where the underlying directed graph implies Granger causality. We provide
uncertainty quantification for the maximum likelihood estimate of the network
multivariate Hawkes process by providing a non-asymptotic confidence set. The
main technique is based on the concentration inequalities of continuous-time
martingales. We compare our method to the previously-derived asymptotic Hawkes
process confidence interval, and demonstrate the strengths of our method in an
application to neuronal connectivity reconstruction.
| stat |
Evaluating probabilistic classifiers: Reliability diagrams and score
decompositions revisited | A probability forecast or probabilistic classifier is reliable or calibrated
if the predicted probabilities are matched by ex post observed frequencies, as
examined visually in reliability diagrams. The classical binning and counting
approach to plotting reliability diagrams has been hampered by a lack of
stability under unavoidable, ad hoc implementation decisions. Here we introduce
the CORP approach, which generates provably statistically Consistent, Optimally
binned, and Reproducible reliability diagrams in an automated way. CORP is
based on non-parametric isotonic regression and implemented via the
Pool-adjacent-violators (PAV) algorithm - essentially, the CORP reliability
diagram shows the graph of the PAV- (re)calibrated forecast probabilities. The
CORP approach allows for uncertainty quantification via either resampling
techniques or asymptotic theory, furnishes a new numerical measure of
miscalibration, and provides a CORP based Brier score decomposition that
generalizes to any proper scoring rule. We anticipate that judicious uses of
the PAV algorithm yield improved tools for diagnostics and inference for a very
wide range of statistical and machine learning methods.
| stat |
Amortized learning of neural causal representations | Causal models can compactly and efficiently encode the data-generating
process under all interventions and hence may generalize better under changes
in distribution. These models are often represented as Bayesian networks and
learning them scales poorly with the number of variables. Moreover, these
approaches cannot leverage previously learned knowledge to help with learning
new causal models. In order to tackle these challenges, we represent a novel
algorithm called \textit{causal relational networks} (CRN) for learning causal
models using neural networks. The CRN represent causal models using continuous
representations and hence could scale much better with the number of variables.
These models also take in previously learned information to facilitate learning
of new causal models. Finally, we propose a decoding-based metric to evaluate
causal models with continuous representations. We test our method on synthetic
data achieving high accuracy and quick adaptation to previously unseen causal
models.
| stat |
Accurate inference in negative binomial regression | Negative binomial regression is commonly employed to analyze overdispersed
count data. With small to moderate sample sizes, the maximum likelihood
estimator of the dispersion parameter may be subject to a significant bias,
that in turn affects inference on mean parameters. This paper proposes
inference for negative binomial regression based on adjustments of the score
function aimed at mean and median bias reduction. The resulting estimating
equations are similar to those available for improved inference in generalized
linear models and, in particular, can be solved using a suitable extension of
iterative weighted least squares. Simulation studies show a remarkable
performance of the new methods, which are also found to solve in many cases
numerical problems of maximum likelihood estimates. The methods are illustrated
and evaluated using two case studies: an Ames salmonella assay data set and
data on epileptic seizures. Inference based on adjusted scores turns out to be
generally preferable to explicit bias correction.
| stat |
Mining Time-Stamped Electronic Health Records Using Referenced Sequences | Electronic Health Records (EHRs) are typically stored as time-stamped
encounter records. Observing temporal relationship between medical records is
an integral part of interpreting the information. Hence, statistical analysis
of EHRs requires that clinically informed time-interdependent analysis
variables (TIAV) be created. Often, formulation and creation of these variables
are iterative and requiring custom codes. We describe a technique of using
sequences of time-referenced entities as the building blocks for TIAVs. These
sequences represent different aspects of patient's medical history in a
contiguous fashion. To illustrate the principles and applications of the
method, we provide examples using Veterans Health Administration's research
databases. In the first example, sequences representing medication exposure
were used to assess patient selection criteria for a treatment comparative
effectiveness study. In the second example, sequences of Charlson Comorbidity
conditions and clinical settings of inpatient or outpatient were used to create
variables with which data anomalies and trends were revealed. The third example
demonstrated the creation of an analysis variable derived from the temporal
dependency of medication exposure and comorbidity. Complex time-interdependent
analysis variables can be created from the sequences with simple, reusable
codes, hence enable unscripted or automation of TIAV creation.
| stat |
Variational Inference of Joint Models using Multivariate Gaussian
Convolution Processes | We present a non-parametric prognostic framework for individualized event
prediction based on joint modeling of both longitudinal and time-to-event data.
Our approach exploits a multivariate Gaussian convolution process (MGCP) to
model the evolution of longitudinal signals and a Cox model to map
time-to-event data with longitudinal data modeled through the MGCP. Taking
advantage of the unique structure imposed by convolved processes, we provide a
variational inference framework to simultaneously estimate parameters in the
joint MGCP-Cox model. This significantly reduces computational complexity and
safeguards against model overfitting. Experiments on synthetic and real world
data show that the proposed framework outperforms state-of-the art approaches
built on two-stage inference and strong parametric assumptions.
| stat |
Sample Size Calculations for SMARTs | Sequential Multiple Assignment Randomized Trials (SMARTs) are considered the
gold standard for estimation and evaluation of treatment regimes. SMARTs are
typically sized to ensure sufficient power for a simple comparison, e.g., the
comparison of two fixed treatment sequences. Estimation of an optimal treatment
regime is conducted as part of a secondary and hypothesis-generating analysis
with formal evaluation of the estimated optimal regime deferred to a follow-up
trial. However, running a follow-up trial to evaluate an estimated optimal
treatment regime is costly and time-consuming; furthermore, the estimated
optimal regime that is to be evaluated in such a follow-up trial may be far
from optimal if the original trial was underpowered for estimation of an
optimal regime. We derive sample size procedures for a SMART that ensure: (i)
sufficient power for comparing the optimal treatment regime with standard of
care; and (ii) the estimated optimal regime is within a given tolerance of the
true optimal regime with high-probability. We establish asymptotic validity of
the proposed procedures and demonstrate their finite sample performance in a
series of simulation experiments.
| stat |
Inference of Polygenic Factors Associated with Breast Cancer Gene
Interaction Networks from Discrete Data Utilizing Poisson Multivariate Mutual
Information | In this work we introduce a new methodology to infer from gene expression
data the complex interactions associated with polygenetic diseases that remain
a major frontier in understanding factors in human health. In many cases
disease may be related to the covariance of several genes, rather than simply
the variance of a single gene, making network inference crucial to the
development of potential treatments. Specifically we investigate the network of
factors and associations involved in developing breast cancer from gene
expression data. Our approach is information theoretic, but a major obstacle
has been the discrete nature of such data that is well described as a
multi-variate Poisson process. In fact despite that mutual information is
generally a well regarded approach for developing networks of association in
data science of complex systems across many disciplines, until now a good
method to accurately and efficiently compute entropies from such processes as
been lacking. Nonparameteric methods such as the popular k-nearest neighbors
(KNN) methods are slow converging and thus require unrealistic amounts of data.
We will use the causation entropy (CSE) principle, together with the associated
greedy search algorithm optimal CSE (oCSE) as a network inference method to
deduce the actual structure, with our multi-variate Poisson estimator developed
here as the core computational engine. We show that the Poisson version of oCSE
outperforms both the Kraskov-St\"ogbauer-Grassberger (KSG) oCSE method (which
is a KNN method for estimating the entropy) and the Gaussian oCSE method on
synthetic data. We present the results for a breast cancer gene expression data
set.
| stat |
fIRTree: An Item Response Theory modeling of fuzzy rating data | In this contribution we describe a novel procedure to represent fuzziness in
rating scales in terms of fuzzy numbers. Following the rationale of fuzzy
conversion scale, we adopted a two-step procedure based on a psychometric model
(i.e., Item Response Theory-based tree) to represent the process of answer
survey questions. This provides a coherent context where fuzzy numbers, and the
related fuzziness, can be interpreted in terms of decision uncertainty that
usually affects the rater's response process. We reported results from a
simulation study and an empirical application to highlight the characteristics
and properties of the proposed approach.
| stat |
Bayesian design for minimising uncertainty in spatial processes | Model-based geostatistical design involves the selection of locations to
collect data to minimise an expected loss function over a set of all possible
locations. The loss function is specified to reflect the aim of data
collection, which, for geostatistical studies, would typically be to minimise
the uncertainty in a spatial process. In this paper, we propose a new approach
to design such studies via a loss function derived through considering the
entropy of model predictions, and we show that this simultaneously addresses
the goal of precise parameter estimation. One drawback of this loss function is
that is it computationally expensive to evaluate, so we provide an efficient
approximation such that it can be used within realistically sized
geostatistical studies. To demonstrate our approach, we apply the proposed
approach to design the collection of spatially dependent multiple responses,
and compare this with either designing for estimation or prediction only. The
results show that our designs remain highly efficient in achieving each
experimental objective individually, and provide an ideal compromise between
the two objectives. Accordingly, we advocate that our design approach should be
used more generally in model-based geostatistical studies.
| stat |
Influence of Vaporfly shoe on sub-2 hour marathon and other top running
performances | In 2019, Eliud Chipgoke ran a sub-two hour marathon in an unofficial race
wearing last-generation shoes. The legitimacy of this feat was naturally
questioned due to unusual racing conditions and suspicions of technological
doping. In this work, we assess the likelihood of a sub-two hour marathon in an
official race, and the potential influence of Vaporfly shoes, by studying the
evolution of running top performances from 2001 to 2019 for distances ranging
from 10k to marathon. The analysis is performed using extreme value theory, a
field of statistics dealing with analysis of records. We find a significant
evidence of technological doping with a 12% increase of the probability that a
new world record for marathon-man discipline is set in 2021. However, results
suggest that achieving a sub-two hour marathon in an official event in 2021 is
still very unlikely, and exceeds 10% probability only by 2025.
| stat |
Physics-aware, probabilistic model order reduction with guaranteed
stability | Given (small amounts of) time-series' data from a high-dimensional,
fine-grained, multiscale dynamical system, we propose a generative framework
for learning an effective, lower-dimensional, coarse-grained dynamical model
that is predictive of the fine-grained system's long-term evolution but also of
its behavior under different initial conditions. We target fine-grained models
as they arise in physical applications (e.g. molecular dynamics, agent-based
models), the dynamics of which are strongly non-stationary but their transition
to equilibrium is governed by unknown slow processes which are largely
inaccessible by brute-force simulations. Approaches based on domain knowledge
heavily rely on physical insight in identifying temporally slow features and
fail to enforce the long-term stability of the learned dynamics. On the other
hand, purely statistical frameworks lack interpretability and rely on large
amounts of expensive simulation data (long and multiple trajectories) as they
cannot infuse domain knowledge. The generative framework proposed achieves the
aforementioned desiderata by employing a flexible prior on the complex plane
for the latent, slow processes, and an intermediate layer of physics-motivated
latent variables that reduces reliance on data and imbues inductive bias. In
contrast to existing schemes, it does not require the a priori definition of
projection operators from the fine-grained description and addresses
simultaneously the tasks of dimensionality reduction and model estimation. We
demonstrate its efficacy and accuracy in multiscale physical systems of
particle dynamics where probabilistic, long-term predictions of phenomena not
contained in the training data are produced.
| stat |
Mastering the body and tail shape of a distribution | The normal distribution and its perturbation has left an immense mark on the
statistical literature. Hence, several generalized forms were developed to
model different skewness, kurtosis, and body shapes. However, it is not easy to
distinguish between changes in the relative body and tail shapes when using
these generalizations. What we propose is a neat integration approach
generalization which enables the visualization and control of the body and the
tail shape separately. This provides a flexible modeling opportunity with an
emphasis on parameter inference and interpretation. Two related models, the
two-piece body-tail generalized normal and the two-piece tail adjusted normal
are swiftly introduced to demonstrate this inferential potential. The
methodology is then demonstrated on heavy and light-tailed data.
| stat |
Variational Multiscale Nonparametric Regression: Algorithms and
Implementation | Many modern statistically efficient methods come with tremendous
computational challenges, often leading to large-scale optimisation problems.
In this work, we examine such computational issues for recently developed
estimation methods in nonparametric regression with a specific view on image
denoising. We consider in particular certain variational multiscale estimators
which are statistically optimal in minimax sense, yet computationally
intensive. Such an estimator is computed as the minimiser of a smoothness
functional (e.g., TV norm) over the class of all estimators such that none of
its coefficients with respect to a given multiscale dictionary is statistically
significant. The so obtained multiscale Nemirowski-Dantzig estimator (MIND) can
incorporate any convex smoothness functional and combine it with a proper
dictionary including wavelets, curvelets and shearlets. The computation of MIND
in general requires to solve a high-dimensional constrained convex optimisation
problem with a specific structure of the constraints induced by the statistical
multiscale testing criterion. To solve this explicitly, we discuss three
different algorithmic approaches: the Chambolle-Pock, ADMM and semismooth
Newton algorithms. Algorithmic details and an explicit implementation is
presented and the solutions are then compared numerically in a simulation study
and on various test images. We thereby recommend the Chambolle-Pock algorithm
in most cases for its fast convergence. We stress that our analysis can also be
transferred to signal recovery and other denoising problems to recover more
general objects whenever it is possible to borrow statistical strength from
data patches of similar object structure.
| stat |
Sparse Canonical Correlation Analysis via Concave Minimization | A new approach to the sparse Canonical Correlation Analysis (sCCA)is proposed
with the aim of discovering interpretable associations in very high-dimensional
multi-view, i.e.observations of multiple sets of variables on the same
subjects, problems. Inspired by the sparse PCA approach of Journee et al.
(2010), we also show that the sparse CCA formulation, while non-convex, is
equivalent to a maximization program of a convex objective over a compact set
for which we propose a first-order gradient method. This result helps us reduce
the search space drastically to the boundaries of the set. Consequently, we
propose a two-step algorithm, where we first infer the sparsity pattern of the
canonical directions using our fast algorithm, then we shrink each view, i.e.
observations of a set of covariates, to contain observations on the sets of
covariates selected in the previous step, and compute their canonical
directions via any CCA algorithm. We also introduceDirected Sparse CCA, which
is able to find associations which are aligned with a specified experiment
design, andMulti-View sCCA which is used to discover associations between
multiple sets of covariates. Our simulations establish the superior convergence
properties and computational efficiency of our algorithm as well as accuracy in
terms of the canonical correlation and its ability to recover the supports of
the canonical directions. We study the associations between metabolomics,
trasncriptomics and microbiomics in a multi-omic study usingMuLe, which is an
R-package that implements our approach, in order to form hypotheses on
mechanisms of adaptations of Drosophila Melanogaster to high doses of
environmental toxicants, specifically Atrazine, which is a commonly used
chemical fertilizer.
| stat |
Learning Hierarchical Priors in VAEs | We propose to learn a hierarchical prior in the context of variational
autoencoders to avoid the over-regularisation resulting from a standard normal
prior distribution. To incentivise an informative latent representation of the
data, we formulate the learning problem as a constrained optimisation problem
by extending the Taming VAEs framework to two-level hierarchical models. We
introduce a graph-based interpolation method, which shows that the topology of
the learned latent representation corresponds to the topology of the data
manifold---and present several examples, where desired properties of latent
representation such as smoothness and simple explanatory factors are learned by
the prior.
| stat |
Generative Models and Model Criticism via Optimized Maximum Mean
Discrepancy | We propose a method to optimize the representation and distinguishability of
samples from two probability distributions, by maximizing the estimated power
of a statistical test based on the maximum mean discrepancy (MMD). This
optimized MMD is applied to the setting of unsupervised learning by generative
adversarial networks (GAN), in which a model attempts to generate realistic
samples, and a discriminator attempts to tell these apart from data samples. In
this context, the MMD may be used in two roles: first, as a discriminator,
either directly on the samples, or on features of the samples. Second, the MMD
can be used to evaluate the performance of a generative model, by testing the
model's samples against a reference data set. In the latter role, the optimized
MMD is particularly helpful, as it gives an interpretable indication of how the
model and data distributions differ, even in cases where individual model
samples are not easily distinguished either by eye or by classifier.
| stat |
First Exit Time Analysis of Stochastic Gradient Descent Under
Heavy-Tailed Gradient Noise | Stochastic gradient descent (SGD) has been widely used in machine learning
due to its computational efficiency and favorable generalization properties.
Recently, it has been empirically demonstrated that the gradient noise in
several deep learning settings admits a non-Gaussian, heavy-tailed behavior.
This suggests that the gradient noise can be modeled by using $\alpha$-stable
distributions, a family of heavy-tailed distributions that appear in the
generalized central limit theorem. In this context, SGD can be viewed as a
discretization of a stochastic differential equation (SDE) driven by a L\'{e}vy
motion, and the metastability results for this SDE can then be used for
illuminating the behavior of SGD, especially in terms of `preferring wide
minima'. While this approach brings a new perspective for analyzing SGD, it is
limited in the sense that, due to the time discretization, SGD might admit a
significantly different behavior than its continuous-time limit. Intuitively,
the behaviors of these two systems are expected to be similar to each other
only when the discretization step is sufficiently small; however, to the best
of our knowledge, there is no theoretical understanding on how small the
step-size should be chosen in order to guarantee that the discretized system
inherits the properties of the continuous-time system. In this study, we
provide formal theoretical analysis where we derive explicit conditions for the
step-size such that the metastability behavior of the discrete-time system is
similar to its continuous-time limit. We show that the behaviors of the two
systems are indeed similar for small step-sizes and we identify how the error
depends on the algorithm and problem parameters. We illustrate our results with
simulations on a synthetic model and neural networks.
| stat |
Bayesian modelling for spatially misaligned health areal data: a
multiple membership approach | Diabetes prevalence is on the rise in the UK, and for public health strategy,
estimation of relative disease risk and subsequent mapping is important. We
consider an application to London data on diabetes prevalence and mortality. In
order to improve the estimation of relative risks we analyse jointly prevalence
and mortality data to ensure borrowing strength over the two outcomes. The
available data involves two spatial frameworks, areas (middle level super
output areas, MSOAs), and general practices (GPs) recruiting patients from
several areas. This raises a spatial misalignment issue that we deal with by
employing the multiple membership principle. Specifically we translate area
spatial effects to explain GP practice prevalence according to proportions of
GP populations resident in different areas. A sparse implementation in Stan of
both the MCAR and GMCAR allows the comparison of these bivariate priors as well
as exploring the different implications for the mapping patterns for both
outcomes. The necessary causal precedence of diabetes prevalence over mortality
allows a specific conditionality assumption in the GMCAR, not always present in
the context of disease mapping.
| stat |
Selfless Sequential Learning | Sequential learning, also called lifelong learning, studies the problem of
learning tasks in a sequence with access restricted to only the data of the
current task. In this paper we look at a scenario with fixed model capacity,
and postulate that the learning process should not be selfish, i.e. it should
account for future tasks to be added and thus leave enough capacity for them.
To achieve Selfless Sequential Learning we study different regularization
strategies and activation functions. We find that imposing sparsity at the
level of the representation (i.e.~neuron activations) is more beneficial for
sequential learning than encouraging parameter sparsity. In particular, we
propose a novel regularizer, that encourages representation sparsity by means
of neural inhibition. It results in few active neurons which in turn leaves
more free neurons to be utilized by upcoming tasks. As neural inhibition over
an entire layer can be too drastic, especially for complex tasks requiring
strong representations, our regularizer only inhibits other neurons in a local
neighbourhood, inspired by lateral inhibition processes in the brain. We
combine our novel regularizer, with state-of-the-art lifelong learning methods
that penalize changes to important previously learned parts of the network. We
show that our new regularizer leads to increased sparsity which translates in
consistent performance improvement %over alternative regularizers we studied on
diverse datasets.
| stat |
Bayesian inference of heterogeneous epidemic models: Application to
COVID-19 spread accounting for long-term care facilities | We propose a high dimensional Bayesian inference framework for learning
heterogeneous dynamics of a COVID-19 model, with a specific application to the
dynamics and severity of COVID-19 inside and outside long-term care (LTC)
facilities. We develop a heterogeneous compartmental model that accounts for
the heterogeneity of the time-varying spread and severity of COVID-19 inside
and outside LTC facilities, which is characterized by time-dependent stochastic
processes and time-independent parameters in $\sim$1500 dimensions after
discretization. To infer these parameters, we use reported data on the number
of confirmed, hospitalized, and deceased cases with suitable post-processing in
both a deterministic inversion approach with appropriate regularization as a
first step, followed by Bayesian inversion with proper prior distributions. To
address the curse of dimensionality and the ill-posedness of the
high-dimensional inference problem, we propose use of a dimension-independent
projected Stein variational gradient descent method, and demonstrate the
intrinsic low-dimensionality of the inverse problem. We present inference
results with quantified uncertainties for both New Jersey and Texas, which
experienced different epidemic phases and patterns. Moreover, we also present
forecasting and validation results based on the empirical posterior samples of
our inference for the future trajectory of COVID-19.
| stat |
Robust estimation of tree structured models | Consider the problem of learning undirected graphical models on trees from
corrupted data. Recently Katiyar et al. showed that it is possible to recover
trees from noisy binary data up to a small equivalence class of possible trees.
Their other paper on the Gaussian case follows a similar pattern. By framing
this as a special phylogenetic recovery problem we largely generalize these two
settings. Using the framework of linear latent tree models we discuss tree
identifiability for binary data under a continuous corruption model. For the
Ising and the Gaussian tree model we also provide a characterisation of when
the Chow-Liu algorithm consistently learns the underlying tree from the noisy
data.
| stat |
A Riemannian Newton Trust-Region Method for Fitting Gaussian Mixture
Models | Gaussian Mixture Models are a powerful tool in Data Science and Statistics
that are mainly used for clustering and density approximation. The task of
estimating the model parameters is in practice often solved by the Expectation
Maximization (EM) algorithm which has its benefits in its simplicity and low
per-iteration costs. However, the EM converges slowly if there is a large share
of hidden information or overlapping clusters. Recent advances in Manifold
Optimization for Gaussian Mixture Models have gained increasing interest. We
introduce a formula for the Riemannian Hessian for Gaussian Mixture Models. On
top, we propose a new Riemannian Newton Trust-Region method which outperforms
current approaches both in terms of runtime and number of iterations.
| stat |
Principled analytic classifier for positive-unlabeled learning via
weighted integral probability metric | We consider the problem of learning a binary classifier from only positive
and unlabeled observations (called PU learning). Recent studies in PU learning
have shown superior performance theoretically and empirically. However, most
existing algorithms may not be suitable for large-scale datasets because they
face repeated computations of a large Gram matrix or require massive
hyperparameter optimization. In this paper, we propose a computationally
efficient and theoretically grounded PU learning algorithm. The proposed PU
learning algorithm produces a closed-form classifier when the hypothesis space
is a closed ball in reproducing kernel Hilbert space. In addition, we establish
upper bounds of the estimation error and the excess risk. The obtained
estimation error bound is sharper than existing results and the derived excess
risk bound has an explicit form, which vanishes as sample sizes increase.
Finally, we conduct extensive numerical experiments using both synthetic and
real datasets, demonstrating improved accuracy, scalability, and robustness of
the proposed algorithm.
| stat |
Fast Best Subset Selection: Coordinate Descent and Local Combinatorial
Optimization Algorithms | The $L_0$-regularized least squares problem (a.k.a. best subsets) is central
to sparse statistical learning and has attracted significant attention across
the wider statistics, machine learning, and optimization communities. Recent
work has shown that modern mixed integer optimization (MIO) solvers can be used
to address small to moderate instances of this problem. In spite of the
usefulness of $L_0$-based estimators and generic MIO solvers, there is a steep
computational price to pay when compared to popular sparse learning algorithms
(e.g., based on $L_1$ regularization). In this paper, we aim to push the
frontiers of computation for a family of $L_0$-regularized problems with
additional convex penalties. We propose a new hierarchy of necessary optimality
conditions for these problems. We develop fast algorithms, based on coordinate
descent and local combinatorial optimization, that are guaranteed to converge
to solutions satisfying these optimality conditions. From a statistical
viewpoint, an interesting story emerges. When the signal strength is high, our
combinatorial optimization algorithms have an edge in challenging statistical
settings. When the signal is lower, pure $L_0$ benefits from additional convex
regularization. We empirically demonstrate that our family of $L_0$-based
estimators can outperform the state-of-the-art sparse learning algorithms in
terms of a combination of prediction, estimation, and variable selection
metrics under various regimes (e.g., different signal strengths, feature
correlations, number of samples and features). Our new open-source sparse
learning toolkit L0Learn (available on CRAN and Github) reaches up to a
three-fold speedup (with $p$ up to $10^6$) when compared to competing toolkits
such as glmnet and ncvreg.
| stat |
Bayesian Joint Modelling of Recurrence and Survival: a Conditional
Approach | Recurrent event processes describe the stochastic repetition of an event over
time. Recurrent event times are often censored with dependence between the
censoring time and recurrence process. For instance, recurrent disease events
are censored by a terminal event such as death, while frailty might affect both
disease recurrence and survival. As such, it is important to model the
recurrent event process and the event time process jointly to better capture
the dependency between them and improve interpretability of the results. We
propose a model in which the number of gap times, i.e. the time between two
consecutive recurrent events, before the terminal event occurs is a random
variable of interest. Then, conditionally on the number of recurrent events, we
specify a joint distribution for the gap times and the survival time.
Dependence between the the recurrence and survival processes is introduced by
specifying a joint distribution on their respective frailty terms. Moreover, an
autoregressive model of order one is assumed to model the evolution of gap
times over time. A non-parametric random effects distribution for the frailty
terms accommodates population heterogeneity and allows for data-driven
clustering of the subjects. Posterior inference is performed through a a Gibbs
sampler strategy involving a reversible jump step and slice sampling. We
illustrate our model on atrial fibrillation data and compare the performance of
our model with existing approaches.
| stat |
Robust Particle Density Tempering for State Space Models | Density tempering (also called density annealing) for state space models is a
sequential Monte Carlo (SMC) approach to Bayesian inference for general state
models, that provides an alternative to MCMC. It moves a collection of
parameters and latent states (which we call particles) through a number of
stages, with each stage having its own target density. Initially, the particles
are generated from a distribution that is easy to sample from, e.g. the prior;
the target density at the final stage is the posterior density of interest.
Tempering is usually carried out either in batch mode, involving all of the
data at each stage, or in sequential mode, where the tempering involves adding
observations at each stage; we call this data tempering. Our article proposes
two innovations for particle based density tempering. First, data tempering is
made more robust to outliers and structural changes by adding batch tempering
at each stage. Second, we propose generating the parameters and states at each
stage using two Gibbs type Markov moves, where the parameters are generated
conditional on the states and conversely. We explain how this allows the
tempering to scale up in terms of the number parameters and states it can
handle. Most of the current literature uses a pseudo-marginal Markov move step
with the states integrated out and the parameters generated by a random walk
proposal; this strategy is inefficient when the states or parameters are high
dimensional. The article demonstrates the performance of the proposed methods
using univariate stochastic volatility models with outliers and structural
breaks and high dimensional factor stochastic volatility models having both a
large number of parameters and a large number of latent state.
| stat |
Intelligent sampling for multiple change-points in exceedingly long time
series with rate guarantees | Change point estimation in its offline version is traditionally performed by
optimizing over the data set of interest, by considering each data point as the
true location parameter and computing a data fit criterion. Subsequently, the
data point that minimizes the criterion is declared as the change point
estimate. For estimating multiple change points, the procedures are analogous
in spirit, but significantly more involved in execution. Since change-points
are local discontinuities, only data points close to the actual change point
provide useful information for estimation, while data points far away are
superfluous, to the point where using only a few points close to the true
parameter is just as precise as using the full data set. Leveraging this
"locality principle", we introduce a two-stage procedure for the problem at
hand, which in the 1st stage uses a sparse subsample to obtain pilot estimates
of the underlying change points, and in the 2nd stage refines these estimates
by sampling densely in appropriately defined neighborhoods around them. We
establish that this method achieves the same rate of convergence and even
virtually the same asymptotic distribution as the analysis of the full data,
while reducing computational complexity to O(N^0.5) time (N being the length of
data set), as opposed to at least O(N) time for all current procedures, making
it promising for the analysis on exceedingly long data sets with adequately
spaced out change points. The main results are established under a signal plus
noise model with independent and identically distributed error terms, but
extensions to dependent data settings, as well as multiple stage (>2)
procedures are also provided. The performance of our procedure -- which is
coined "intelligent sampling" -- is illustrated on both synthetic and real
Internet data streams.
| stat |
Testing for Balance in Social Networks | Friendship and antipathy exist in concert with one another in real social
networks. Despite the role they play in social interactions, antagonistic ties
are poorly understood and infrequently measured. One important theory of
negative ties that has received relatively little empirical evaluation is
balance theory, the codification of the adage `the enemy of my enemy is my
friend' and similar sayings. Unbalanced triangles are those with an odd number
of negative ties, and the theory posits that such triangles are rare. To test
for balance, previous works have utilized a permutation test on the edge signs.
The flaw in this method, however, is that it assumes that negative and positive
edges are interchangeable. In reality, they could not be more different. Here,
we propose a novel test of balance that accounts for this discrepancy and show
that our test is more accurate at detecting balance. Along the way, we prove
asymptotic normality of the test statistic under our null model, which is of
independent interest. Our case study is a novel dataset of signed networks we
collected from 32 isolated, rural villages in Honduras. Contrary to previous
results, we find that there is only marginal evidence for balance in social tie
formation in this setting.
| stat |
Bayesian analysis of mixture autoregressive models covering the complete
parameter space | Mixture autoregressive (MAR) models provide a flexible way to model time
series with predictive distributions which depend on the recent history of the
process and are able to accommodate asymmetry and multimodality. Bayesian
inference for such models offers the additional advantage of incorporating the
uncertainty in the estimated models into the predictions. We introduce a new
way of sampling from the posterior distribution of the parameters of MAR models
which allows for covering the complete parameter space of the models, unlike
previous approaches. We also propose a relabelling algorithm to deal a
posteriori with label switching. We apply our new method to simulated and real
datasets, discuss the accuracy and performance of our new method, as well as
its advantages over previous studies. The idea of density forecasting using
MCMC output is also introduced.
| stat |
Variable Selection with Random Survival Forest and Bayesian Additive
Regression Tree for Survival Data | In this paper we utilize a survival analysis methodology incorporating
Bayesian additive regression trees to account for nonlinear and additive
covariate effects. We compare the performance of Bayesian additive regression
trees, Cox proportional hazards and random survival forests models for censored
survival data, using simulation studies and survival analysis for breast cancer
with U.S. SEER database for the year 2005. In simulation studies, we compare
the three models across varying sample sizes and censoring rates on the basis
of bias and prediction accuracy. In survival analysis for breast cancer, we
retrospectively analyze a subset of 1500 patients having invasive ductal
carcinoma that is a common form of breast cancer mostly affecting older woman.
Predictive potential of the three models are then compared using some widely
used performance assessment measures in survival literature.
| stat |
PAC-Bayesian High Dimensional Bipartite Ranking | This paper is devoted to the bipartite ranking problem, a classical
statistical learning task, in a high dimensional setting. We propose a scoring
and ranking strategy based on the PAC-Bayesian approach. We consider nonlinear
additive scoring functions, and we derive non-asymptotic risk bounds under a
sparsity assumption. In particular, oracle inequalities in probability holding
under a margin condition assess the performance of our procedure, and prove its
minimax optimality. An MCMC-flavored algorithm is proposed to implement our
method, along with its behavior on synthetic and real-life datasets.
| stat |
Deducing neighborhoods of classes from a fitted model | In todays world the request for very complex models for huge data sets is
rising steadily. The problem with these models is that by raising the
complexity of the models, it gets much harder to interpret them. The growing
field of \emph{interpretable machine learning} tries to make up for the lack of
interpretability in these complex (or even blackbox-)models by using specific
techniques that can help to understand those models better. In this article a
new kind of interpretable machine learning method is presented, which can help
to understand the partitioning of the feature space into predicted classes in a
classification model using quantile shifts. To illustrate in which situations
this quantile shift method (QSM) could become beneficial, it is applied to a
theoretical medical example and a real data example. Basically, real data
points (or specific points of interest) are used and the changes of the
prediction after slightly raising or decreasing specific features are observed.
By comparing the predictions before and after the manipulations, under certain
conditions the observed changes in the predictions can be interpreted as
neighborhoods of the classes with regard to the manipulated features.
Chordgraphs are used to visualize the observed changes.
| stat |
Efficiency of Observed Information Adaptive Designs | In this work the primary objective is to maximize the precision of the
maximum likelihood estimate in a linear regression model through the efficient
design of the experiment. One common measure of precision is the unconditional
mean square error. Unconditional mean square error has been a primary motivator
for optimal designs; commonly, defined as the design that maximizes a concave
function of the expected Fisher information. The inverse of expected Fisher
information is asymptotically equal to the mean square error of the maximum
likelihood estimate. There is a substantial amount of existing literature that
argues the mean square error conditioned on an appropriate ancillary statistic
better represents the precision of the maximum likelihood estimate. Despite
evidence in favor of conditioning, limited effort has been made to find designs
that are optimal with respect to conditional mean square error. The inverse of
observed Fisher information is a higher order approximation of the conditional
mean square error than the inverse of expected Fisher information [Efron and
Hinkley (1978)]. In light of this, a more relevant objective is to find designs
that optimize observed Fisher information. Unlike expected Fisher information,
observed Fisher information depends on the observed data and cannot be used to
design an experiment completely in advance of data collection. In a sequential
experiment the observed Fisher information from past observations is available
to inform the design of the next observation. In this work an adaptive design
that incorporates observed Fisher information is proposed for linear regression
models. It is shown that the proposed design is more efficient, at the limit,
than any fixed design, including the optimal design, with respect to
conditional mean square error.
| stat |
On missing label patterns in semi-supervised learning | We investigate model based classification with partially labelled training
data. In many biostatistical applications, labels are manually assigned by
experts, who may leave some observations unlabelled due to class uncertainty.
We analyse semi-supervised learning as a missing data problem and identify
situations where the missing label pattern is non-ignorable for the purposes of
maximum likelihood estimation. In particular, we find that a relationship
between classification difficulty and the missing label pattern implies a
non-ignorable missingness mechanism. We examine a number of real datasets and
conclude the pattern of missing labels is related to the difficulty of
classification. We propose a joint modelling strategy involving the observed
data and the missing label mechanism to account for the systematic missing
labels. Full likelihood inference including the missing label mechanism can
improve the efficiency of parameter estimation, and increase classification
accuracy.
| stat |
Variance propagation for density surface models | Spatially-explicit estimates of population density, together with appropriate
estimates of uncertainty, are required in many management contexts. Density
Surface Models (DSMs) are a two-stage approach for estimating spatially-varying
density from distance-sampling data. First, detection probabilities -- perhaps
depending on covariates -- are estimated based on details of individual
encounters; next, local densities are estimated using a GAM, by fitting local
encounter rates to location and/or spatially-varying covariates while allowing
for the estimated detectabilities. One criticism of DSMs has been that
uncertainty from the two stages is not usually propagated correctly into the
final variance estimates. We show how to reformulate a DSM so that the
uncertainty in detection probability from the distance sampling stage
(regardless of its complexity) is captured as an extra random effect in the GAM
stage. In effect, we refit an approximation to the detection function model at
the same time as fitting the spatial model. This allows straightforward
computation of the overall variance via exactly the same software already
needed to fit the GAM. A further extension allows for spatial variation in
group size, which can be an important covariate for detectability as well as
directly affecting abundance. We illustrate these models using point transect
survey data of Island Scrub-Jays on Santa Cruz Island, CA and harbour porpoise
from the SCANS-II line transect survey of European waters.
| stat |
Maximum approximate likelihood estimation of general continuous-time
state-space models | Continuous-time state-space models (SSMs) are flexible tools for analysing
irregularly sampled sequential observations that are driven by an underlying
state process. Corresponding applications typically involve restrictive
assumptions concerning linearity and Gaussianity to facilitate inference on the
model parameters via the Kalman filter. In this contribution, we provide a
general continuous-time SSM framework, allowing both the observation and the
state process to be non-linear and non-Gaussian. Statistical inference is
carried out by maximum approximate likelihood estimation, where multiple
numerical integration within the likelihood evaluation is performed via a fine
discretisation of the state process. The corresponding reframing of the SSM as
a continuous-time hidden Markov model, with structured state transitions,
enables us to apply the associated efficient algorithms for parameter
estimation and state decoding. We illustrate the modelling approach in a case
study using data from a longitudinal study on delinquent behaviour of
adolescents in Germany, revealing temporal persistence in the deviation of an
individual's delinquency level from the population mean.
| stat |
Variable selection in social-environmental data: Sparse regression and
tree ensemble machine learning approaches | Objective: Social-environmental data obtained from the U.S. Census is an
important resource for understanding health disparities, but rarely is the full
dataset utilized for analysis. A barrier to incorporating the full data is a
lack of solid recommendations for variable selection, with researchers often
hand-selecting a few variables. Thus, we evaluated the ability of empirical
machine learning approaches to identify social-environmental factors having a
true association with a health outcome.
Materials and Methods: We compared several popular machine learning methods,
including penalized regressions (e.g. lasso, elastic net), and tree ensemble
methods. Via simulation, we assessed the methods' ability to identify census
variables truly associated with binary and continuous outcomes while minimizing
false positive results (10 true associations, 1,000 total variables). We
applied the most promising method to the full census data (p=14,663 variables)
linked to prostate cancer registry data (n=76,186 cases) to identify
social-environmental factors associated with advanced prostate cancer.
Results: In simulations, we found that elastic net identified many
true-positive variables, while lasso provided good control of false positives.
Using a combined measure of accuracy, hierarchical clustering based on
Spearman's correlation with sparse group lasso regression performed the best
overall. Bayesian Adaptive Regression Trees outperformed other tree ensemble
methods, but not the sparse group lasso. In the full dataset, the sparse group
lasso successfully identified a subset of variables, three of which replicated
earlier findings.
Discussion: This analysis demonstrated the potential of empirical machine
learning approaches to identify a small subset of census variables having a
true association with the outcome, and that replicate across empiric methods.
| stat |
Algorithmic Analysis and Statistical Estimation of SLOPE via Approximate
Message Passing | SLOPE is a relatively new convex optimization procedure for high-dimensional
linear regression via the sorted l1 penalty: the larger the rank of the fitted
coefficient, the larger the penalty. This non-separable penalty renders many
existing techniques invalid or inconclusive in analyzing the SLOPE solution. In
this paper, we develop an asymptotically exact characterization of the SLOPE
solution under Gaussian random designs through solving the SLOPE problem using
approximate message passing (AMP). This algorithmic approach allows us to
approximate the SLOPE solution via the much more amenable AMP iterates.
Explicitly, we characterize the asymptotic dynamics of the AMP iterates relying
on a recently developed state evolution analysis for non-separable penalties,
thereby overcoming the difficulty caused by the sorted l1 penalty. Moreover, we
prove that the AMP iterates converge to the SLOPE solution in an asymptotic
sense, and numerical simulations show that the convergence is surprisingly
fast. Our proof rests on a novel technique that specifically leverages the
SLOPE problem. In contrast to prior literature, our work not only yields an
asymptotically sharp analysis but also offers an algorithmic, flexible, and
constructive approach to understanding the SLOPE problem.
| stat |
Explanations can be manipulated and geometry is to blame | Explanation methods aim to make neural networks more trustworthy and
interpretable. In this paper, we demonstrate a property of explanation methods
which is disconcerting for both of these purposes. Namely, we show that
explanations can be manipulated arbitrarily by applying visually hardly
perceptible perturbations to the input that keep the network's output
approximately constant. We establish theoretically that this phenomenon can be
related to certain geometrical properties of neural networks. This allows us to
derive an upper bound on the susceptibility of explanations to manipulations.
Based on this result, we propose effective mechanisms to enhance the robustness
of explanations.
| stat |
Aggregating predictions from experts: a scoping review of statistical
methods, experiments, and applications | Forecasts support decision making in a variety of applications. Statistical
models can produce accurate forecasts given abundant training data, but when
data is sparse, rapidly changing, or unavailable, statistical models may not be
able to make accurate predictions. Expert judgmental forecasts---models that
combine expert-generated predictions into a single forecast---can make
predictions when training data is limited by relying on expert intuition to
take the place of concrete training data. Researchers have proposed a wide
array of algorithms to combine expert predictions into a single forecast, but
there is no consensus on an optimal aggregation model. This scoping review
surveyed recent literature on aggregating expert-elicited predictions. We
gathered common terminology, aggregation methods, and forecasting performance
metrics, and offer guidance to strengthen future work that is growing at an
accelerated pace.
| stat |
Learning CHARME models with neural networks | In this paper, we consider a model called CHARME (Conditional Heteroscedastic
Autoregressive Mixture of Experts), a class of generalized mixture of nonlinear
nonparametric AR-ARCH time series. Under certain Lipschitz-type conditions on
the autoregressive and volatility functions, we prove that this model is
stationary, ergodic and $\tau$-weakly dependent. These conditions are much
weaker than those presented in the literature that treats this model. Moreover,
this result forms the theoretical basis for deriving an asymptotic theory of
the underlying (non)parametric estimation, which we present for this model. As
an application, from the universal approximation property of neural networks
(NN), we develop a learning theory for the NN-based autoregressive functions of
the model, where the strong consistency and asymptotic normality of the
considered estimator of the NN weights and biases are guaranteed under weak
conditions.
| stat |
Representation Learning: A Statistical Perspective | Learning representations of data is an important problem in statistics and
machine learning. While the origin of learning representations can be traced
back to factor analysis and multidimensional scaling in statistics, it has
become a central theme in deep learning with important applications in computer
vision and computational neuroscience. In this article, we review recent
advances in learning representations from a statistical perspective. In
particular, we review the following two themes: (a) unsupervised learning of
vector representations and (b) learning of both vector and matrix
representations.
| stat |
The Hyv\"arinen scoring rule in Gaussian linear time series models | Likelihood-based estimation methods involve the normalising constant of the
model distributions, expressed as a function of the parameter. However in many
problems this function is not easily available, and then less efficient but
more easily computed estimators may be attractive. In this work we study
stationary time-series models, and construct and analyse "score-matching''
estimators, that do not involve the normalising constant. We consider two
scenarios: a single series of increasing length, and an increasing number of
independent series of fixed length. In the latter case there are two variants,
one based on the full data, and another based on a sufficient statistic. We
study the empirical performance of these estimators in three special cases,
autoregressive (\AR), moving average (MA) and fractionally differenced white
noise (\ARFIMA) models, and make comparisons with full and pairwise likelihood
estimators. The results are somewhat model-dependent, with the new estimators
doing well for $\MA$ and \ARFIMA\ models, but less so for $\AR$ models.
| stat |
Interpretable Model Summaries Using the Wasserstein Distance | Statistical models often include thousands of parameters. However, large
models decrease the investigator's ability to interpret and communicate the
estimated parameters. Reducing the dimensionality of the parameter space in the
estimation phase is a commonly used approach, but less work has focused on
selecting subsets of the parameters for interpreting the estimated model --
especially in settings such as Bayesian inference and model averaging.
Importantly, many models do not have straightforward interpretations and create
another layer of obfuscation. To solve this gap, we introduce a new method that
uses the Wasserstein distance to identify a low-dimensional interpretable model
projection. After the estimation of complex models, users can budget how many
parameters they wish to interpret and the proposed generates a simplified model
of the desired dimension minimizing the distance to the full model. We provide
simulation results to illustrate the method and apply it to cancer datasets.
| stat |
Comparison of Multi-response Prediction Methods | While data science is battling to extract information from the enormous
explosion of data, many estimators and algorithms are being developed for
better prediction. Researchers and data scientists often introduce new methods
and evaluate them based on various aspects of data. However, studies on the
impact of/on a model with multiple response variables are limited. This study
compares some newly-developed (envelope) and well-established (PLS, PCR)
prediction methods based on real data and simulated data specifically designed
by varying properties such as multicollinearity, the correlation between
multiple responses and position of relevant principal components of predictors.
This study aims to give some insight into these methods and help the researcher
to understand and use them in further studies.
| stat |
How are emergent constraints quantifying uncertainty and what do they
leave behind? | The use of emergent constraints to quantify uncertainty for key policy
relevant quantities such as Equilibrium Climate Sensitivity (ECS) has become
increasingly widespread in recent years. Many researchers, however, claim that
emergent constraints are inappropriate or even under-report uncertainty. In
this paper we contribute to this discussion by examining the emergent
constraints methodology in terms of its underpinning statistical assumptions.
We argue that the existing frameworks are based on indefensible assumptions,
then show how weakening them leads to a more transparent Bayesian framework
wherein hitherto ignored sources of uncertainty, such as how reality might
differ from models, can be quantified. We present a guided framework for the
quantification of additional uncertainties that is linked to the confidence we
can have in the underpinning physical arguments for using linear constraints.
We provide a software tool for implementing our general framework for emergent
constraints and use it to illustrate the framework on a number of recent
emergent constraints for ECS. We find that the robustness of any constraint to
additional uncertainties depends strongly on the confidence we can have in the
underpinning physics, allowing a future framing of the debate over the validity
of a particular constraint around the underlying physical arguments, rather
than statistical assumptions.
| stat |
Avoiding Bias Due to Nonrandom Scheduling When Modeling Trends in
Home-Field Advantage | Existing approaches for estimating home-field advantage (HFA) include
modeling the difference between home and away scores as a function of the
difference between home and away team ratings that are treated either as fixed
or random effects. We uncover an upward bias in the mixed model HFA estimates
that is due to the nonrandom structure of the schedule -- and thus the random
effect design matrix -- and explore why the fixed effects model is not subject
to the same bias. Intraconference HFAs and standard errors are calculated for
each of 3 college sports and 3 professional sports over 18 seasons and then
fitted with conference-specific slopes and intercepts to measure the potential
linear population trend in HFA.
| stat |
Application of Clustering Analysis for Investigation of Food
Accessibility | Access to food assistance programs such as food pantries and food banks needs
focus in order to mitigate food insecurity. Accessibility to the food
assistance programs is impacted by demographics of the population and geography
of the location. It hence becomes imperative to define and identify food
assistance deserts (Under-served areas) within a given region to find out the
ways to improve the accessibility of food. Food banks, the supplier of food to
the food agencies serving the people, can manage its resources more efficiently
by targeting the food assistance deserts and increase the food supply in those
regions. This paper will examine the characteristics and structure of the food
assistance network in the region of Ohio by presenting the possible reasons of
food insecurity in this region and identify areas wherein food agencies are
needed or may not be needed. Gaussian Mixture Model (GMM) clustering technique
is employed to identify the possible reasons and address this problem of food
accessibility.
| stat |
When does gradient descent with logistic loss find interpolating
two-layer networks? | We study the training of finite-width two-layer smoothed ReLU networks for
binary classification using the logistic loss. We show that gradient descent
drives the training loss to zero if the initial loss is small enough. When the
data satisfies certain cluster and separation conditions and the network is
wide enough, we show that one step of gradient descent reduces the loss
sufficiently that the first result applies.
| stat |
The rational SPDE approach for Gaussian random fields with general
smoothness | A popular approach for modeling and inference in spatial statistics is to
represent Gaussian random fields as solutions to stochastic partial
differential equations (SPDEs) of the form $L^{\beta}u = \mathcal{W}$, where
$\mathcal{W}$ is Gaussian white noise, $L$ is a second-order differential
operator, and $\beta>0$ is a parameter that determines the smoothness of $u$.
However, this approach has been limited to the case $2\beta\in\mathbb{N}$,
which excludes several important models and makes it necessary to keep $\beta$
fixed during inference.
We propose a new method, the rational SPDE approach, which in spatial
dimension $d\in\mathbb{N}$ is applicable for any $\beta>d/4$, and thus remedies
the mentioned limitation. The presented scheme combines a finite element
discretization with a rational approximation of the function $x^{-\beta}$ to
approximate $u$. For the resulting approximation, an explicit rate of
convergence to $u$ in mean-square sense is derived. Furthermore, we show that
our method has the same computational benefits as in the restricted case
$2\beta\in\mathbb{N}$. Several numerical experiments and a statistical
application are used to illustrate the accuracy of the method, and to show that
it facilitates likelihood-based inference for all model parameters including
$\beta$.
| stat |
Recycling intermediate steps to improve Hamiltonian Monte Carlo | Hamiltonian Monte Carlo (HMC) and related algorithms have become routinely
used in Bayesian computation. In this article, we present a simple and provably
accurate method to improve the efficiency of HMC and related algorithms with
essentially no extra computational cost. This is achieved by {recycling the
intermediate states along simulated trajectories of Hamiltonian dynamics.
Standard algorithms use only the end points of trajectories, wastefully
discarding all the intermediate states. Compared to the alternative methods for
utilizing the intermediate states, our algorithm is simpler to apply in
practice and requires little programming effort beyond the usual
implementations of HMC and related algorithms. Our algorithm applies
straightforwardly to the no-U-turn sampler, arguably the most popular variant
of HMC. Through a variety of experiments, we demonstrate that our recycling
algorithm yields substantial computational efficiency gains.
| stat |
Scalable Grouped Gaussian Processes via Direct Cholesky Functional
Representations | We consider multi-task regression models where observations are assumed to be
a linear combination of several latent node and weight functions, all drawn
from Gaussian process (GP) priors that allow nonzero covariance between grouped
latent functions. We show that when these grouped functions are conditionally
independent given a group-dependent pivot, it is possible to parameterize the
prior through sparse Cholesky factors directly, hence avoiding their
computation during inference. Furthermore, we establish that kernels that are
multiplicatively separable over input points give rise to such sparse
parameterizations naturally without any additional assumptions. Finally, we
extend the use of these sparse structures to approximate posteriors within
variational inference, further improving scalability on the number of
functions. We test our approach on multi-task datasets concerning distributed
solar forecasting and show that it outperforms several multi-task GP baselines
and that our sparse specifications achieve the same or better accuracy than
non-sparse counterparts.
| stat |
Robust multi-stage model-based design of optimal experiments for
nonlinear estimation | We study approaches to robust model-based design of experiments in the
context of maximum-likelihood estimation. These approaches provide
robustification of model-based methodologies for the design of optimal
experiments by accounting for the effect of the parametric uncertainty. We
study the problem of robust optimal design of experiments in the framework of
nonlinear least-squares parameter estimation using linearized confidence
regions. We investigate several well-known robustification frameworks in this
respect and propose a novel methodology based on multi-stage robust
optimization. The proposed methodology aims at problems, where the experiments
are designed sequentially with a possibility of re-estimation in-between the
experiments. The multi-stage formalism aids in identifying experiments that are
better conducted in the early phase of experimentation, where parameter
knowledge is poor. We demonstrate the findings and effectiveness of the
proposed methodology using four case studies of varying complexity.
| stat |
Forecasting asylum-related migration flows with machine learning and
data at scale | The effects of the so-called "refugee crisis" of 2015-16 continue to dominate
the political agenda in Europe. Migration flows were sudden and unexpected,
leaving governments unprepared and exposing significant shortcomings in the
field of migration forecasting. Migration is a complex system typified by
episodic variation, underpinned by causal factors that are interacting, highly
context dependent and short-lived. Correspondingly, migration monitoring relies
on scattered data, while approaches to forecasting focus on specific migration
flows and often have inconsistent results that are difficult to generalise at
the regional or global levels.
Here we show that adaptive machine learning algorithms that integrate
official statistics and non-traditional data sources at scale can effectively
forecast asylum-related migration flows. We focus on asylum applications lodged
in countries of the European Union (EU) by nationals of all countries of origin
worldwide; the same approach can be applied in any context provided adequate
migration or asylum data are available.
We exploit three tiers of data - geolocated events and internet searches in
countries of origin, detections of irregular crossings at the EU border, and
asylum recognition rates in countries of destination - to effectively forecast
individual asylum-migration flows up to four weeks ahead with high accuracy.
Uniquely, our approach a) monitors potential drivers of migration in countries
of origin to detect changes early onset; b) models individual
country-to-country migration flows separately and on moving time windows; c)
estimates the effects of individual drivers, including lagged effects; d)
provides forecasts of asylum applications up to four weeks ahead; e) assesses
how patterns of drivers shift over time to describe the functioning and change
of migration systems.
| stat |
Getting Started with Particle Metropolis-Hastings for Inference in
Nonlinear Dynamical Models | This tutorial provides a gentle introduction to the particle
Metropolis-Hastings (PMH) algorithm for parameter inference in nonlinear
state-space models together with a software implementation in the statistical
programming language R. We employ a step-by-step approach to develop an
implementation of the PMH algorithm (and the particle filter within) together
with the reader. This final implementation is also available as the package
pmhtutorial in the CRAN repository. Throughout the tutorial, we provide some
intuition as to how the algorithm operates and discuss some solutions to
problems that might occur in practice. To illustrate the use of PMH, we
consider parameter inference in a linear Gaussian state-space model with
synthetic data and a nonlinear stochastic volatility model with real-world
data.
| stat |
Detecting and Analyzing Mobility Hotspots using Surface Networks | Capabilities for collecting and storing data on mobile objects have increased
dramatically over the past few decades. A persistent difficulty is summarizing
large collections of mobile objects. This paper develops methods for extracting
and analyzing hotspots or locations with relatively high levels of mobility
activity. We use kernel density estimation (KDE) to convert a large collection
of mobile objects into a smooth, continuous surface. We then develop a
topological algorithm to extract critical geometric features of the surface;
these include critical points (peaks, pits and passes) and critical lines
(ridgelines and course-lines). We connect the peaks and corresponding
ridgelines to produce a surface network that summarizes the topological
structure of the surface. We apply graph theoretic indices to analytically
characterize the surface and its changes over time. To illustrate our approach,
we apply the techniques to taxi cab data collected in Shanghai, China. We find
increases in the complexity of the hotspot spatial distribution during normal
activity hours in the late morning, afternoon and evening and a spike in the
connectivity of the hotspot spatial distribution in the morning as taxis
concentrate on servicing travel to work. These results match with scientific
and anecdotal knowledge about human activity patterns in the study area.
| stat |
A Basis Approach to Surface Clustering | This paper presents a novel method for clustering surfaces. The proposal
involves first using basis functions in a tensor product to smooth the data and
thus reduce the dimension to a finite number of coefficients, and then using
these estimated coefficients to cluster the surfaces via the k-means algorithm.
An extension of the algorithm to clustering tensors is also discussed. We show
that the proposed algorithm exhibits the property of strong consistency, with
or without measurement errors, in correctly clustering the data as the sample
size increases. Simulation studies suggest that the proposed method outperforms
the benchmark k-means algorithm which uses the original vectorized data. In
addition, an EGG real data example is considered to illustrate the practical
application of the proposal.
| stat |
Transparency, Auditability and eXplainability of Machine Learning Models
in Credit Scoring | A major requirement for credit scoring models is to provide a maximally
accurate risk prediction. Additionally, regulators demand these models to be
transparent and auditable. Thus, in credit scoring, very simple predictive
models such as logistic regression or decision trees are still widely used and
the superior predictive power of modern machine learning algorithms cannot be
fully leveraged. Significant potential is therefore missed, leading to higher
reserves or more credit defaults. This paper works out different dimensions
that have to be considered for making credit scoring models understandable and
presents a framework for making ``black box'' machine learning models
transparent, auditable and explainable. Following this framework, we present an
overview of techniques, demonstrate how they can be applied in credit scoring
and how results compare to the interpretability of score cards. A real world
case study shows that a comparable degree of interpretability can be achieved
while machine learning techniques keep their ability to improve predictive
power.
| stat |
Skills to not fall behind in school | Many recent studies emphasize how important the role of cognitive and
social-emotional skills can be in determining people's quality of life.
Although skills are of great importance in many aspects, in this paper we will
focus our efforts to better understand the relationship between several types
of skills with academic progress delay. Our dataset contains the same students
in 2012 and 2017, and we consider that there was a academic progress delay for
a specific student if he or she progressed less than expected in school grades.
Our methodology primarily includes the use of a Bayesian logistic regression
model and our results suggest that both cognitive and social-emotional skills
may impact the conditional probability of falling behind in school, and the
magnitude of the impact between the two types of skills can be comparable.
| stat |
Compositionally-warped additive mixed modeling for a wide variety of
non-Gaussian spatial data | As with the advancement of geographical information systems, non-Gaussian
spatial data is getting larger and more diverse. Considering this background,
this study develops a general framework for fast and flexible non-Gaussian
regression, especially for spatial/spatiotemporal modeling. The developed
model, termed the compositionally-warped additive mixed model (CAMM), combines
an additive mixed model (AMM) and the compositionally-warped Gaussian process
to model a wide variety of non-Gaussian continuous data including spatial and
other effects. Specific advantages of the proposed CAMM requires no explicit
assumption of data distribution unlike existing AMMs, and fast estimation
through a restricted likelihood maximization balancing the modeling accuracy
and complexity. Monte Carlo experiments show the estimation accuracy and
computational efficiency of CAMM for modeling non-Gaussian data including
fat-tailed and/or skewed distributions. Finally, the proposed approach is
applied to crime data to examine the empirical performance of the regression
analysis and prediction. The proposed approach is implemented in an R package
spmoran. See details on how to implement CAMM, see
https://github.com/dmuraka/spmoran.
| stat |
Estimating Causal Effects with the Neural Autoregressive Density
Estimator | Estimation of causal effects is fundamental in situations were the underlying
system will be subject to active interventions. Part of building a causal
inference engine is defining how variables relate to each other, that is,
defining the functional relationship between variables given conditional
dependencies. In this paper, we deviate from the common assumption of linear
relationships in causal models by making use of neural autoregressive density
estimators and use them to estimate causal effects within the Pearl's
do-calculus framework. Using synthetic data, we show that the approach can
retrieve causal effects from non-linear systems without explicitly modeling the
interactions between the variables.
| stat |
A Meta-Learning Framework for Generalized Zero-Shot Learning | Learning to classify unseen class samples at test time is popularly referred
to as zero-shot learning (ZSL). If test samples can be from training (seen) as
well as unseen classes, it is a more challenging problem due to the existence
of strong bias towards seen classes. This problem is generally known as
\emph{generalized} zero-shot learning (GZSL). Thanks to the recent advances in
generative models such as VAEs and GANs, sample synthesis based approaches have
gained considerable attention for solving this problem. These approaches are
able to handle the problem of class bias by synthesizing unseen class samples.
However, these ZSL/GZSL models suffer due to the following key limitations:
$(i)$ Their training stage learns a class-conditioned generator using only
\emph{seen} class data and the training stage does not \emph{explicitly} learn
to generate the unseen class samples; $(ii)$ They do not learn a generic
optimal parameter which can easily generalize for both seen and unseen class
generation; and $(iii)$ If we only have access to a very few samples per seen
class, these models tend to perform poorly. In this paper, we propose a
meta-learning based generative model that naturally handles these limitations.
The proposed model is based on integrating model-agnostic meta learning with a
Wasserstein GAN (WGAN) to handle $(i)$ and $(iii)$, and uses a novel task
distribution to handle $(ii)$. Our proposed model yields significant
improvements on standard ZSL as well as more challenging GZSL setting. In ZSL
setting, our model yields 4.5\%, 6.0\%, 9.8\%, and 27.9\% relative improvements
over the current state-of-the-art on CUB, AWA1, AWA2, and aPY datasets,
respectively.
| stat |
Optimal Best-arm Identification in Linear Bandits | We study the problem of best-arm identification with fixed confidence in
stochastic linear bandits. The objective is to identify the best arm with a
given level of certainty while minimizing the sampling budget. We devise a
simple algorithm whose sampling complexity matches known instance-specific
lower bounds, asymptotically almost surely and in expectation. The algorithm
relies on an arm sampling rule that tracks an optimal proportion of arm draws,
and that remarkably can be updated as rarely as we wish, without compromising
its theoretical guarantees. Moreover, unlike existing best-arm identification
strategies, our algorithm uses a stopping rule that does not depend on the
number of arms. Experimental results suggest that our algorithm significantly
outperforms existing algorithms. The paper further provides a first analysis of
the best-arm identification problem in linear bandits with a continuous set of
arms.
| stat |
Stochastic Recursive Variance Reduction for Efficient Smooth Non-Convex
Compositional Optimization | Stochastic compositional optimization arises in many important machine
learning tasks such as value function evaluation in reinforcement learning and
portfolio management. The objective function is the composition of two
expectations of stochastic functions, and is more challenging to optimize than
vanilla stochastic optimization problems. In this paper, we investigate the
stochastic compositional optimization in the general smooth non-convex setting.
We employ a recently developed idea of \textit{Stochastic Recursive Gradient
Descent} to design a novel algorithm named SARAH-Compositional, and prove a
sharp Incremental First-order Oracle (IFO) complexity upper bound for
stochastic compositional optimization: $\mathcal{O}((n+m)^{1/2}
\varepsilon^{-2})$ in the finite-sum case and $\mathcal{O}(\varepsilon^{-3})$
in the online case. Such a complexity is known to be the best one among IFO
complexity results for non-convex stochastic compositional optimization, and is
believed to be optimal. Our experiments validate the theoretical performance of
our algorithm.
| stat |
Deriving Priors for Bayesian Prediction of Daily Response Propensity in
Responsive Survey Design: Historical Data Analysis vs. Literature Review | Responsive Survey Design (RSD) aims to increase the efficiency of survey data
collection via live monitoring of paradata and the introduction of protocol
changes when survey errors and increased costs seem imminent. Daily predictions
of response propensity for all active sampled cases are among the most
important quantities for live monitoring of data collection outcomes, making
sound predictions of these propensities essential for the success of RSD.
Because it relies on real-time updates of prior beliefs about key design
quantities, such as predicted response propensities, RSD stands to benefit from
Bayesian approaches. However, empirical evidence of the merits of these
approaches is lacking in the literature, and the derivation of informative
prior distributions is required for these approaches to be effective. In this
paper, we evaluate the ability of two approaches to deriving prior
distributions for the coefficients defining daily response propensity models to
improve predictions of daily response propensity in a real data collection
employing RSD. The first approach involves analyses of historical data from the
same survey, and the second approach involves literature review. We find that
Bayesian methods based on these two approaches result in higher-quality
predictions of response propensity than more standard approaches ignoring prior
information. This is especially true during the early-to-middle periods of data
collection when interventions are often considered in an RSD framework.
| stat |
K-bMOM: a robust Lloyd-type clustering algorithm based on bootstrap
Median-of-Means | We propose a new clustering algorithm that is robust to the presence of
outliers in the dataset. We perform Lloyd-type iterations with robust estimates
of the centroids. More precisely, we build on the idea of median-of-means
statistics to estimate the centroids, but allow for replacement while
constructing the blocks. We call this methodology the bootstrap median-of-means
(bMOM) and prove that if enough blocks are generated through the bootstrap
sampling, then it has a better breakdown point for mean estimation than the
classical median-of-means (MOM), where the blocks form a partition of the
dataset. From a clustering perspective, bMOM enables to take many blocks of a
desired size, thus avoiding possible disappearance of clusters in some blocks,
a pitfall that can occur for the partition-based generation of blocks of the
classical median-of-means. Experiments on simulated datasets show that the
proposed approach, called K-bMOM, performs better than existing robust K-means
based methods. Guidelines are provided for tuning the hyper-parameters K-bMOM
in practice. It is also recommended to the practitionner to use such a robust
approach to initialize their clustering algorithm. Finally, considering a
simplified and theoretical version of our estimator, we prove its robustness to
adversarial contamination by deriving robust rates of convergence for the
K-means distorsion. To our knowledge, it is the first result of this kind for
the K-means distorsion.
| stat |
A New Correlation Coefficient for Aggregating Non-strict and Incomplete
Rankings | We introduce a correlation coefficient that is designed to deal with a
variety of ranking formats including those containing non-strict (i.e.,
with-ties) and incomplete (i.e., unknown) preferences. The correlation
coefficient is designed to enforce a neutral treatment of incompleteness
whereby no assumptions are made about individual preferences involving unranked
objects. The new measure, which can be regarded as a generalization of the
seminal Kendall tau correlation coefficient, is proven to satisfy a set of
metric-like axioms and to be equivalent to a recently developed ranking
distance function associated with Kemeny aggregation. In an effort to further
unify and enhance both robust ranking methodologies, this work proves the
equivalence of an additional distance and correlation-coefficient pairing in
the space of non-strict incomplete rankings. These connections induce new exact
optimization methodologies: a specialized branch and bound algorithm and an
exact integer programming formulation. Moreover, the bridging of these
complementary theories reinforces the singular suitability of the featured
correlation coefficient to solve the general consensus ranking problem. The
latter premise is bolstered by an accompanying set of experiments on random
instances, which are generated via a herein developed sampling technique
connected with the classic Mallows distribution of ranking data. Associated
experiments with the branch and bound algorithm demonstrate that, as data
becomes noisier, the featured correlation coefficient yields relatively fewer
alternative optimal solutions and that the aggregate rankings tend to be closer
to an underlying ground truth shared by a majority.
| stat |
Recurrent Flow Networks: A Recurrent Latent Variable Model for
Spatio-Temporal Density Modelling | When modelling real-valued sequences, a typical approach in current RNN
architectures is to use a Gaussian mixture model to describe the conditional
output distribution. In this paper, we argue that mixture-based distributions
could exhibit structural limitations when faced with highly complex data
distributions such as for spatial densities. To address this issue, we
introduce recurrent flow networks which combine deterministic and stochastic
recurrent hidden states with conditional normalizing flows to form a
probabilistic neural generative model capable of describing the kind of
variability observed in highly structured spatio-temporal data. Inspired by the
model's factorization, we further devise a structured variational inference
network to approximate the intractable posterior distribution by exploiting a
spatial representation of the data. We empirically evaluate our model against
other generative models for sequential data on three real-world datasets for
the task of spatio-temporal transportation demand modelling. Results show how
the added flexibility allows our model to generate distributions matching
potentially complex urban topologies.
| stat |
Hop Sampling: A Simple Regularized Graph Learning for Non-Stationary
Environments | Graph representation learning is gaining popularity in a wide range of
applications, such as social networks analysis, computational biology, and
recommender systems. However, different with positive results from many
academic studies, applying graph neural networks (GNNs) in a real-world
application is still challenging due to non-stationary environments. The
underlying distribution of streaming data changes unexpectedly, resulting in
different graph structures (a.k.a., concept drift). Therefore, it is essential
to devise a robust graph learning technique so that the model does not overfit
to the training graphs. In this work, we present Hop Sampling, a
straightforward regularization method that can effectively prevent GNNs from
overfishing. The hop sampling randomly selects the number of propagation steps
rather than fixing it, and by doing so, it encourages the model to learn
meaningful node representation for all intermediate propagation layers and to
experience a variety of plausible graphs that are not in the training set.
Particularly, we describe the use case of our method in recommender systems, a
representative example of the real-world non-stationary case. We evaluated hop
sampling on a large-scale real-world LINE dataset and conducted an online A/B/n
test in LINE Coupon recommender systems of LINE Wallet Tab. Experimental
results demonstrate that the proposed scheme improves the prediction accuracy
of GNNs. We observed hop sampling provides 7.97% and 16.93% improvements for
NDCG and MAP compared to non-regularized GNN models in our online service.
Furthermore, models using hop sampling alleviate the oversmoothing issue in
GNNs enabling a deeper model as well as more diversified representation.
| stat |
Bayesian inference for stochastic differential equation mixed effects
models of a tumor xenography study | We consider Bayesian inference for stochastic differential equation mixed
effects models (SDEMEMs) exemplifying tumor response to treatment and regrowth
in mice. We produce an extensive study on how a SDEMEM can be fitted using both
exact inference based on pseudo-marginal MCMC and approximate inference via
Bayesian synthetic likelihoods (BSL). We investigate a two-compartments SDEMEM,
these corresponding to the fractions of tumor cells killed by and survived to a
treatment, respectively. Case study data considers a tumor xenography study
with two treatment groups and one control, each containing 5-8 mice. Results
from the case study and from simulations indicate that the SDEMEM is able to
reproduce the observed growth patterns and that BSL is a robust tool for
inference in SDEMEMs. Finally, we compare the fit of the SDEMEM to a similar
ordinary differential equation model. Due to small sample sizes, strong prior
information is needed to identify all model parameters in the SDEMEM and it
cannot be determined which of the two models is the better in terms of
predicting tumor growth curves. In a simulation study we find that with a
sample of 17 mice per group BSL is able to identify all model parameters and
distinguish treatment groups.
| stat |
The effect of geographic sampling on evaluation of extreme precipitation
in high resolution climate models | Traditional approaches for comparing global climate models and observational
data products typically fail to account for the geographic location of the
underlying weather station data. For modern high-resolution models, this is an
oversight since there are likely grid cells where the physical output of a
climate model is compared with a statistically interpolated quantity instead of
actual measurements of the climate system. In this paper, we quantify the
impact of geographic sampling on the relative performance of high resolution
climate models' representation of precipitation extremes in Boreal winter (DJF)
over the contiguous United States (CONUS), comparing model output from five
early submissions to the HighResMIP subproject of the CMIP6 experiment. We find
that properly accounting for the geographic sampling of weather stations can
significantly change the assessment of model performance. Across the models
considered, failing to account for sampling impacts the different metrics
(extreme bias, spatial pattern correlation, and spatial variability) in
different ways (both increasing and decreasing). We argue that the geographic
sampling of weather stations should be accounted for in order to yield a more
straightforward and appropriate comparison between models and observational
data sets, particularly for high resolution models. While we focus on the CONUS
in this paper, our results have important implications for other global land
regions where the sampling problem is more severe.
| stat |
The Bayesian Method of Tensor Networks | Bayesian learning is a powerful learning framework which combines the
external information of the data (background information) with the internal
information (training data) in a logically consistent way in inference and
prediction. By Bayes rule, the external information (prior distribution) and
the internal information (training data likelihood) are combined coherently,
and the posterior distribution and the posterior predictive (marginal)
distribution obtained by Bayes rule summarize the total information needed in
the inference and prediction, respectively. In this paper, we study the
Bayesian framework of the Tensor Network from two perspective. First, we
introduce the prior distribution to the weights in the Tensor Network and
predict the labels of the new observations by the posterior predictive
(marginal) distribution. Since the intractability of the parameter integral in
the normalization constant computation, we approximate the posterior predictive
distribution by Laplace approximation and obtain the out-product approximation
of the hessian matrix of the posterior distribution of the Tensor Network
model. Second, to estimate the parameters of the stationary mode, we propose a
stable initialization trick to accelerate the inference process by which the
Tensor Network can converge to the stationary path more efficiently and stably
with gradient descent method. We verify our work on the MNIST, Phishing Website
and Breast Cancer data set. We study the Bayesian properties of the Bayesian
Tensor Network by visualizing the parameters of the model and the decision
boundaries in the two dimensional synthetic data set. For a application
purpose, our work can reduce the overfitting and improve the performance of
normal Tensor Network model.
| stat |
Directed Networks with a Differentially Private Bi-degree Sequence | Although a lot of approaches are developed to release network data with a
differentially privacy guarantee, inference using noisy data in many network
models is still unknown or not properly explored. In this paper, we release the
bi-degree sequences of directed networks using the Laplace mechanism and use
the $p_0$ model for inferring the degree parameters. The $p_0$ model is an
exponential random graph model with the bi-degree sequence as its exclusively
sufficient statistic. We show that the estimator of the parameter without the
denoised process is asymptotically consistent and normally distributed. This is
contrast sharply with some known results that valid inference such as the
existence and consistency of the estimator needs the denoised process. Along
the way, a new phenomenon is revealed in which an additional variance factor
appears in the asymptotic variance of the estimator when the noise becomes
large. Further, we propose an efficient algorithm for finding the closet point
lying in the set of all graphical bi-degree sequences under the global $L_1$
optimization problem. Numerical studies demonstrate our theoretical findings.
| stat |
Estimation and selection for high-order Markov chains with Bayesian
mixture transition distribution models | We develop two models for Bayesian estimation and selection in high-order,
discrete-state Markov chains. Both are based on the mixture transition
distribution, which constructs a transition probability tensor with additive
mixing of probabilities from first-order transition matrices. We demonstrate
two uses for the proposed models: parsimonious approximation of high-order
dynamics by mixing lower-order transition models, and order/lag selection
through over-specification and shrinkage via priors for sparse probability
vectors. The priors further shrink all models to an identifiable and
interpretable parameterization, useful for data analysis. We discuss properties
of the models and demonstrate their utility with simulation studies. We further
apply the methodology to a data analysis from the high-order Markov chain
literature and to a time series of pink salmon abundance in a creek in Alaska,
U.S.A.
| stat |
Matrix Completion with Quantified Uncertainty through Low Rank Gaussian
Copula | Modern large scale datasets are often plagued with missing entries. For
tabular data with missing values, a flurry of imputation algorithms solve for a
complete matrix which minimizes some penalized reconstruction error. However,
almost none of them can estimate the uncertainty of its imputations. This paper
proposes a probabilistic and scalable framework for missing value imputation
with quantified uncertainty. Our model, the Low Rank Gaussian Copula, augments
a standard probabilistic model, Probabilistic Principal Component Analysis,
with marginal transformations for each column that allow the model to better
match the distribution of the data. It naturally handles Boolean, ordinal, and
real-valued observations and quantifies the uncertainty in each imputation. The
time required to fit the model scales linearly with the number of rows and the
number of columns in the dataset. Empirical results show the method yields
state-of-the-art imputation accuracy across a wide range of data types,
including those with high rank. Our uncertainty measure predicts imputation
error well: entries with lower uncertainty do have lower imputation error (on
average). Moreover, for real-valued data, the resulting confidence intervals
are well-calibrated.
| stat |
Cluster validation by measurement of clustering characteristics relevant
to the user | There are many cluster analysis methods that can produce quite different
clusterings on the same dataset. Cluster validation is about the evaluation of
the quality of a clustering; "relative cluster validation" is about using such
criteria to compare clusterings. This can be used to select one of a set of
clusterings from different methods, or from the same method ran with different
parameters such as different numbers of clusters.
There are many cluster validation indexes in the literature. Most of them
attempt to measure the overall quality of a clustering by a single number, but
this can be inappropriate. There are various different characteristics of a
clustering that can be relevant in practice, depending on the aim of
clustering, such as low within-cluster distances and high between-cluster
separation.
In this paper, a number of validation criteria will be introduced that refer
to different desirable characteristics of a clustering, and that characterise a
clustering in a multidimensional way. In specific applications the user may be
interested in some of these criteria rather than others. A focus of the paper
is on methodology to standardise the different characteristics so that users
can aggregate them in a suitable way specifying weights for the various
criteria that are relevant in the clustering application at hand.
| stat |
Bayesian Temporal Factorization for Multidimensional Time Series
Prediction | Large-scale and multidimensional spatiotemporal data sets are becoming
ubiquitous in many real-world applications such as monitoring urban traffic and
air quality. Making predictions on these time series has become a critical
challenge due to not only the large-scale and high-dimensional nature but also
the considerable amount of missing data. In this paper, we propose a Bayesian
temporal factorization (BTF) framework for modeling multidimensional time
series -- in particular spatiotemporal data -- in the presence of missing
values. By integrating low-rank matrix/tensor factorization and vector
autoregressive (VAR) process into a single probabilistic graphical model, this
framework can characterize both global and local consistencies in large-scale
time series data. The graphical model allows us to effectively perform
probabilistic predictions and produce uncertainty estimates without imputing
those missing values. We develop efficient Gibbs sampling algorithms for model
inference and model updating for real-time prediction and test the proposed BTF
framework on several real-world spatiotemporal data sets for both missing data
imputation and multi-step rolling prediction tasks. The numerical experiments
demonstrate the superiority of the proposed BTF approaches over existing
state-of-the-art methods.
| stat |
Spatiotemporal dynamic of COVID-19 mortality in the city of Sao Paulo,
Brazil: shifting the high risk from the best to the worst socio-economic
conditions | Currently, Brazil has one of the fastest increasing COVID-19 epidemics in the
world, that has caused at least 94 thousand confirmed deaths until now. The
city of Sao Paulo is particularly vulnerable because it is the most populous in
the country. Analyzing the spatiotemporal dynamics of COVID-19 is important to
help the urgent need to integrate better actions to face the pandemic. Thus,
this study aimed to analyze the COVID-19 mortality, from March to July 2020,
considering the spatio-time architectures, the socio-economic context of the
population, and using a fine granular level, in the most populous city in
Brazil. For this, we conducted an ecological study, using secondary public data
from the mortality information system. We describe mortality rates for each
epidemiological week and the entire period by sex and age. We modelled the
deaths using spatiotemporal and spatial architectures and Poisson probability
distributions in a latent Gaussian Bayesian model approach. We obtained the
relative risks for temporal and spatiotemporal trends and socio-economic
conditions. To reduce possible sub notification, we considered the confirmed
and suspected deaths. Our findings showed an apparent stabilization of the
temporal trend, at the end of the period, but that may change in the future.
Mortality rate increased with increasing age and was higher in men. The risk of
death was greater in areas with the worst social conditions throughout the
study period. However, this was not a uniform pattern over time, since we
identified a shift from the high risk in the areas with best socio-economic
conditions to the worst ones. Our study contributed by emphasizing the
importance of geographic screening in areas with a higher risk of death, and,
currently, worse socio-economic contexts, as a crucial aspect to reducing
disease mortality and health inequities, through integrated public health
actions.
| stat |
Locally associated graphical models | The notion of multivariate total positivity has proved to be useful in
finance and psychology but may be too restrictive in other applications. In
this paper we propose a concept of local association, where highly connected
components in a graphical model are positively associated and study its
properties. Our main motivation comes from gene expression data, where
graphical models have become a popular exploratory tool. We focus the
exposition on Gaussian distributions but our methods readily extend to
non-paranormal distributions. Motivated by a convex optimization problem that
arises in this context, we develop a GOLAZO approach that generalizes a number
of optimization procedures that arise in the context of graphical models (e.g.
the GLASSO) and propose a simple block-coordinate descent optimization
procedure for solving the dual problem. Our results on existence of the optimum
for such problems are of separate interest.
| stat |
Integrative High Dimensional Multiple Testing with Heterogeneity under
Data Sharing Constraints | Identifying informative predictors in a high dimensional regression model is
a critical step for association analysis and predictive modeling. Signal
detection in the high dimensional setting often fails due to the limited sample
size. One approach to improve power is through meta-analyzing multiple studies
on the same scientific question. However, integrative analysis of high
dimensional data from multiple studies is challenging in the presence of
between study heterogeneity. The challenge is even more pronounced with
additional data sharing constraints under which only summary data but not
individual level data can be shared across different sites. In this paper, we
propose a novel data shielding integrative large-scale testing (DSILT) approach
to signal detection by allowing between study heterogeneity and not requiring
sharing of individual level data. Assuming the underlying high dimensional
regression models of the data differ across studies yet share similar support,
the DSILT approach incorporates proper integrative estimation and debiasing
procedures to construct test statistics for the overall effects of specific
covariates. We also develop a multiple testing procedure to identify
significant effects while controlling for false discovery rate (FDR) and false
discovery proportion (FDP). Theoretical comparisons of the DSILT procedure with
the ideal individual--level meta--analysis (ILMA) approach and other
distributed inference methods are investigated. Simulation studies demonstrate
that the DSILT procedure performs well in both false discovery control and
attaining power. The proposed method is applied to a real example on detecting
interaction effect of the genetic variants for statins and obesity on the risk
for Type 2 Diabetes.
| stat |
A dynamic stochastic blockmodel for interaction lengths | We propose a new dynamic stochastic blockmodel that focuses on the analysis
of interaction lengths in networks. The model does not rely on a discretization
of the time dimension and may be used to analyze networks that evolve
continuously over time. The framework relies on a clustering structure on the
nodes, whereby two nodes belonging to the same latent group tend to create
interactions and non-interactions of similar lengths. We introduce a fast
variational expectation-maximization algorithm to perform inference, and adapt
a widely used clustering criterion to perform model choice. Finally, we test
our methodology on artificial data, and propose a demonstration on a dataset
concerning face-to-face interactions between students in a high-school.
| stat |
Measuring the Stability of Learned Features | Many modern datasets don't fit neatly into $n \times p$ matrices, but most
techniques for measuring statistical stability expect rectangular data. We
study methods for stability assessment on non-rectangular data, using
statistical learning algorithms to extract rectangular latent features. We
design controlled simulations to characterize the power and practicality of
competing approaches. This motivates new strategies for visualizing feature
stability. Our stability curves supplement the direct analysis, providing
information about the reliability of inferences based on learned features.
Finally, we illustrate our approach using a spatial proteomics dataset, where
machine learning tools can augment the scientist's workflow, but where
guarantees of statistical reproducibility are still central. Our raw data,
packaged code, and experimental outputs are publicly available.
| stat |
Bayesian Hierarchical Models for High-Dimensional Mediation Analysis
with Coordinated Selection of Correlated Mediators | We consider Bayesian high-dimensional mediation analysis to identify among a
large set of correlated potential mediators the active ones that mediate the
effect from an exposure variable to an outcome of interest. Correlations among
mediators are commonly observed in modern data analysis; examples include the
activated voxels within connected regions in brain image data, regulatory
signals driven by gene networks in genome data and correlated exposure data
from the same source. When correlations are present among active mediators,
mediation analysis that fails to account for such correlation can be
sub-optimal and may lead to a loss of power in identifying active mediators.
Building upon a recent high-dimensional mediation analysis framework, we
propose two Bayesian hierarchical models, one with a Gaussian mixture prior
that enables correlated mediator selection and the other with a Potts mixture
prior that accounts for the correlation among active mediators in mediation
analysis. We develop efficient sampling algorithms for both methods. Various
simulations demonstrate that our methods enable effective identification of
correlated active mediators, which could be missed by using existing methods
that assume prior independence among active mediators. The proposed methods are
applied to the LIFECODES birth cohort and the Multi-Ethnic Study of
Atherosclerosis (MESA) and identified new active mediators with important
biological implications.
| stat |
Estimating FARIMA models with uncorrelated but non-independent error
terms | In this paper we derive the asymptotic properties of the least squares
estimator (LSE) of fractionally integrated autoregressive moving-average
(FARIMA) models under the assumption that the errors are uncorrelated but not
necessarily independent nor martingale differences. We relax considerably the
independence and even the martingale difference assumptions on the innovation
process to extend the range of application of the FARIMA models. We propose a
consistent estimator of the asymptotic covariance matrix of the LSE which may
be very different from that obtained in the standard framework. A
self-normalized approach to confidence interval construction for weak FARIMA
model parameters is also presented. All our results are done under a mixing
assumption on the noise. Finally, some simulation studies and an application to
the daily returns of stock market indices are presented to corroborate our
theoretical work.
| stat |
Adaptive Randomization in Network Data | Network data have appeared frequently in recent research. For example, in
comparing the effects of different types of treatment, network models have been
proposed to improve the quality of estimation and hypothesis testing. In this
paper, we focus on efficiently estimating the average treatment effect using an
adaptive randomization procedure in networks. We work on models of causal
frameworks, for which the treatment outcome of a subject is affected by its own
covariate as well as those of its neighbors. Moreover, we consider the case in
which, when we assign treatments to the current subject, only the subnetwork of
existing subjects is revealed. New randomized procedures are proposed to
minimize the mean squared error of the estimated differences between treatment
effects. In network data, it is usually difficult to obtain theoretical
properties because the numbers of nodes and connections increase
simultaneously. Under mild assumptions, our proposed procedure is closely
related to a time-varying inhomogeneous Markov chain. We then use Lyapunov
functions to derive the theoretical properties of the proposed procedures. The
advantages of the proposed procedures are also demonstrated by extensive
simulations and experiments on real network data.
| stat |
Using extreme value theory for the estimation of risk metrics for
capacity adequacy assessment | This paper investigates the use of extreme value theory for modelling the
distribution of demand-net-of-wind for capacity adequacy assessment. Extreme
value theory approaches are well-established and mathematically justified
methods for estimating the tails of a distribution and so are ideally suited
for problems in capacity adequacy, where normally only the tails of the
relevant distributions are significant. The extreme value theory peaks over
threshold approach is applied directly to observations of demand-net-of-wind,
meaning that no assumption is needed about the nature of any dependence between
demand and wind.
The methodology is tested on data from Great Britain and compared to two
alternative approaches: use of the empirical distribution of demand-net-of-wind
and use of a model which assumes independence between demand and wind. Extreme
value theory is shown to produce broadly similar estimates of risk metrics to
the use of the above empirical distribution but with smaller sampling
uncertainty. Estimates of risk metrics differ when the approach assuming
independence is used, especially when data across different historical years
are pooled, suggesting that assuming independence might result in the over- or
under-estimation of risk metrics.
| 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.