title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
A fast tunable blurring algorithm for scattered data | A blurring algorithm with linear time complexity can reduce the small-scale
content of data observed at scattered locations in a spatially extended domain
of arbitrary dimension. The method works by forming a Gaussian interpolant of
the input data, and then convolving the interpolant with a multiresolution
Gaussian approximation of the Green's function to a differential operator whose
spectrum can be tuned for problem-specific considerations. Like conventional
blurring algorithms, which the new algorithm generalizes to data measured at
locations other than a uniform grid, applications include deblurring and
separation of spatial scales. An example illustrates a possible application
toward enabling importance sampling approaches to data assimilation of
geophysical observations, which are often scattered over a spatial domain,
since blurring observations can make particle filters more effective at state
estimation of large scales. Another example, motivated by data analysis of
dynamics like ocean eddies that have strong separation of spatial scales, uses
the algorithm to decompose scattered oceanographic float measurements into
large-scale and small-scale components.
| stat |
Globally-centered autocovariances in MCMC | Autocovariances are a fundamental quantity of interest in Markov chain Monte
Carlo (MCMC) simulations with autocorrelation function (ACF) plots being an
integral visualization tool for performance assessment. Unfortunately, for slow
mixing Markov chains, the empirical autocovariance can highly underestimate the
truth. For multiple-chain MCMC sampling, we propose a globally-centered
estimator of the autocovariance function (G-ACvF) that exhibits significant
theoretical and empirical improvements. We show that the bias of the G-ACvF
estimator is smaller than the bias of the current state-of-the-art. The impact
of this improved estimator is evident in three critical output analysis
applications: (1) ACF plots, (2) estimates of the Monte Carlo asymptotic
covariance matrix, and (3) estimates of the effective sample size. Under weak
conditions, we establish strong consistency of our improved asymptotic
covariance estimator, and obtain its large-sample bias and variance. The
performance of the new estimators is demonstrated through various examples.
| stat |
Look-Ahead Screening Rules for the Lasso | The lasso is a popular method to induce shrinkage and sparsity in the
solution vector (coefficients) of regression problems, particularly when there
are many predictors relative to the number of observations. Solving the lasso
in this high-dimensional setting can, however, be computationally demanding.
Fortunately, this demand can be alleviated via the use of screening rules that
discard predictors prior to fitting the model, leading to a reduced problem to
be solved. In this paper, we present a new screening strategy: look-ahead
screening. Our method uses safe screening rules to find a range of penalty
values for which a given predictor cannot enter the model, thereby screening
predictors along the remainder of the path. In experiments we show that these
look-ahead screening rules improve the performance of existing screening
strategies.
| stat |
An anisotropic model for global climate data | We present a new, elementary way to obtain axially symmetric Gaussian
processes on the sphere, in order to accommodate for the directional anisotropy
of global climate data in geostatistical analysis.
| stat |
pexm: a JAGS module for applications involving the piecewise exponential
distribution | In this study, we present a new module built for users interested in a
programming language similar to BUGS to fit a Bayesian model based on the
piecewise exponential (PE) distribution. The module is an extension to the
open-source program JAGS by which a Gibbs sampler can be applied without
requiring the derivation of complete conditionals and the subsequent
implementation of strategies to draw samples from unknown distributions. The PE
distribution is widely used in the fields of survival analysis and reliability.
Currently, it can only be implemented in JAGS through methods to indirectly
specify the likelihood based on the Poisson or Bernoulli probabilities. Our
module provides a more straightforward implementation and is thus more
attractive to the researchers aiming to spend more time exploring the results
from the Bayesian inference rather than implementing the Markov Chain Monte
Carlo (MCMC) algorithm. For those interested in extending JAGS, this work can
be seen as a tutorial including important information not well investigated or
organized in other materials. Here, we describe how to use the module taking
advantage of the interface between R and JAGS. A short simulation study is
developed to ensure that the module behaves well and a real illustration,
involving two PE models, exhibits a context where the module can be used in
practice.
| stat |
A study on tuning parameter selection for the high-dimensional lasso | High-dimensional predictive models, those with more measurements than
observations, require regularization to be well defined, perform well
empirically, and possess theoretical guarantees. The amount of regularization,
often determined by tuning parameters, is integral to achieving good
performance. One can choose the tuning parameter in a variety of ways, such as
through resampling methods or generalized information criteria. However, the
theory supporting many regularized procedures relies on an estimate for the
variance parameter, which is complicated in high dimensions. We develop a suite
of information criteria for choosing the tuning parameter in lasso regression
by leveraging the literature on high-dimensional variance estimation. We derive
intuition showing that existing information-theoretic approaches work poorly in
this setting. We compare our risk estimators to existing methods with an
extensive simulation and derive some theoretical justification. We find that
our new estimators perform well across a wide range of simulation conditions
and evaluation criteria.
| stat |
Bayesian hierarchical modeling and analysis for physical activity
trajectories using actigraph data | Rapid developments in streaming data technologies are continuing to generate
increased interest in monitoring human activity. Wearable devices, such as
wrist-worn sensors that monitor gross motor activity (actigraphy), have become
prevalent. An actigraph unit continually records the activity level of an
individual, producing a very large amount of data at a high-resolution that can
be immediately downloaded and analyzed. While this kind of \textit{big data}
includes both spatial and temporal information, the variation in such data
seems to be more appropriately modeled by considering stochastic evolution
through time while accounting for spatial information separately. We propose a
comprehensive Bayesian hierarchical modeling and inferential framework for
actigraphy data reckoning with the massive sizes of such databases while
attempting to offer full inference. Building upon recent developments in this
field, we construct Nearest Neighbour Gaussian Processes (NNGPs) for actigraphy
data to compute at large temporal scales. More specifically, we construct a
temporal NNGP and we focus on the optimized implementation of the collapsed
algorithm in this specific context. This approach permits improved model
scaling while also offering full inference. We test and validate our methods on
simulated data and subsequently apply and verify their predictive ability on an
original dataset concerning a health study conducted by the Fielding School of
Public Health of the University of California, Los Angeles.
| stat |
Deep Generalized Method of Moments for Instrumental Variable Analysis | Instrumental variable analysis is a powerful tool for estimating causal
effects when randomization or full control of confounders is not possible. The
application of standard methods such as 2SLS, GMM, and more recent variants are
significantly impeded when the causal effects are complex, the instruments are
high-dimensional, and/or the treatment is high-dimensional. In this paper, we
propose the DeepGMM algorithm to overcome this. Our algorithm is based on a new
variational reformulation of GMM with optimal inverse-covariance weighting that
allows us to efficiently control very many moment conditions. We further
develop practical techniques for optimization and model selection that make it
particularly successful in practice. Our algorithm is also computationally
tractable and can handle large-scale datasets. Numerical results show our
algorithm matches the performance of the best tuned methods in standard
settings and continues to work in high-dimensional settings where even recent
methods break.
| stat |
Supporting Multi-point Fan Design with Dimension Reduction | Motivated by the idea of turbomachinery active subspace performance maps,
this paper studies dimension reduction in turbomachinery 3D CFD simulations.
First, we show that these subspaces exist across different blades---under the
same parametrization---largely independent of their Mach number or Reynolds
number. This is demonstrated via a numerical study on three different blades.
Then, in an attempt to reduce the computational cost of identifying a suitable
dimension reducing subspace, we examine statistical sufficient dimension
reduction methods, including sliced inverse regression, sliced average variance
estimation, principal Hessian directions and contour regression. Unsatisfied by
these results, we evaluate a new idea based on polynomial variable
projection---a non-linear least squares problem. Our results using polynomial
variable projection clearly demonstrate that one can accurately identify
dimension reducing subspaces for turbomachinery functionals at a fraction of
the cost associated with prior methods. We apply these subspaces to the problem
of comparing design configurations across different flight points on a working
line of a fan blade. We demonstrate how designs that offer a healthy compromise
between performance at cruise and sea-level conditions can be easily found by
visually inspecting their subspaces.
| stat |
Do Methodological Birds of a Feather Flock Together? | Quasi-experimental methods have proliferated over the last two decades, as
researchers develop causal inference tools for settings in which randomization
is infeasible. Two popular such methods, difference-in-differences (DID) and
comparative interrupted time series (CITS), compare observations before and
after an intervention in a treated group to an untreated comparison group
observed over the same period. Both methods rely on strong, untestable
counterfactual assumptions. Despite their similarities, the methodological
literature on CITS lacks the mathematical formality of DID. In this paper, we
use the potential outcomes framework to formalize two versions of CITS - a
general version described by Bloom (2005) and a linear version often used in
health services research. We then compare these to two corresponding DID
formulations - one with time fixed effects and one with time fixed effects and
group trends. We also re-analyze three previously published studies using these
methods. We demonstrate that the most general versions of CITS and DID impute
the same counterfactuals and estimate the same treatment effects. The only
difference between these two designs is the language used to describe them and
their popularity in distinct disciplines. We also show that these designs
diverge when one constrains them using linearity (CITS) or parallel trends
(DID). We recommend defaulting to the more flexible versions and provide advice
to practitioners on choosing between the more constrained versions by
considering the data-generating mechanism. We also recommend greater attention
to specifying the outcome model and counterfactuals in papers, allowing for
transparent evaluation of the plausibility of causal assumptions.
| stat |
Hard Shape-Constrained Kernel Machines | Shape constraints (such as non-negativity, monotonicity, convexity) play a
central role in a large number of applications, as they usually improve
performance for small sample size and help interpretability. However enforcing
these shape requirements in a hard fashion is an extremely challenging problem.
Classically, this task is tackled (i) in a soft way (without out-of-sample
guarantees), (ii) by specialized transformation of the variables on a
case-by-case basis, or (iii) by using highly restricted function classes, such
as polynomials or polynomial splines. In this paper, we prove that hard affine
shape constraints on function derivatives can be encoded in kernel machines
which represent one of the most flexible and powerful tools in machine learning
and statistics. Particularly, we present a tightened second-order cone
constrained reformulation, that can be readily implemented in convex solvers.
We prove performance guarantees on the solution, and demonstrate the efficiency
of the approach in joint quantile regression with applications to economics and
to the analysis of aircraft trajectories, among others.
| stat |
Statistical modelling of conidial discharge of entomophthoralean fungi
using a newly discovered Pandora species | Entomophthoralean fungi are insect pathogenic fungi and are characterized by
their active discharge of infective conidia that infect insects. Our aim was to
study the effects of temperature on the discharge and to characterize the
variation in the associated temporal pattern of a newly discovered Pandora
species with focus on peak location and shape of the discharge. Mycelia were
incubated at various temperatures in darkness, and conidial discharge was
measured over time. We used a novel modification of a statistical model
(pavpop), that simultaneously estimates phase and amplitude effects, into a
setting of generalized linear models. This model is used to test hypotheses of
peak location and discharge of conidia. The statistical analysis showed that
high temperature leads to an early and fast decreasing peak, whereas there were
no significant differences in total number of discharged conidia. Using the
proposed model we also quantified the biological variation in the timing of the
peak location at a fixed temperature.
| stat |
Variance Reduction with Array-RQMC for Tau-Leaping Simulation of
Stochastic Biological and Chemical Reaction Networks | We explore the use of Array-RQMC, a randomized quasi-Monte Carlo method
designed for the simulation of Markov chains, to reduce the variance when
simulating stochastic biological or chemical reaction networks with
$\tau$-leaping. The task is to estimate the expectation of a function of
molecule copy numbers at a given future time $T$ by the sample average over $n$
sample paths, and the goal is to reduce the variance of this sample-average
estimator. We find that when the method is properly applied, variance
reductions by factors in the thousands can be obtained. These factors are much
larger than those observed previously by other authors who tried RQMC methods
for the same examples. Array-RQMC simulates an array of realizations of the
Markov chain and requires a sorting function to reorder these chains according
to their states, after each step. The choice of sorting function is a key
ingredient for the efficiency of the method. We illustrate this by comparing
various choices. The expected number of reactions of each type per step also
has an impact on the efficiency gain.
| stat |
Bifidelity data-assisted neural networks in nonintrusive reduced-order
modeling | In this paper, we present a new nonintrusive reduced basis method when a
cheap low-fidelity model and expensive high-fidelity model are available. The
method relies on proper orthogonal decomposition (POD) to generate the
high-fidelity reduced basis and a shallow multilayer perceptron to learn the
high-fidelity reduced coefficients. In contrast to other methods, one distinct
feature of the proposed method is to incorporate the features extracted from
the low-fidelity data as the input feature, this approach not only improves the
predictive capability of the neural network but also enables the decoupling the
high-fidelity simulation from the online stage. Due to its nonintrusive nature,
it is applicable to general parameterized problems. We also provide several
numerical examples to illustrate the effectiveness and performance of the
proposed method.
| stat |
A novel dowscaling procedure for compositional data in the Aitchison
geometry with application to soil texture data | In this work, we present a novel downscaling procedure for compositional
quantities based on the Aitchison geometry. The method is able to naturally
consider compositional constraints, i.e. unit-sum and positivity. We show that
the method can be used in a block sequential Gaussian simulation framework in
order to assess the variability of downscaled quantities. Finally, to validate
the method, we test it first in an idealized scenario and then apply it for the
downscaling of digital soil maps on a more realistic case study. The digital
soil maps for the realistic case study are obtained from SoilGrids, a system
for automated soil mapping based on state-of-the-art spatial predictions
methods.
| stat |
Crime Prediction by Data-Driven Green's Function method | We develop an algorithm that forecasts cascading events, by employing a
Green's function scheme on the basis of the self-exciting point process model.
This method is applied to open data of 10 types of crimes happened in Chicago.
It shows a good prediction accuracy superior to or comparable to the standard
methods which are the expectation-maximization method and prospective hotspot
maps method. We find a cascade influence of the crimes that has a long-time,
logarithmic tail; this result is consistent with an earlier study on
burglaries. This long-tail feature cannot be reproduced by the other standard
methods. In addition, a merit of the Green's function method is the low
computational cost in the case of high density of events and/or large amount of
the training data.
| stat |
Climate Change and Grain Production Fluctuations | 50 year-long time series from a Long Term Agronomic Experiment have been used
to to investigate the effects of climate change on yields of Wheat and Maize.
Trends and fluctuations, useful to estimate production forecasts and related
risks are compared to national ones, a classical regional climatic index as
Western Mediterranean Oscillation Index, and a global one given by Sun Spot
Number. Data, denoised by EMD and SSA, show how SSN oscillations slowing down
in the last decades, affects regional scale dynamics, where in the last two
decades a range of fluctuations (7-16 years) are also evident. Both signals
reflects on yield fluctuations of Wheat and Maize both a national and local
level.
| stat |
Towards a Complete Picture of Stationary Covariance Functions on Spheres
Cross Time | With the advent of wide-spread global and continental-scale spatiotemporal
datasets, increased attention has been given to covariance functions on spheres
over time. This paper provides results for stationary covariance functions of
random fields defined over $d$-dimensional spheres cross time. Specifically, we
provide a bridge between the characterization in \cite{berg-porcu} for
covariance functions on spheres cross time and Gneiting's lemma
\citep{gneiting2002} that deals with planar surfaces.
We then prove that there is a valid class of covariance functions similar in
form to the Gneiting class of space-time covariance functions
\citep{gneiting2002} that replaces the squared Euclidean distance with the
great circle distance. Notably, the provided class is shown to be positive
definite on every $d$-dimensional sphere cross time, while the Gneiting class
is positive definite over $\R^d \times \R$ for fixed $d$ only.
In this context, we illustrate the value of our adapted Gneiting class by
comparing examples from this class to currently established nonseparable
covariance classes using out-of-sample predictive criteria. These comparisons
are carried out on two climate reanalysis datasets from the National Centers
for Environmental Prediction and National Center for Atmospheric Research. For
these datasets, we show that examples from our covariance class have better
predictive performance than competing models.
| stat |
First derivatives at the optimum analysis (\textit{fdao}): An approach
to estimate the uncertainty in nonlinear regression involving stochastically
independent variables | An important problem of optimization analysis surges when parameters such as
$ \{\theta_j\}_{j=1,\, \dots \,,k }$, determining a function $
y=f(x\given\{\theta_j\}) $, must be estimated from a set of observables $ \{
x_i,y_i\}_{i=1,\, \dots \,,m} $. Where $ \{x_i\} $ are independent variables
assumed to be uncertainty-free. It is known that analytical solutions are
possible if $ y=f(x\given\theta_j) $ is a linear combination of $
\{\theta_{j=1,\, \dots \,,k} \}.$ Here it is proposed that determining the
uncertainty of parameters that are not \textit{linearly independent} may be
achieved from derivatives $ \tfrac{\partial f(x \given \{\theta_j\})}{\partial
\theta_j} $ at an optimum, if the parameters are \textit{stochastically
independent}.
| stat |
Faster Non-Convex Federated Learning via Global and Local Momentum | In this paper, we propose \texttt{FedGLOMO}, the first (first-order) FL
algorithm that achieves the optimal iteration complexity (i.e matching the
known lower bound) on smooth non-convex objectives -- without using clients'
full gradient in each round. Our key algorithmic idea that enables attaining
this optimal complexity is applying judicious momentum terms that promote
variance reduction in both the local updates at the clients, and the global
update at the server. Our algorithm is also provably optimal even with
compressed communication between the clients and the server, which is an
important consideration in the practical deployment of FL algorithms. Our
experiments illustrate the intrinsic variance reduction effect of
\texttt{FedGLOMO} which implicitly suppresses client-drift in heterogeneous
data distribution settings and promotes communication-efficiency. As a prequel
to \texttt{FedGLOMO}, we propose \texttt{FedLOMO} which applies momentum only
in the local client updates. We establish that \texttt{FedLOMO} enjoys improved
convergence rates under common non-convex settings compared to prior work, and
with fewer assumptions.
| stat |
Comparing methods addressing multi-collinearity when developing
prediction models | Clinical prediction models are developed widely across medical disciplines.
When predictors in such models are highly collinear, unexpected or spurious
predictor-outcome associations may occur, thereby potentially reducing
face-validity and explainability of the prediction model. Collinearity can be
dealt with by exclusion of collinear predictors, but when there is no a priori
motivation (besides collinearity) to include or exclude specific predictors,
such an approach is arbitrary and possibly inappropriate. We compare different
methods to address collinearity, including shrinkage, dimensionality reduction,
and constrained optimization. The effectiveness of these methods is illustrated
via simulations. In the conducted simulations, no effect of collinearity was
observed on predictive outcomes. However, a negative effect of collinearity on
the stability of predictor selection was found, affecting all compared methods,
but in particular methods that perform strong predictor selection (e.g.,
Lasso).}
| stat |
Anomaly Detection with Inexact Labels | We propose a supervised anomaly detection method for data with inexact
anomaly labels, where each label, which is assigned to a set of instances,
indicates that at least one instance in the set is anomalous. Although many
anomaly detection methods have been proposed, they cannot handle inexact
anomaly labels. To measure the performance with inexact anomaly labels, we
define the inexact AUC, which is our extension of the area under the ROC curve
(AUC) for inexact labels. The proposed method trains an anomaly score function
so that the smooth approximation of the inexact AUC increases while anomaly
scores for non-anomalous instances become low. We model the anomaly score
function by a neural network-based unsupervised anomaly detection method, e.g.,
autoencoders. The proposed method performs well even when only a small number
of inexact labels are available by incorporating an unsupervised anomaly
detection mechanism with inexact AUC maximization. Using various datasets, we
experimentally demonstrate that our proposed method improves the anomaly
detection performance with inexact anomaly labels, and outperforms existing
unsupervised and supervised anomaly detection and multiple instance learning
methods.
| stat |
On the Use of Auxiliary Variables in Multilevel Regression and
Poststratification | Multilevel regression and poststratification (MRP) has been a popular
approach for selection bias adjustment and subgroup estimation, with successful
and widespread applications from social sciences to health sciences. We
demonstrate the capability of MRP to handle the methodological and
computational issues in data integration and inferences of probability and
nonprobability-based surveys, and the broad extensions in practical
applications. Our development is motivated by the Adolescent Brain Cognitive
Development (ABCD) Study that has collected children across 21 U.S. geographic
locations for national representation but is subject to selection bias, a
common problem of nonprobability samples. Though treated as the gold standard
in public opinion research, MRP is a statistical technique that has assumptions
and pitfalls, the validity of which prominently depends on the quality of
available auxiliary information. In this paper, we develop the statistical
foundation of how to incorporate auxiliary variables under MRP. We build up a
systematic framework under MRP for statistical data integration and inferences.
Our simulation studies indicate the statistical validity of MRP with a tradeoff
between robustness and efficiency and present the improvement over alternative
methods. We apply the approach to evaluate cognition performances of diverse
groups of children in the ABCD study and find that the adjustment of auxiliary
variables has a substantial effect on the inference results.
| stat |
Instance-Dependent Cost-Sensitive Learning for Detecting Transfer Fraud | Card transaction fraud is a growing problem affecting card holders worldwide.
Financial institutions increasingly rely upon data-driven methods for
developing fraud detection systems, which are able to automatically detect and
block fraudulent transactions. From a machine learning perspective, the task of
detecting fraudulent transactions is a binary classification problem.
Classification models are commonly trained and evaluated in terms of
statistical performance measures, such as likelihood and AUC, respectively.
These measures, however, do not take into account the actual business
objective, which is to minimize the financial losses due to fraud. Fraud
detection is to be acknowledged as an instance-dependent cost-sensitive
classification problem, where the costs due to misclassification vary between
instances, and requiring adapted approaches for learning a classification
model. In this article, an instance-dependent threshold is derived, based on
the instance-dependent cost matrix for transfer fraud detection, that allows
for making the optimal cost-based decision for each transaction. Two novel
classifiers are presented, based on lasso-regularized logistic regression and
gradient tree boosting, which directly minimize the proposed instance-dependent
cost measure when learning a classification model. The proposed methods are
implemented in the R packages cslogit and csboost, and compared against
state-of-the-art methods on a publicly available data set from the machine
learning competition website Kaggle and a proprietary card transaction data
set. The results of the experiments highlight the potential of reducing fraud
losses by adopting the proposed methods.
| stat |
Reluctant generalized additive modeling | Sparse generalized additive models (GAMs) are an extension of sparse
generalized linear models which allow a model's prediction to vary non-linearly
with an input variable. This enables the data analyst build more accurate
models, especially when the linearity assumption is known to be a poor
approximation of reality. Motivated by reluctant interaction modeling (Yu et
al. 2019), we propose a multi-stage algorithm, called $\textit{reluctant
generalized additive modeling (RGAM)}$, that can fit sparse generalized
additive models at scale. It is guided by the principle that, if all else is
equal, one should prefer a linear feature over a non-linear feature. Unlike
existing methods for sparse GAMs, RGAM can be extended easily to binary, count
and survival data. We demonstrate the method's effectiveness on real and
simulated examples.
| stat |
Sparse Regression at Scale: Branch-and-Bound rooted in First-Order
Optimization | We consider the least squares regression problem, penalized with a
combination of the $\ell_{0}$ and squared $\ell_{2}$ penalty functions (a.k.a.
$\ell_0 \ell_2$ regularization). Recent work shows that the resulting
estimators are of key importance in many high-dimensional statistical settings.
However, exact computation of these estimators remains a major challenge.
Indeed, modern exact methods, based on mixed integer programming (MIP), face
difficulties when the number of features $p \sim 10^4$. In this work, we
present a new exact MIP framework for $\ell_0\ell_2$-regularized regression
that can scale to $p \sim 10^7$, achieving speedups of at least $5000$x,
compared to state-of-the-art exact methods. Unlike recent work, which relies on
modern commercial MIP solvers, we design a specialized nonlinear
branch-and-bound (BnB) framework, by critically exploiting the problem
structure. A key distinguishing component in our framework lies in efficiently
solving the node relaxations using a specialized first-order method, based on
coordinate descent (CD). Our CD-based method effectively leverages information
across the BnB nodes, through using warm starts, active sets, and gradient
screening. In addition, we design a novel method for obtaining dual bounds from
primal CD solutions, which certifiably works in high dimensions. Experiments on
synthetic and real high-dimensional datasets demonstrate that our framework is
not only significantly faster than the state of the art, but can also deliver
certifiably optimal solutions to statistically challenging instances that
cannot be handled with existing methods. We open source the implementation
through our toolkit L0BnB.
| stat |
Risk Convergence of Centered Kernel Ridge Regression with Large
Dimensional Data | This paper carries out a large dimensional analysis of a variation of kernel
ridge regression that we call \emph{centered kernel ridge regression} (CKRR),
also known in the literature as kernel ridge regression with offset. This
modified technique is obtained by accounting for the bias in the regression
problem resulting in the old kernel ridge regression but with \emph{centered}
kernels. The analysis is carried out under the assumption that the data is
drawn from a Gaussian distribution and heavily relies on tools from random
matrix theory (RMT). Under the regime in which the data dimension and the
training size grow infinitely large with fixed ratio and under some mild
assumptions controlling the data statistics, we show that both the empirical
and the prediction risks converge to a deterministic quantities that describe
in closed form fashion the performance of CKRR in terms of the data statistics
and dimensions. Inspired by this theoretical result, we subsequently build a
consistent estimator of the prediction risk based on the training data which
allows to optimally tune the design parameters. A key insight of the proposed
analysis is the fact that asymptotically a large class of kernels achieve the
same minimum prediction risk. This insight is validated with both synthetic and
real data.
| stat |
A Compartment Model of Human Mobility and Early Covid-19 Dynamics in NYC | In this paper, we build a mechanistic system to understand the relation
between a reduction in human mobility and Covid-19 spread dynamics within New
York City. To this end, we propose a multivariate compartmental model that
jointly models smartphone mobility data and case counts during the first 90
days of the epidemic. Parameter calibration is achieved through the formulation
of a general Bayesian hierarchical model to provide uncertainty quantification
of resulting estimates. The open-source probabilistic programming language Stan
is used for the requisite computation. With the additional benefit of modeling
a mechanism by which human mobility and infection counts relate, we find our
simple and interpretable model is able to recover epidemiological parameters
that are consistent with current literature.
| stat |
A novel approach to bivariate meta-analysis of binary outcomes and its
application in the context of surrogate endpoints | Bivariate meta-analysis provides a useful framework for combining information
across related studies and has been widely utilised to combine evidence from
clinical studies in order to evaluate treatment efficacy. Bivariate
meta-analysis has also been used to investigate surrogacy patterns between
treatment effects on the surrogate and the final outcome. Surrogate endpoints
play an important role in drug development when they can be used to measure
treatment effect early compared to the final clinical outcome and to predict
clinical benefit or harm. The standard bivariate meta-analytic approach models
the observed treatment effects on the surrogate and final outcomes jointly, at
both the within-study and between-studies levels, using a bivariate normal
distribution. For binomial data a normal approximation can be used on log odds
ratio scale, however, this method may lead to biased results when the
proportions of events are close to one or zero, affecting the validation of
surrogate endpoints. In this paper, two Bayesian meta-analytic approaches are
introduced which allow for modelling the within-study variability using
binomial data directly. The first uses independent binomial likelihoods to
model the within-study variability avoiding to approximate the observed
treatment effects, however, ignores the within-study association. The second,
models the summarised events in each arm jointly using a bivariate copula with
binomial marginals. This allows the model to take into account the within-study
association through the copula dependence parameter. We applied the methods to
an illustrative example in chronic myeloid leukemia to investigate the
surrogate relationship between complete cytogenetic response (CCyR) and
event-free-survival (EFS).
| stat |
The Impact of Time Series Length and Discretization on Longitudinal
Causal Estimation Methods | The use of observational time series data to assess the impact of multi-time
point interventions is becoming increasingly common as more health and activity
data are collected and digitized via wearables, social media, and electronic
health records. Such time series may involve hundreds or thousands of
irregularly sampled observations. One common analysis approach is to simplify
such time series by first discretizing them into sequences before applying a
discrete-time estimation method that adjusts for time-dependent confounding. In
certain settings, this discretization results in sequences with many time
points; however, the empirical properties of longitudinal causal estimators
have not been systematically compared on long sequences. We compare three
representative longitudinal causal estimation methods on simulated and real
clinical data. Our simulations and analyses assume a Markov structure and that
longitudinal treatments/exposures are binary-valued and have at most a single
jump point. We identify sources of bias that arise from temporally discretizing
the data and provide practical guidance for discretizing data and choosing
between methods when working with long sequences. Additionally, we compare
these estimators on real electronic health record data, evaluating the impact
of early treatment for patients with a life-threatening complication of
infection called sepsis.
| stat |
Initialization methods for optimum average silhouette width clustering | A unified clustering approach that can estimate number of clusters and
produce clustering against this number simultaneously is proposed. Average
silhouette width (ASW) is a widely used standard cluster quality index. A
distance based objective function that optimizes ASW for clustering is defined.
The proposed algorithm named as OSil, only, needs data observations as an input
without any prior knowledge of the number of clusters. This work is about
thorough investigation of the proposed methodology, its usefulness and
limitations. A vast spectrum of clustering structures were generated, and
several well-known clustering methods including partitioning, hierarchical,
density based, and spatial methods were consider as the competitor of the
proposed methodology. Simulation reveals that OSil algorithm has shown superior
performance in terms of clustering quality than all clustering methods included
in the study. OSil can find well separated, compact clusters and have shown
better performance for the estimation of number of clusters as compared to
several methods. Apart from the proposal of the new methodology and it's
investigation the paper offers a systematic analysis on the estimation of
cluster indices, some of which never appeared together in comparative
simulation setup before. The study offers many insightful findings useful for
the selection of the clustering methods and indices for clustering quality
assessment.
| stat |
Estimating the Number of Fatal Victims of the Peruvian Internal Armed
Conflict, 1980-2000: an application of modern multi-list Capture-Recapture
techniques | We estimate the number of fatal victims of the Peruvian internal armed
conflict between 1980-2000 using stratified seven-list Capture-Recapture
methods based on Dirichlet process mixtures, which we extend to accommodate
incomplete stratification information. We use matched data from six sources,
originally analyzed by the Peruvian Truth and Reconciliation Commission in
2003, together with a new large dataset, originally published in 2006 by the
Peruvian government. We deal with missing stratification labels by developing a
general framework and estimation methods based on MCMC sampling for jointly
fitting generic Bayesian Capture-Recapture models and the missing labels.
Through a detailed exploration driven by domain-knowledge, modeling and
refining, with special precautions to avoid cherry-picking of results, we
arrive to a conservative posterior estimate of 58,234 (CI95% = [56,741,
61,289]), and a more liberal estimate of 65,958 (CI95% = [61,462, 75,387])
fatal victims. We also determine that the Shining Path guerrillas killed more
people than the Peruvian armed forces. We additionally explore and discuss
estimates based on log-linear modeling and multiple-imputation. We finish by
discussing several lessons learned about the use of Capture-Recapture methods
for estimating casualties in conflicts.
| stat |
Four-Point, 2D, Free-Ranging, IMSPE-Optimal, Twin-Point Designs | We report the discovery of a set of four-point, two-factor, free-ranging,
putatively IMSPE-optimal designs with a pair of twin points, in the statistical
design of computer experiments, under Gaussian-process,
fixed-Gaussian-covariance parameter, and zero-nugget assumptions. We conjecture
this is the set of free-ranging, twin-point designs with the smallest number of
degrees of freedom.
| stat |
Variational State and Parameter Estimation | This paper considers the problem of computing Bayesian estimates of both
states and model parameters for nonlinear state-space models. Generally, this
problem does not have a tractable solution and approximations must be utilised.
In this work, a variational approach is used to provide an assumed density
which approximates the desired, intractable, distribution. The approach is
deterministic and results in an optimisation problem of a standard form. Due to
the parametrisation of the assumed density selected first- and second-order
derivatives are readily available which allows for efficient solutions. The
proposed method is compared against state-of-the-art Hamiltonian Monte Carlo in
two numerical examples.
| stat |
High-dimensional nonparametric monotone function estimation using BART | For the estimation of a regression relationship between Y and a large set of
potential predictors x 1 , . . . , x p , the flexible nature of a nonparametric
approach such as BART (Bayesian Additive Regression Trees) allows for a much
richer set of possibilities than a more restrictive parametric approach.
However, it may often occur that subject matter considerations suggest the
relationship will be monotone in one or more of the predictors. For such
situations, we propose monotone BART, a constrained version of BART that uses
the monotonicity information to improve function estimation without the need of
using a parametric form. Imposing monotonicity, when appropriate, results in
(i) function estimates that are smoother and more interpretable, (ii) better
out-of-sample predictive performance, (iii) less uncertainty, and (iv) less
sensitivity to prior choice. While some of the key aspects of the unconstrained
BART model carry over directly to monotone BART, the imposition of the
monotonicity constraints necessitates a fundamental rethinking of how the model
is implemented. In particular, in the original BART algorithm, the Markov Chain
Monte Carlo algorithm relied on a conditional conjugacy that is no longer
available in a high-dimensional, constrained space.
| stat |
Generalization bounds for deep learning | Generalization in deep learning has been the topic of much recent theoretical
and empirical research. Here we introduce desiderata for techniques that
predict generalization errors for deep learning models in supervised learning.
Such predictions should 1) scale correctly with data complexity; 2) scale
correctly with training set size; 3) capture differences between architectures;
4) capture differences between optimization algorithms; 5) be quantitatively
not too far from the true error (in particular, be non-vacuous); 6) be
efficiently computable; and 7) be rigorous. We focus on generalization error
upper bounds, and introduce a categorisation of bounds depending on assumptions
on the algorithm and data. We review a wide range of existing approaches, from
classical VC dimension to recent PAC-Bayesian bounds, commenting on how well
they perform against the desiderata.
We next use a function-based picture to derive a marginal-likelihood
PAC-Bayesian bound. This bound is, by one definition, optimal up to a
multiplicative constant in the asymptotic limit of large training sets, as long
as the learning curve follows a power law, which is typically found in practice
for deep learning problems. Extensive empirical analysis demonstrates that our
marginal-likelihood PAC-Bayes bound fulfills desiderata 1-3 and 5. The results
for 6 and 7 are promising, but not yet fully conclusive, while only desideratum
4 is currently beyond the scope of our bound. Finally, we comment on why this
function-based bound performs significantly better than current parameter-based
PAC-Bayes bounds.
| stat |
Multiple Imputation and Synthetic Data Generation with the R package
NPBayesImputeCat | In many contexts, missing data and disclosure control are ubiquitous and
challenging issues. In particular at statistical agencies, the respondent-level
data they collect from surveys and censuses can suffer from high rates of
missingness. Furthermore, agencies are obliged to protect respondents' privacy
when publishing the collected data for public use. The NPBayesImputeCat R
package, introduced in this paper, provides routines to i) create multiple
imputations for missing data, and ii) create synthetic data for statistical
disclosure control, for multivariate categorical data, with or without
structural zeros. We describe the Dirichlet process mixture of products of
multinomial distributions model used in the package, and illustrate various
uses of the package using data samples from the American Community Survey
(ACS). We also compare results of the missing data imputation to the mice R
package and those of the synthetic data generation to the synthpop R package.
| stat |
A non-convex regularization approach for stable estimation of loss
development factors | In this article, we apply non-convex regularization methods in order to
obtain stable estimation of loss development factors in insurance claims
reserving. Among the non-convex regularization methods, we focus on the use of
the log-adjusted absolute deviation (LAAD) penalty and provide discussion on
optimization of LAAD penalized regression model, which we prove to converge
with a coordinate descent algorithm under mild conditions. This has the
advantage of obtaining a consistent estimator for the regression coefficients
while allowing for the variable selection, which is linked to the stable
estimation of loss development factors. We calibrate our proposed model using a
multi-line insurance dataset from a property and casualty insurer where we
observed reported aggregate loss along accident years and development periods.
When compared to other regression models, our LAAD penalized regression model
provides very promising results.
| stat |
Latency Analysis of Consortium Blockchained Federated Learning | A decentralized federated learning architecture is proposed to apply to the
Businesses-to-Businesses scenarios by introducing the consortium blockchain in
this paper. We introduce a model verification mechanism to ensure the quality
of local models trained by participators. To analyze the latency of the system,
a latency model is constructed by considering the work flow of the
architecture. Finally the experiment results show that our latency model does
well in quantifying the actual delays.
| stat |
Insights into the "cross-world" independence assumption of causal
mediation analysis | Causal mediation analysis is a useful tool for epidemiological research, but
it has been criticized for relying on a "cross-world" independence assumption
that is empirically difficult to verify and problematic to justify based on
background knowledge. In the present article we aim to assist the applied
researcher in understanding this assumption. Synthesizing what is known about
the cross-world independence assumption, we discuss the relationship between
assumptions for causal mediation analyses, causal models, and non-parametric
identification of natural direct and indirect effects. In particular we give a
practical example of an applied setting where the cross-world independence
assumption is violated even without any post-treatment confounding. Further, we
review possible alternatives to the cross-world independence assumption,
including the use of computation of bounds that avoid the assumption
altogether. Finally, we carry out a numerical study in which the cross-world
independence assumption is violated to assess the ensuing bias in estimating
natural direct and indirect effects. We conclude with recommendations for
carrying out causal mediation analyses.
| stat |
A Bayesian framework for functional calibration of expensive
computational models through non-isometric matching | We study statistical calibration, i.e., adjusting features of a computational
model that are not observable or controllable in its associated physical
system. We focus on functional calibration, which arises in many manufacturing
processes where the unobservable features, called calibration variables, are a
function of the input variables. A major challenge in many applications is that
computational models are expensive and can only be evaluated a limited number
of times. Furthermore, without making strong assumptions, the calibration
variables are not identifiable. We propose Bayesian non-isometric matching
calibration (BNMC) that allows calibration of expensive computational models
with only a limited number of samples taken from a computational model and its
associated physical system. BNMC replaces the computational model with a
dynamic Gaussian process (GP) whose parameters are trained in the calibration
procedure. To resolve the identifiability issue, we present the calibration
problem from a geometric perspective of non-isometric curve to surface
matching, which enables us to take advantage of combinatorial optimization
techniques to extract necessary information for constructing prior
distributions. Our numerical experiments demonstrate that in terms of
prediction accuracy BNMC outperforms, or is comparable to, other existing
calibration frameworks.
| stat |
Deep Learning in current Neuroimaging: a multivariate approach with
power and type I error control but arguable generalization ability | Discriminative analysis in neuroimaging by means of deep/machine learning
techniques is usually tested with validation techniques, whereas the associated
statistical significance remains largely under-developed due to their
computational complexity. In this work, a non-parametric framework is proposed
that estimates the statistical significance of classifications using deep
learning architectures. In particular, a combination of autoencoders (AE) and
support vector machines (SVM) is applied to: (i) a one-condition, within-group
designs often of normal controls (NC) and; (ii) a two-condition, between-group
designs which contrast, for example, Alzheimer's disease (AD) patients with NC
(the extension to multi-class analyses is also included). A random-effects
inference based on a label permutation test is proposed in both studies using
cross-validation (CV) and resubstitution with upper bound correction (RUB) as
validation methods. This allows both false positives and classifier overfitting
to be detected as well as estimating the statistical power of the test. Several
experiments were carried out using the Alzheimer's Disease Neuroimaging
Initiative (ADNI) dataset, the Dominantly Inherited Alzheimer Network (DIAN)
dataset, and a MCI prediction dataset. We found in the permutation test that CV
and RUB methods offer a false positive rate close to the significance level and
an acceptable statistical power (although lower using cross-validation). A
large separation between training and test accuracies using CV was observed,
especially in one-condition designs. This implies a low generalization ability
as the model fitted in training is not informative with respect to the test
set. We propose as solution by applying RUB, whereby similar results are
obtained to those of the CV test set, but considering the whole set and with a
lower computational cost per iteration.
| stat |
Guided Visual Exploration of Relations in Data Sets | Efficient explorative data analysis systems must take into account both what
a user knows and wants to know. This paper proposes a principled framework for
interactive visual exploration of relations in data, through views most
informative given the user's current knowledge and objectives. The user can
input pre-existing knowledge of relations in the data and also formulate
specific exploration interests, then taken into account in the exploration. The
idea is to steer the exploration process towards the interests of the user,
instead of showing uninteresting or already known relations. The user's
knowledge is modelled by a distribution over data sets parametrised by subsets
of rows and columns of data, called tile constraints. We provide a
computationally efficient implementation of this concept based on constrained
randomisation. Furthermore, we describe a novel dimensionality reduction method
for finding the views most informative to the user, which at the limit of no
background knowledge and with generic objectives reduces to PCA. We show that
the method is suitable for interactive use and robust to noise, outperforms
standard projection pursuit visualisation methods, and gives understandable and
useful results in analysis of real-world data. We have released an open-source
implementation of the framework.
| stat |
Stochastic Mirror Descent for Low-Rank Tensor Decomposition Under
Non-Euclidean Losses | This work considers low-rank canonical polyadic decomposition (CPD) under a
class of non-Euclidean loss functions that frequently arise in statistical
machine learning and signal processing. These loss functions are often used for
certain types of tensor data, e.g., count and binary tensors, where the least
squares loss is considered unnatural.Compared to the least squares loss, the
non-Euclidean losses are generally more challenging to handle. Non-Euclidean
CPD has attracted considerable interests and a number of prior works exist.
However, pressing computational and theoretical challenges, such as scalability
and convergence issues, still remain. This work offers a unified stochastic
algorithmic framework for large-scale CPD decomposition under a variety of
non-Euclidean loss functions. Our key contribution lies in a tensor fiber
sampling strategy-based flexible stochastic mirror descent framework.
Leveraging the sampling scheme and the multilinear algebraic structure of
low-rank tensors, the proposed lightweight algorithm ensures global convergence
to a stationary point under reasonable conditions. Numerical results show that
our framework attains promising non-Euclidean CPD performance. The proposed
framework also exhibits substantial computational savings compared to
state-of-the-art methods.
| stat |
Augmented Probability Simulation Methods for Non-cooperative Games | We present a robust decision support framework with computational algorithms
for decision makers in non-cooperative sequential setups. Existing simulation
based approaches can be inefficient when there is a large number of feasible
decisions and uncertain outcomes. Hence, we provide a novel alternative to
solve non-cooperative sequential games based on augmented probability
simulation. We propose approaches to approximate subgame perfect equilibria
under complete information, assess the robustness of such solutions and,
finally, approximate adversarial risk analysis solutions when lacking complete
information. This framework could be especially beneficial in application
domains such as cybersecurity and counter-terrorism.
| stat |
A Bayesian latent allocation model for clustering compositional data
with application to the Great Barrier Reef | Relative abundance is a common metric to estimate the composition of species
in ecological surveys reflecting patterns of commonness and rarity of
biological assemblages. Measurements of coral reef compositions formed by four
communities along Australia's Great Barrier Reef (GBR) gathered between 2012
and 2017 are the focus of this paper. We undertake the task of finding clusters
of transect locations with similar community composition and investigate
changes in clustering dynamics over time. During these years, an unprecedented
sequence of extreme weather events (cyclones and coral bleaching) impacted the
58 surveyed locations. The dependence between constituent parts of a
composition presents a challenge for existing multivariate clustering
approaches. In this paper, we introduce a finite mixture of Dirichlet
distributions with group-specific parameters, where cluster memberships are
dictated by unobserved latent variables. The inference is carried in a Bayesian
framework, where MCMC strategies are outlined to sample from the posterior
model. Simulation studies are presented to illustrate the performance of the
model in a controlled setting. The application of the model to the 2012 coral
reef data reveals that clusters were spatially distributed in similar ways
across reefs which indicates a potential influence of wave exposure at the
origin of coral reef community composition. The number of clusters estimated by
the model decreased from four in 2012 to two from 2014 until 2017. Posterior
probabilities of transect allocations to the same cluster substantially
increase through time showing a potential homogenization of community
composition across the whole GBR. The Bayesian model highlights the diversity
of coral reef community composition within a coral reef and rapid changes
across large spatial scales that may contribute to undermining the future of
the GBR's biodiversity.
| stat |
A/B Testing in Dense Large-Scale Networks: Design and Inference | Design of experiments and estimation of treatment effects in large-scale
networks, in the presence of strong interference, is a challenging and
important problem. Most existing methods' performance deteriorates as the
density of the network increases. In this paper, we present a novel strategy
for accurately estimating the causal effects of a class of treatments in a
dense large-scale network. First, we design an approximate randomized
controlled experiment by solving an optimization problem to allocate treatments
in the presence of competition among neighboring nodes. Then we apply an
importance sampling adjustment to correct for any leftover bias (from the
approximation) in estimating average treatment effects. We provide theoretical
guarantees, verify robustness in a simulation study, and validate the
scalability and usefulness of our procedure in a real-world experiment on a
large social network.
| stat |
Double-matched matrix decomposition for multi-view data | We consider the problem of extracting joint and individual signals from
multi-view data, that is data collected from different sources on matched
samples. While existing methods for multi-view data decomposition explore
single matching of data by samples, we focus on double-matched multi-view data
(matched by both samples and source features). Our motivating example is the
miRNA data collected from both primary tumor and normal tissues of the same
subjects; the measurements from two tissues are thus matched both by subjects
and by miRNAs. Our proposed double-matched matrix decomposition allows to
simultaneously extract joint and individual signals across subjects, as well as
joint and individual signals across miRNAs. Our estimation approach takes
advantage of double-matching by formulating a new type of optimization problem
with explicit row space and column space constraints, for which we develop an
efficient iterative algorithm. Numerical studies indicate that taking advantage
of double-matching leads to superior signal estimation performance compared to
existing multi-view data decomposition based on single-matching. We apply our
method to miRNA data as well as data from the English Premier League soccer
matches, and find joint and individual multi-view signals that align with
domain specific knowledge.
| stat |
Optimal Statistical Inference for Individualized Treatment Effects in
High-dimensional Models | The ability to predict individualized treatment effects (ITEs) based on a
given patient's profile is essential for personalized medicine. We propose a
hypothesis testing approach to choosing between two potential treatments for a
given individual in the framework of high-dimensional linear models. The
methodological novelty lies in the construction of a debiased estimator of the
ITE and establishment of its asymptotic normality uniformly for an arbitrary
future high-dimensional observation, while the existing methods can only handle
certain specific forms of observations. We introduce a testing procedure with
the type-I error controlled and establish its asymptotic power. The proposed
method can be extended to making inference for general linear contrasts,
including both the average treatment effect and outcome prediction. We
introduce the optimality framework for hypothesis testing from both the
minimaxity and adaptivity perspectives and establish the optimality of the
proposed procedure. An extension to high-dimensional approximate linear models
is also considered. The finite sample performance of the procedure is
demonstrated in simulation studies and further illustrated through an analysis
of electronic health records data from patients with rheumatoid arthritis.
| stat |
Regularized K-means through hard-thresholding | We study a framework of regularized $K$-means methods based on direct
penalization of the size of the cluster centers. Different penalization
strategies are considered and compared through simulation and theoretical
analysis. Based on the results, we propose HT $K$-means, which uses an $\ell_0$
penalty to induce sparsity in the variables. Different techniques for selecting
the tuning parameter are discussed and compared. The proposed method stacks up
favorably with the most popular regularized $K$-means methods in an extensive
simulation study. Finally, HT $K$-means is applied to several real data
examples. Graphical displays are presented and used in these examples to gain
more insight into the datasets.
| stat |
Organic direct and indirect effects with post-treatment common causes of
mediator and outcome | Most of the literature on direct and indirect effects assumes that there are
no post-treatment common causes of the mediator and the outcome. In contrast to
natural direct and indirect effects, organic direct and indirect effects, which
were introduced in Lok (2016, 2020), can be extended to provide an
identification result for settings with post-treatment common causes of the
mediator and the outcome. This article provides a definition and an
identification result for organic direct and indirect effects in the presence
of post-treatment common causes of mediator and outcome. These new organic
indirect and direct effects have interpretations in terms of intervention
effects. Organic indirect effects in the presence of post-treatment common
causes are an addition to indirect effects through multivariate mediators.
Organic indirect effects in the presence of post-treatment common causes can be
used e.g. 1. to predict the effect of the initial treatment if its side affects
are suppressed through additional interventions or 2. to predict the effect of
a treatment that does not affect the post-treatment common cause and affects
the mediator the same way as the initial treatment.
| stat |
Maximum likelihood estimation of parameters of spherical particle size
distributions from profile size measurements and its application for small
samples | Microscopy research often requires recovering particle-size distributions in
three dimensions from only a few (10 - 200) profile measurements in the
section. This problem is especially relevant for petrographic and mineralogical
studies, where parametric assumptions are reasonable and finding distribution
parameters from the microscopic study of small sections is essential. This
paper deals with the specific case where particles are approximately spherical
(i.e. Wicksell's problem). The paper presents a novel approximation of the
probability density of spherical particle profile sizes. This approximation
uses the actual non-smoothness of mineral particles rather than perfect
spheres. The new approximation facilitates the numerically efficient use of the
maximum likelihood method, a generally powerful method that provides the
distribution parameter estimates of the minimal variance in most practical
cases. The variance and bias of the estimates by the maximum likelihood method
were compared numerically for several typical particle-size distributions with
those by alternative parametric methods (method of moments and minimum distance
estimation), and the maximum likelihood estimation was found to be preferable
for both small and large samples. The maximum likelihood method, along with the
suggested approximation, may also be used for selecting a model, for
constructing narrow confidence intervals for distribution parameters using all
the profiles without random sampling and for including the measurements of the
profiles intersected by section boundaries. The utility of the approach is
illustrated using an example from glacier ice petrography.
| stat |
On Focal Loss for Class-Posterior Probability Estimation: A Theoretical
Perspective | The focal loss has demonstrated its effectiveness in many real-world
applications such as object detection and image classification, but its
theoretical understanding has been limited so far. In this paper, we first
prove that the focal loss is classification-calibrated, i.e., its minimizer
surely yields the Bayes-optimal classifier and thus the use of the focal loss
in classification can be theoretically justified. However, we also prove a
negative fact that the focal loss is not strictly proper, i.e., the confidence
score of the classifier obtained by focal loss minimization does not match the
true class-posterior probability and thus it is not reliable as a
class-posterior probability estimator. To mitigate this problem, we next prove
that a particular closed-form transformation of the confidence score allows us
to recover the true class-posterior probability. Through experiments on
benchmark datasets, we demonstrate that our proposed transformation
significantly improves the accuracy of class-posterior probability estimation.
| stat |
Improve the Robustness and Accuracy of Deep Neural Network with
$L_{2,\infty}$ Normalization | In this paper, the robustness and accuracy of the deep neural network (DNN)
was enhanced by introducing the $L_{2,\infty}$ normalization of the weight
matrices of the DNN with Relu as the activation function. It is proved that the
$L_{2,\infty}$ normalization leads to large dihedral angles between two
adjacent faces of the polyhedron graph of the DNN function and hence smoother
DNN functions, which reduces over-fitting. A measure is proposed for the
robustness of a classification DNN, which is the average radius of the maximal
robust spheres with the sample data as centers. A lower bound for the
robustness measure is given in terms of the $L_{2,\infty}$ norm. Finally, an
upper bound for the Rademacher complexity of DNN with $L_{2,\infty}$
normalization is given. An algorithm is given to train a DNN with the
$L_{2,\infty}$ normalization and experimental results are used to show that the
$L_{2,\infty}$ normalization is effective to improve the robustness and
accuracy.
| stat |
Adaptive Quantile Computation for Brownian Bridge in Change-Point
Analysis | As an example for the fast calculation of distributional parameters of
Gaussian processes, we propose a new Monte Carlo algorithm for the computation
of quantiles of the supremum norm of weighted Brownian bridges. As it is known,
the corresponding distributions arise asymptotically for weighted CUSUM
statistics for change-point detection. The new algorithm employs an adaptive
(sequential) time discretization for the trajectories of the Brownian bridge. A
simulation study shows that the new algorithm by far outperforms the standard
approach, which employs a uniform time discretization.
| stat |
Active multi-fidelity Bayesian online changepoint detection | Online algorithms for detecting changepoints, or abrupt shifts in the
behavior of a time series, are often deployed with limited resources, e.g., to
edge computing settings such as mobile phones or industrial sensors. In these
scenarios it may be beneficial to trade the cost of collecting an environmental
measurement against the quality or "fidelity" of this measurement and how the
measurement affects changepoint estimation. For instance, one might decide
between inertial measurements or GPS to determine changepoints for motion. A
Bayesian approach to changepoint detection is particularly appealing because we
can represent our posterior uncertainty about changepoints and make active,
cost-sensitive decisions about data fidelity to reduce this posterior
uncertainty. Moreover, the total cost could be dramatically lowered through
active fidelity switching, while remaining robust to changes in data
distribution. We propose a multi-fidelity approach that makes cost-sensitive
decisions about which data fidelity to collect based on maximizing information
gain with respect to changepoints. We evaluate this framework on synthetic,
video, and audio data and show that this information-based approach results in
accurate predictions while reducing total cost.
| stat |
Quantifying replicability and consistency in systematic reviews | Systematic reviews of interventions are important tools for synthesizing
evidence from multiple studies. They serve to increase power and improve
precision, in the same way that larger studies can do, but also to establish
the consistency of effects and replicability of results across studies which
are not identical. In this work we suggest to incorporate replicability
analysis tools to quantify the consistency and conflict. These are offered both
for the fixed-effect and for the random-effects meta-analyses. We motivate and
demonstrate our approach and its implications by examples from systematic
reviews from the Cochrane library, and offer a way to incorporate our
suggestions in their standard reporting system.
| stat |
JointAI: Joint Analysis and Imputation of Incomplete Data in R | Missing data occur in many types of studies and typically complicate the
analysis. Multiple imputation, either using joint modelling or the more
flexible fully conditional specification approach, are popular and work well in
standard settings. In settings involving non-linear associations or
interactions, however, incompatibility of the imputation model with the
analysis model is an issue often resulting in bias. Similarly, complex outcomes
such as longitudinal or survival outcomes cannot be adequately handled by
standard implementations. In this paper, we introduce the R package JointAI,
which utilizes the Bayesian framework to perform simultaneous analysis and
imputation in regression models with incomplete covariates. Using a fully
Bayesian joint modelling approach it overcomes the issue of uncongeniality
while retaining the attractive flexibility of fully conditional specification
multiple imputation by specifying the joint distribution of analysis and
imputation models as a sequence of univariate models that can be adapted to the
type of variable. JointAI provides functions for Bayesian inference with
generalized linear and generalized linear mixed models and extensions thereof
as well as survival models and joint models for longitudinal and survival data,
that take arguments analogous to corresponding well known functions for the
analysis of complete data from base R and other packages. Usage and features of
JointAI are described and illustrated using various examples and the
theoretical background is outlined.
| stat |
Generalized Autoregressive Neural Network Models | A time series is a sequence of observations taken sequentially in time. The
autoregressive integrated moving average is a class of the model more used for
times series data. However, this class of model has two critical limitations.
It fits well onlyGaussian data with the linear structure of correlation. Here,
I present a new model named as generalized autoregressive neural networks,
GARNN. The GARNN is an extension of the generalized linear model where the mean
marginal depends on the lagged values via the inclusion of the neural network
in the link function. A practical application of the model is shown using a
well-known poliomyelitis case number, originated analyzed by Zeger and Qaqish
(1988),
| stat |
Generalized interpretation and identification of separable effects in
competing event settings | In competing event settings, a counterfactual contrast of cause-specific
cumulative incidences quantifies the total causal effect of a treatment on the
event of interest. However, effects of treatment on the competing event may
indirectly contribute to this total effect, complicating its interpretation. We
previously proposed the separable effects (Stensrud et al, 2019) to define
direct and indirect effects of the treatment on the event of interest. This
definition presupposes a treatment decomposition into two components acting
along two separate causal pathways, one exclusively outside of the competing
event and the other exclusively through it. Unlike previous definitions of
direct and indirect effects, the separable effects can be subject to empirical
scrutiny in a study where separate interventions on the treatment components
are available. Here we extend and generalize the notion of the separable
effects in several ways, allowing for interpretation, identification and
estimation under considerably weaker assumptions. We propose and discuss a
definition of separable effects that is applicable to general time-varying
structures, where the separable effects can still be meaningfully interpreted,
even when they cannot be regarded as direct and indirect effects. We further
derive weaker conditions for identification of separable effects in
observational studies where decomposed treatments are not yet available; in
particular, these conditions allow for time-varying common causes of the event
of interest, the competing events and loss to follow-up. For these general
settings, we propose semi-parametric weighted estimators that are
straightforward to implement. As an illustration, we apply the estimators to
study the separable effects of intensive blood pressure therapy on acute kidney
injury, using data from a randomized clinical trial.
| stat |
Spatio-temporal methods for estimating subsurface ocean thermal response
to tropical cyclones | Prevailing theory contends that tropical cyclones (TCs) are primarily driven
by wind-induced heat exchange between the air and sea. Accurate
characterization of the subsurface ocean thermal response to TC passage is
crucial for accurate TC intensity forecasts and climatological modeling, yet
that characterization is complicated by the high-noise ocean environment,
correlations inherent in spatio-temporal data, relative scarcity of in situ
observations, and entanglement of the TC-induced signal with seasonal signals.
We present a general methodological framework that addresses these
difficulties, integrating existing techniques in seasonal mean field
estimation, Gaussian process modeling, and nonparametric regression into a
functional ANOVA model. Importantly, we improve upon past work by properly
handling seasonality, providing rigorous uncertainty quantification, and
treating time as a continuous variable, rather than producing estimates that
are binned in time. This functional ANOVA model is estimated using in situ
subsurface temperature profiles from the Argo fleet of autonomous floats
through a multi-step procedure, which (1) characterizes the upper ocean
seasonal shift during the TC season; (2) models the variability in the
temperature observations; (3) fits a thin plate spline using the variability
estimates to account for heteroskedasticity and correlation between the
observations. This spline fit reveals the ocean thermal response to TC passage.
Through this framework, we obtain new scientific insights into the interaction
between TCs and the ocean on a global scale, including a three-dimensional
characterization of the near-surface and subsurface cooling along the TC storm
track and the mixing-induced subsurface warming on the track's right side.
| stat |
Multi-task Learning with High-Dimensional Noisy Images | Recent medical imaging studies have given rise to distinct but inter-related
datasets corresponding to multiple experimental tasks or longitudinal visits.
Standard scalar-on-image regression models that fit each dataset separately are
not equipped to leverage information across inter-related images, and existing
multi-task learning approaches are compromised by the inability to account for
the noise that is often observed in images. We propose a novel joint
scalar-on-image regression framework involving wavelet-based image
representations with grouped penalties that are designed to pool information
across inter-related images for joint learning, and which explicitly accounts
for noise in high-dimensional images via a projection-based approach. In the
presence of non-convexity arising due to noisy images, we derive non-asymptotic
error bounds under non-convex as well as convex grouped penalties, even when
the number of voxels increases exponentially with sample size. A projected
gradient descent algorithm is used for computation, which is shown to
approximate the optimal solution via well-defined non-asymptotic optimization
error bounds under noisy images. Extensive simulations and application to a
motivating longitudinal Alzheimer's disease study illustrate significantly
improved predictive ability and greater power to detect true signals, that are
simply missed by existing methods without noise correction due to the
attenuation to null phenomenon.
| stat |
The MELODIC family for simultaneous binary logistic regression in a
reduced space | Logistic regression is a commonly used method for binary classification.
Researchers often have more than a single binary response variable and
simultaneous analysis is beneficial because it provides insight into the
dependencies among response variables as well as between the predictor
variables and the responses. Moreover, in such a simultaneous analysis the
equations can lend each other strength, which might increase predictive
accuracy. In this paper, we propose the MELODIC family for simultaneous binary
logistic regression modeling. In this family, the regression models are defined
in a Euclidean space of reduced dimension, based on a distance rule. The model
may be interpreted in terms of logistic regression coefficients or in terms of
a biplot. We discuss a fast iterative majorization (or MM) algorithm for
parameter estimation. Two applications are shown in detail: one relating
personality characteristics to drug consumption profiles and one relating
personality characteristics to depressive and anxiety disorders. We present a
thorough comparison of our MELODIC family with alternative approaches for
multivariate binary data.
| stat |
Reparametrization Invariance in non-parametric Causal Discovery | Causal discovery estimates the underlying physical process that generates the
observed data: does X cause Y or does Y cause X? Current methodologies use
structural conditions to turn the causal query into a statistical query, when
only observational data is available. But what if these statistical queries are
sensitive to causal invariants? This study investigates one such invariant: the
causal relationship between X and Y is invariant to the marginal distributions
of X and Y. We propose an algorithm that uses a non-parametric estimator that
is robust to changes in the marginal distributions. This way we may marginalize
the marginals, and inspect what relationship is intrinsically there. The
resulting causal estimator is competitive with current methodologies and has
high emphasis on the uncertainty in the causal query; an aspect just as
important as the query itself.
| stat |
Uncertainty Quantification with Generative Models | We develop a generative model-based approach to Bayesian inverse problems,
such as image reconstruction from noisy and incomplete images. Our framework
addresses two common challenges of Bayesian reconstructions: 1) It makes use of
complex, data-driven priors that comprise all available information about the
uncorrupted data distribution. 2) It enables computationally tractable
uncertainty quantification in the form of posterior analysis in latent and data
space. The method is very efficient in that the generative model only has to be
trained once on an uncorrupted data set, after that, the procedure can be used
for arbitrary corruption types.
| stat |
An Upper Bound of the Bias of Nadaraya-Watson Kernel Regression under
Lipschitz Assumptions | The Nadaraya-Watson kernel estimator is among the most popular nonparameteric
regression technique thanks to its simplicity. Its asymptotic bias has been
studied by Rosenblatt in 1969 and has been reported in a number of related
literature. However, Rosenblatt's analysis is only valid for infinitesimal
bandwidth. In contrast, we propose in this paper an upper bound of the bias
which holds for finite bandwidths. Moreover, contrarily to the classic analysis
we allow for discontinuous first order derivative of the regression function,
we extend our bounds for multidimensional domains and we include the knowledge
of the bound of the regression function when it exists and if it is known, to
obtain a tighter bound. We believe that this work has potential applications in
those fields where some hard guarantees on the error are needed
| stat |
Joint Stochastic Modeling and Statistical Analysis of Recurrent
Competing Risks, Longitudinal Marker, and Health Status | Consider a subject or unit being monitored over a period of random duration
in a longitudinal time-to-event study in a biomedical, public health, or
engineering setting. As time moves forward, this unit experiences recurrent
events of several types and a longitudinal marker transitions over a discrete
state-space. In addition, its "health" status also transitions over a discrete
state-space containing at least one absorbing state. A vector of covariates
will also be associated with this unit. Of major interest for this unit is the
time-to-absorption of its health status process, which represents this unit's
lifetime. Aside from being affected by the covariate vector, there is a synergy
among the recurrent competing risks processes, the longitudinal marker process,
and the health status process in the sense that the time-evolution of each
process is affected by the other processes. To exploit this synergy in order to
obtain more realistic models and enhance inferential performance, a joint
stochastic model for these components is proposed and the proper statistical
inference methods for this model are developed. This joint model has the
potential of facilitating precision interventions, thereby enhancing precision
or personalized medicine. A stochastic process approach, using counting
processes and continuous-time Markov chains, is utilized, which allows for
modeling the dynamicity arising from the synergy among the model components and
the impact of performed interventions after event occurrences and the
increasing number of event occurrences. Likelihood-based inferential methods
are developed based on observing a sample of these units. Properties of the
inferential procedures are examined through simulations and illustrated using
some real data sets.
| stat |
On a Multi-Year Microlevel Collective Risk Model | For a typical insurance portfolio, the claims process for a short period,
typically one year, is characterized by observing frequency of claims together
with the associated claims severities. The collective risk model describes this
portfolio as a random sum of the aggregation of the claim amounts. In the
classical framework, for simplicity, the claim frequency and claim severities
are assumed to be mutually independent. However, there is a growing interest in
relaxing this independence assumption which is more realistic and useful for
the practical insurance ratemaking. While the common thread has been capturing
the dependence between frequency and aggregate severity within a single period,
the work of Oh et al. (2020a) provides an interesting extension to the addition
of capturing dependence among individual severities. In this paper, we extend
these works within a framework where we have a portfolio of microlevel
frequencies and severities for multiple years. This allows us to develop a
factor copula model framework that captures various types of dependence between
claim frequencies and claim severities over multiple years. It is therefore a
clear extension of earlier works on one-year dependent frequency-severity
models and on random effects model for capturing serial dependence of claims.
We focus on the results using a family of elliptical copulas to model the
dependence. The paper further describes how to calibrate the proposed model
using illustrative claims data arising from a Singapore insurance company. The
estimated results provide strong evidence of all forms of dependencies captured
by our model.
| stat |
Link prediction in dynamic networks using random dot product graphs | The problem of predicting links in large networks is an important task in a
variety of practical applications, including social sciences, biology and
computer security. In this paper, statistical techniques for link prediction
based on the popular random dot product graph model are carefully presented,
analysed and extended to dynamic settings. Motivated by a practical application
in cyber-security, this paper demonstrates that random dot product graphs not
only represent a powerful tool for inferring differences between multiple
networks, but are also efficient for prediction purposes and for understanding
the temporal evolution of the network. The probabilities of links are obtained
by fusing information at two stages: spectral methods provide estimates of
latent positions for each node, and time series models are used to capture
temporal dynamics. In this way, traditional link prediction methods, usually
based on decompositions of the entire network adjacency matrix, are extended
using temporal information. The methods presented in this article are applied
to a number of simulated and real-world computer network graphs, showing
promising results.
| stat |
Diffusion Means and Heat Kernel on Manifolds | We introduce diffusion means as location statistics on manifold data spaces.
A diffusion mean is defined as the starting point of an isotropic diffusion
with a given diffusivity. They can therefore be defined on all spaces on which
a Brownian motion can be defined and numerical calculation of sample diffusion
means is possible on a variety of spaces using the heat kernel expansion. We
present several classes of spaces, for which the heat kernel is known and
sample diffusion means can therefore be calculated. As an example, we
investigate a classic data set from directional statistics, for which the
sample Fr\'echet mean exhibits finite sample smeariness.
| stat |
A Non-Asymptotic Analysis for Stein Variational Gradient Descent | We study the Stein Variational Gradient Descent (SVGD) algorithm, which
optimises a set of particles to approximate a target probability distribution
$\pi\propto e^{-V}$ on $\mathbb{R}^d$. In the population limit, SVGD performs
gradient descent in the space of probability distributions on the KL divergence
with respect to $\pi$, where the gradient is smoothed through a kernel integral
operator. In this paper, we provide a novel finite time analysis for the SVGD
algorithm. We provide a descent lemma establishing that the algorithm decreases
the objective at each iteration, and rates of convergence for the average Stein
Fisher divergence (also referred to as Kernel Stein Discrepancy). We also
provide a convergence result of the finite particle system corresponding to the
practical implementation of SVGD to its population version.
| stat |
Stochastic Neighbor Embedding with Gaussian and Student-t Distributions:
Tutorial and Survey | Stochastic Neighbor Embedding (SNE) is a manifold learning and dimensionality
reduction method with a probabilistic approach. In SNE, every point is consider
to be the neighbor of all other points with some probability and this
probability is tried to be preserved in the embedding space. SNE considers
Gaussian distribution for the probability in both the input and embedding
spaces. However, t-SNE uses the Student-t and Gaussian distributions in these
spaces, respectively. In this tutorial and survey paper, we explain SNE,
symmetric SNE, t-SNE (or Cauchy-SNE), and t-SNE with general degrees of
freedom. We also cover the out-of-sample extension and acceleration for these
methods. Some simulations to visualize the embeddings are also provided.
| stat |
Interpretation of point forecasts with unkown directive | Point forecasts can be interpreted as functionals (i.e., point summaries) of
predictive distributions. We consider the situation where forecasters'
directives are hidden and develop methodology for the identification of the
unknown functional based on time series data of point forecasts and associated
realizations. Focusing on the natural cases of state-dependent quantiles and
expectiles, we provide a generalized method of moments estimator for the
functional, along with tests of optimality relative to information sets that
are specified by instrumental variables. Using simulation, we demonstrate that
our optimality test is better calibrated and more powerful than existing
solutions. In empirical examples, Greenbook gross domestic product (GDP)
forecasts of the US Federal Reserve and model output for precipitation from the
European Centre for Medium-Range Weather Forecasts (ECMWF) are indicative of
overstatement in anticipation of extreme events.
| stat |
Hierarchical approaches for flexible and interpretable binary regression
models | Binary regression models are ubiquitous in virtually every scientific field.
Frequently, traditional generalized linear models fail to capture the
variability in the probability surface that gives rise to the binary
observations and novel methodology is required. This has generated a
substantial literature comprised of binary regression models motivated by
various applications. We describe a novel organization of generalizations to
traditional binary regression methods based on the familiar three-part
structure of generalized linear models (random component, systematic component,
link function). This new perspective facilitates both the comparison of
existing approaches, and the development of novel, flexible models with
interpretable parameters that capture application-specific data generating
mechanisms. We use our proposed organizational structure to discuss some
concerns with certain existing models for binary data based on quantile
regression. We then use the framework to develop several new binary regression
models tailored to occupancy data for European red squirrels (Sciurus
vulgaris).
| stat |
An Adapted Geographically Weighted Lasso(Ada-GWL) model for estimating
metro ridership | Ridership estimation at station level plays a critical role in metro
transportation planning. Among various existing ridership estimation methods,
direct demand model has been recognized as an effective approach. However,
existing direct demand models including Geographically Weighted Regression
(GWR) have rarely included local model selection in ridership estimation. In
practice, acquiring insights into metro ridership under multiple influencing
factors from a local perspective is important for passenger volume management
and transportation planning operations adapting to local conditions. In this
study, we propose an Adapted Geographically Weighted Lasso (Ada-GWL) framework
for modelling metro ridership, which performs regression-coefficient shrinkage
and local model selection. It takes metro network connection intermedia into
account and adopts network-based distance metric instead of Euclidean-based
distance metric, making it so-called adapted to the context of metro networks.
The real-world case of Shenzhen Metro is used to validate the superiority of
our proposed model. The results show that the Ada-GWL model performs the best
compared with the global model (Ordinary Least Square (OLS), GWR, GWR
calibrated with network-based distance metric and GWL in terms of estimation
error of the dependent variable and goodness-of-fit. Through understanding the
variation of each coefficient across space (elasticities) and variables
selection of each station, it provides more realistic conclusions based on
local analysis. Besides, through clustering analysis of the stations according
to the regression coefficients, clusters' functional characteristics are found
to be in compliance with the facts of the functional land use policy of
Shenzhen. These results of the proposed Ada-GWL model demonstrate a great
spatial explanatory power in transportation planning.
| stat |
Crop yield probability density forecasting via quantile random forest
and Epanechnikov Kernel function | A reliable and accurate forecasting model for crop yields is of crucial
importance for efficient decision-making process in the agricultural sector.
However, due to weather extremes and uncertainties, most forecasting models for
crop yield are not reliable and accurate. For measuring the uncertainty and
obtaining further information of future crop yields, a probability density
forecasting model based on quantile random forest and Epanechnikov kernel
function (QRF-SJ) is proposed. The nonlinear structure of random forest is
applied to change the quantile regression model for building the probabilistic
forecasting model. Epanechnikov kernel function and solve-the equation plug-in
approach of Sheather and Jones are used in the kernel density estimation. A
case study using the annual crop yield of groundnut and millet in Ghana is
presented to illustrate the efficiency and robustness of the proposed
technique. The values of the prediction interval coverage probability and
prediction interval normalized average width for the two crops show that the
constructed prediction intervals capture the observed yields with high coverage
probability. The probability density curves show that QRF-SJ method has a very
high ability to forecast quality prediction intervals with a higher coverage
probability. The feature importance gave a score of the importance of each
weather variable in building the quantile regression forest model. The farmer
and other stakeholders are able to realize the specific weather variable that
affect the yield of a selected crop through feature importance. The proposed
method and its application on crop yield dataset are the first of its kind in
literature.
| stat |
Diffeomorphic Learning | We introduce in this paper a learning paradigm in which the training data is
transformed by a diffeomorphic transformation before prediction. The learning
algorithm minimizes a cost function evaluating the prediction error on the
training set penalized by the distance between the diffeomorphism and the
identity. The approach borrows ideas from shape analysis where diffeomorphisms
are estimated for shape and image alignment, and brings them in a previously
unexplored setting, estimating, in particular diffeomorphisms in much larger
dimensions. After introducing the concept and describing a learning algorithm,
we present diverse applications, mostly with synthetic examples, demonstrating
the potential of the approach, as well as some insight on how it can be
improved.
| stat |
Localization processes for functional data analysis | We propose an alternative to k-nearest neighbors for functional data whereby
the approximating neighbor curves are piecewise functions built from a
functional sample. Instead of a distance on a function space we use a locally
defined distance function that satisfies stabilization criteria. We exploit
this feature to develop the asymptotic theory when the number of curves is
large enough or when a finite number of curves is observed at time-points
coinciding with the realization of a point process with intensity increasing to
infinity. We use these results to investigate the problem of estimating
unobserved segments of a partially observed functional data sample as well as
to study the problem of functional classification and outlier detection. For
these problems, we discuss methods that are competitive with and often superior
to benchmark predictions in the field.
| stat |
Numerical Gaussian process Kalman filtering | In this manuscript we introduce numerical Gaussian process Kalman filtering
(GPKF). Numerical Gaussian processes have recently been developed to simulate
spatiotemporal models. The contribution of this paper is to embed numerical
Gaussian processes into the recursive Kalman filter equations. This embedding
enables us to do Kalman filtering on infinite-dimensional systems using
Gaussian processes. This is possible because i) we are obtaining a linear model
from numerical Gaussian processes, and ii) the states of this model are by
definition Gaussian distributed random variables. Convenient properties of the
numerical GPKF are that no spatial discretization of the model is necessary,
and manual setting up of the Kalman filter, that is fine-tuning the process and
measurement noise levels by hand is not required, as they are learned online
from the data stream. We showcase the capability of the numerical GPKF in a
simulation study of the advection equation.
| stat |
Distributional Anchor Regression | Prediction models often fail if train and test data do not stem from the same
distribution. Out-of-distribution (OOD) generalization to unseen, perturbed
test data is a desirable but difficult-to-achieve property for prediction
models and in general requires strong assumptions on the data generating
process (DGP). In a causally inspired perspective on OOD generalization, the
test data arise from a specific class of interventions on exogenous random
variables of the DGP, called anchors. Anchor regression models, introduced by
Rothenh\"ausler et al. (2018), protect against distributional shifts in the
test data by employing causal regularization. However, so far anchor regression
has only been used with a squared-error loss which is inapplicable to common
responses such as censored continuous or ordinal data. Here, we propose a
distributional version of anchor regression which generalizes the method to
potentially censored responses with at least an ordered sample space. To this
end, we combine a flexible class of parametric transformation models for
distributional regression with an appropriate causal regularizer under a more
general notion of residuals. In an exemplary application and several simulation
scenarios we demonstrate the extent to which OOD generalization is possible.
| stat |
Leave-one-out Approach for Matrix Completion: Primal and Dual Analysis | In this paper, we introduce a powerful technique based on Leave-one-out
analysis to the study of low-rank matrix completion problems. Using this
technique, we develop a general approach for obtaining fine-grained, entrywise
bounds for iterative stochastic procedures in the presence of probabilistic
dependency. We demonstrate the power of this approach in analyzing two of the
most important algorithms for matrix completion: (i) the non-convex approach
based on Projected Gradient Descent (PGD) for a rank-constrained formulation,
also known as the Singular Value Projection algorithm, and (ii) the convex
relaxation approach based on nuclear norm minimization (NNM).
Using this approach, we establish the first convergence guarantee for the
original form of PGD without regularization or sample splitting}, and in
particular shows that it converges linearly in the infinity norm. For NNM, we
use this approach to study a fictitious iterative procedure that arises in the
dual analysis. Our results show that \NNM recovers an $ d $-by-$ d $ rank-$ r $
matrix with $\mathcal{O}(\mu r \log(\mu r) d \log d )$ observed entries. This
bound has optimal dependence on the matrix dimension and is independent of the
condition number. To the best of our knowledge, this is the first sample
complexity result for a tractable matrix completion algorithm that satisfies
these two properties simultaneously.
| stat |
Nested conformal prediction and quantile out-of-bag ensemble methods | Conformal prediction is a popular tool for providing valid prediction sets
for classification and regression problems, without relying on any
distributional assumptions on the data. While the traditional description of
conformal prediction starts with a nonconformity score, we provide an alternate
(but equivalent) view that starts with a sequence of nested sets and calibrates
them to find a valid prediction set. The nested framework subsumes all
nonconformity scores, including recent proposals based on quantile regression
and density estimation. While these ideas were originally derived based on
sample splitting, our framework seamlessly extends them to other aggregation
schemes like cross-conformal, jackknife+ and out-of-bag methods. We use the
framework to derive a new algorithm (QOOB, pronounced cube) that combines four
ideas: quantile regression, cross-conformalization, ensemble methods and
out-of-bag predictions. We develop a computationally efficient implementation
of cross-conformal, that is also used by QOOB. In a detailed numerical
investigation, QOOB performs either the best or close to the best on all
simulated and real datasets.
| stat |
Sensitivity Prewarping for Local Surrogate Modeling | In the continual effort to improve product quality and decrease operations
costs, computational modeling is increasingly being deployed to determine
feasibility of product designs or configurations. Surrogate modeling of these
computer experiments via local models, which induce sparsity by only
considering short range interactions, can tackle huge analyses of complicated
input-output relationships. However, narrowing focus to local scale means that
global trends must be re-learned over and over again. In this article, we
propose a framework for incorporating information from a global sensitivity
analysis into the surrogate model as an input rotation and rescaling
preprocessing step. We discuss the relationship between several sensitivity
analysis methods based on kernel regression before describing how they give
rise to a transformation of the input variables. Specifically, we perform an
input warping such that the "warped simulator" is equally sensitive to all
input directions, freeing local models to focus on local dynamics. Numerical
experiments on observational data and benchmark test functions, including a
high-dimensional computer simulator from the automotive industry, provide
empirical validation.
| stat |
Randomization Tests for Weak Null Hypotheses in Randomized Experiments | The Fisher randomization test (FRT) is appropriate for any test statistic,
under a sharp null hypothesis that can recover all missing potential outcomes.
However, it is often sought after to test a weak null hypothesis that the
treatment does not affect the units on average. To use the FRT for a weak null
hypothesis, we must address two issues. First, we need to impute the missing
potential outcomes although the weak null hypothesis cannot determine all of
them. Second, we need to choose a proper test statistic. For a general weak
null hypothesis, we propose an approach to imputing missing potential outcomes
under a compatible sharp null hypothesis. Building on this imputation scheme,
we advocate a studentized statistic. The resulting FRT has multiple desirable
features. First, it is model-free. Second, it is finite-sample exact under the
sharp null hypothesis that we use to impute the potential outcomes. Third, it
conservatively controls large-sample type I errors under the weak null
hypothesis of interest. Therefore, our FRT is agnostic to the treatment effect
heterogeneity. We establish a unified theory for general factorial experiments.
We also extend it to stratified and clustered experiments.
| stat |
Large scale Lasso with windowed active set for convolutional spike
sorting | Spike sorting is a fundamental preprocessing step in neuroscience that is
central to access simultaneous but distinct neuronal activities and therefore
to better understand the animal or even human brain. But numerical complexity
limits studies that require processing large scale datasets in terms of number
of electrodes, neurons, spikes and length of the recorded signals. We propose
in this work a novel active set algorithm aimed at solving the Lasso for a
classical convolutional model. Our algorithm can be implemented efficiently on
parallel architecture and has a linear complexity w.r.t. the temporal
dimensionality which ensures scaling and will open the door to online spike
sorting. We provide theoretical results about the complexity of the algorithm
and illustrate it in numerical experiments along with results about the
accuracy of the spike recovery and robustness to the regularization parameter.
| stat |
Estimating densities with nonlinear support using Fisher-Gaussian
kernels | Current tools for multivariate density estimation struggle when the density
is concentrated near a nonlinear subspace or manifold. Most approaches require
choice of a kernel, with the multivariate Gaussian by far the most commonly
used. Although heavy-tailed and skewed extensions have been proposed, such
kernels cannot capture curvature in the support of the data. This leads to poor
performance unless the sample size is very large relative to the dimension of
the data. This article proposes a novel generalization of the Gaussian
distribution, which includes an additional curvature parameter. We refer to the
proposed class as Fisher-Gaussian (FG) kernels, since they arise by sampling
from a von Mises-Fisher density on the sphere and adding Gaussian noise. The FG
density has an analytic form, and is amenable to straightforward implementation
within Bayesian mixture models using Markov chain Monte Carlo. We provide
theory on large support, and illustrate gains relative to competitors in
simulated and real data applications.
| stat |
Sampling of Bayesian posteriors with a non-Gaussian probabilistic
learning on manifolds from a small dataset | This paper tackles the challenge presented by small-data to the task of
Bayesian inference. A novel methodology, based on manifold learning and
manifold sampling, is proposed for solving this computational statistics
problem under the following assumptions: 1) neither the prior model nor the
likelihood function are Gaussian and neither can be approximated by a Gaussian
measure; 2) the number of functional input (system parameters) and functional
output (quantity of interest) can be large; 3) the number of available
realizations of the prior model is small, leading to the small-data challenge
typically associated with expensive numerical simulations; the number of
experimental realizations is also small; 4) the number of the posterior
realizations required for decision is much larger than the available initial
dataset. The method and its mathematical aspects are detailed. Three
applications are presented for validation: The first two involve mathematical
constructions aimed to develop intuition around the method and to explore its
performance. The third example aims to demonstrate the operational value of the
method using a more complex application related to the statistical inverse
identification of the non-Gaussian matrix-valued random elasticity field of a
damaged biological tissue (osteoporosis in a cortical bone) using ultrasonic
waves.
| stat |
Adaptive Sparse Estimation with Side Information | The article considers the problem of estimating a high-dimensional sparse
parameter in the presence of side information that encodes the sparsity
structure. We develop a general framework that involves first using an
auxiliary sequence to capture the side information, and then incorporating the
auxiliary sequence in inference to reduce the estimation risk. The proposed
method, which carries out adaptive SURE-thresholding using side information
(ASUS), is shown to have robust performance and enjoy optimality properties. We
develop new theories to characterize regimes in which ASUS far outperforms
competitive shrinkage estimators, and establish precise conditions under which
ASUS is asymptotically optimal. Simulation studies are conducted to show that
ASUS substantially improves the performance of existing methods in many
settings. The methodology is applied for analysis of data from single cell
virology studies and microarray time course experiments.
| stat |
Testing new property of elliptical model for stock returns distribution | Wide class of elliptically contoured distributions is a popular model of
stock returns distribution. However the important question of adequacy of the
model is open. There are some results which reject and approve such model. Such
results are obtained by testing some properties of elliptical model for each
pair of stocks from some markets. New property of equality of $\tau$ Kendall
correlation coefficient and probability of sign coincidence for any pair of
random variables with elliptically contoured distribution is proved in the
paper. Distribution free statistical tests for testing this property for any
pair of stocks are constructed. Holm multiple hypotheses testing procedure
based on the individual tests is constructed and applied for stock markets data
for the concrete year. New procedure of testing the elliptical model for stock
returns distribution for all years of observation for some period is proposed.
The procedure is applied for the stock markets data of China, USA, Great
Britain and Germany for the period from 2003 to 2014. It is shown that for USA,
Great Britain and Germany stock markets the hypothesis of elliptical model of
stock returns distribution could be accepted but for Chinese stock market is
rejected for some cases.
| stat |
Information-Theoretic Approximation to Causal Models | Inferring the causal direction and causal effect between two discrete random
variables X and Y from a finite sample is often a crucial problem and a
challenging task. However, if we have access to observational and
interventional data, it is possible to solve that task. If X is causing Y, then
it does not matter if we observe an effect in Y by observing changes in X or by
intervening actively on X. This invariance principle creates a link between
observational and interventional distributions in a higher dimensional
probability space. We embed distributions that originate from samples of X and
Y into that higher dimensional space such that the embedded distribution is
closest to the distributions that follow the invariance principle, with respect
to the relative entropy. This allows us to calculate the best
information-theoretic approximation for a given empirical distribution, that
follows an assumed underlying causal model. We show that this
information-theoretic approximation to causal models (IACM) can be done by
solving a linear optimization problem. In particular, by approximating the
empirical distribution to a monotonic causal model, we can calculate
probabilities of causation. We can also use IACM for causal discovery problems
in the bivariate, discrete case. However, experimental results on labeled
synthetic data from additive noise models show that our causal discovery
approach is lagging behind state-of-the-art approaches because the invariance
principle encodes only a necessary condition for causal relations.
Nevertheless, for synthetic multiplicative noise data and real-world data, our
approach can compete in some cases with alternative methods.
| stat |
MetFlow: A New Efficient Method for Bridging the Gap between Markov
Chain Monte Carlo and Variational Inference | In this contribution, we propose a new computationally efficient method to
combine Variational Inference (VI) with Markov Chain Monte Carlo (MCMC). This
approach can be used with generic MCMC kernels, but is especially well suited
to \textit{MetFlow}, a novel family of MCMC algorithms we introduce, in which
proposals are obtained using Normalizing Flows. The marginal distribution
produced by such MCMC algorithms is a mixture of flow-based distributions, thus
drastically increasing the expressivity of the variational family. Unlike
previous methods following this direction, our approach is amenable to the
reparametrization trick and does not rely on computationally expensive reverse
kernels. Extensive numerical experiments show clear computational and
performance improvements over state-of-the-art methods.
| stat |
Learning Nonlinear Brain Dynamics: van der Pol Meets LSTM | Many real-world data sets, especially in biology, are produced by complex
nonlinear dynamical systems. In this paper, we focus on brain calcium imaging
(CaI) of different organisms (zebrafish and rat), aiming to build a model of
joint activation dynamics in large neuronal populations, including the whole
brain of zebrafish. We propose a new approach for capturing dynamics of
temporal SVD components that uses the coupled (multivariate) van der Pol (VDP)
oscillator, a nonlinear ordinary differential equation (ODE) model describing
neural activity, with a new parameter estimation technique that combines
variable projection optimization and stochastic search. We show that the
approach successfully handles nonlinearities and hidden state variables in the
coupled VDP. The approach is accurate, achieving 0.82 to 0.94 correlation
between the actual and model-generated components, and interpretable, as VDP's
coupling matrix reveals anatomically meaningful positive (excitatory) and
negative (inhibitory) interactions across different brain subsystems
corresponding to spatial SVD components. Moreover, VDP is comparable to (or
sometimes better than) recurrent neural networks (LSTM) for (short-term)
prediction of future brain activity; VDP needs less parameters to train, which
was a plus on our small training data. Finally, the overall best predictive
method, greatly outperforming both VDP and LSTM in short- and long-term
predictive settings on both datasets, was the new hybrid VDP-LSTM approach that
used VDP to simulate large domain-specific dataset for LSTM pretraining; note
that simple LSTM data-augmentation via noisy versions of training data was much
less effective.
| stat |
Robust statistical modeling of monthly rainfall: The minimum density
power divergence approach | Statistical modeling of rainfall is an important challenge in meteorology,
particularly from the perspective of rainfed agriculture where a proper
assessment of the future availability of rainwater is necessary. The
probability models mostly used for this purpose are exponential, gamma, Weibull
and lognormal distributions, where the unknown model parameters are routinely
estimated using the maximum likelihood estimator (MLE). However, presence of
outliers or extreme observations is quite common in rainfall data and the MLEs
being highly sensitive to them often leads to spurious inference. In this
paper, we discuss a robust parameter estimation approach based on the minimum
density power divergence estimators (MDPDEs) which provides a class of
estimates through a tuning parameter including the MLE as a special case. The
underlying tuning parameter controls the trade-offs between efficiency and
robustness of the resulting inference; we also discuss a procedure for
data-driven optimal selection of this tuning parameter as well as robust
selection of an appropriate model that provides best fit to some specific
rainfall data. We fit the above four parametric models to the areally-weighted
monthly rainfall data from the 36 meteorological subdivisions of India for the
years 1951-2014 and compare the fits based on the MLE and the proposed optimum
MDPDE; the superior performances of the MDPDE based approach are illustrated
for several cases. For all month-subdivision combinations, the best-fit models
and the estimated median rainfall amounts are provided. Software (written in R)
for calculating MDPDEs and their standard errors, optimal tuning parameter
selection and model selection are also provided.
| stat |
Latent Theme Dictionary Model for Finding Co-occurrent Patterns in
Process Data | Process data, temporally ordered categorical observations, are of recent
interest due to its increasing abundance and the desire to extract useful
information. A process is a collection of time-stamped events of different
types, recording how an individual behaves in a given time period. The process
data are too complex in terms of size and irregularity for the classical
psychometric models to be applicable, at least directly, and, consequently, it
is desirable to develop new ways for modeling and analysis. We introduce herein
a latent theme dictionary model (LTDM) for processes that identifies
co-occurrent event patterns and individuals with similar behavioral patterns.
Theoretical properties are established under certain regularity conditions for
the likelihood based estimation and inference. A non-parametric Bayes LTDM
algorithm using the Markov Chain Monte Carlo method is proposed for
computation. Simulation studies show that the proposed approach performs well
in a range of situations. The proposed method is applied to an item in the 2012
Programme for International Student Assessment with interpretable findings.
| stat |
Probabilistic learning of boolean functions applied to the binary
classification problem with categorical covariates | In this work we cast the problem of binary classification in terms of
estimating a partition on Bernoulli data. When the explanatory variables are
all categorical, the problem can be modelled using the language of boolean
functions. We offer a probabilistic analysis of the problem, and propose two
algorithms for learning boolean functions from binary data.
| stat |
Prediction of Hilbertian autoregressive processes : a Recurrent Neural
Network approach | The autoregressive Hilbertian model (ARH) was introduced in the early 90's by
Denis Bosq. It was the subject of a vast literature and gave birth to numerous
extensions. The model generalizes the classical multidimensional autoregressive
model, widely used in Time Series Analysis. It was successfully applied in
numerous fields such as finance, industry, biology. We propose here to compare
the classical prediction methodology based on the estimation of the
autocorrelation operator with a neural network learning approach. The latter is
based on a popular version of Recurrent Neural Networks : the Long Short Term
Memory networks. The comparison is carried out through simulations and real
datasets.
| stat |
Low-rank multi-parametric covariance identification | We propose a differential geometric construction for families of low-rank
covariance matrices, via interpolation on low-rank matrix manifolds. In
contrast with standard parametric covariance classes, these families offer
significant flexibility for problem-specific tailoring via the choice of
"anchor" matrices for the interpolation. Moreover, their low-rank facilitates
computational tractability in high dimensions and with limited data. We employ
these covariance families for both interpolation and identification, where the
latter problem comprises selecting the most representative member of the
covariance family given a data set. In this setting, standard procedures such
as maximum likelihood estimation are nontrivial because the covariance family
is rank-deficient; we resolve this issue by casting the identification problem
as distance minimization. We demonstrate the power of these differential
geometric families for interpolation and identification in a practical
application: wind field covariance approximation for unmanned aerial vehicle
navigation.
| stat |
T-optimal designs for multi-factor polynomial regression models via a
semidefinite relaxation method | We consider T-optimal experiment design problems for discriminating
multi-factor polynomial regression models where the design space is defined by
polynomial inequalities and the regression parameters are constrained to given
convex sets. Our proposed optimality criterion is formulated as a convex
optimization problem with a moment cone constraint. When the regression models
have one factor, an exact semidefinite representation of the moment cone
constraint can be applied to obtain an equivalent semidefinite program. When
there are two or more factors in the models, we apply a moment relaxation
technique and approximate the moment cone constraint by a hierarchy of
semidefinite-representable outer approximations. When the relaxation hierarchy
converges, an optimal discrimination design can be recovered from the optimal
moment matrix, and its optimality is confirmed by an equivalence theorem. The
methodology is illustrated with several examples.
| stat |
Better Exploration with Optimistic Actor-Critic | Actor-critic methods, a type of model-free Reinforcement Learning, have been
successfully applied to challenging tasks in continuous control, often
achieving state-of-the art performance. However, wide-scale adoption of these
methods in real-world domains is made difficult by their poor sample
efficiency. We address this problem both theoretically and empirically. On the
theoretical side, we identify two phenomena preventing efficient exploration in
existing state-of-the-art algorithms such as Soft Actor Critic. First,
combining a greedy actor update with a pessimistic estimate of the critic leads
to the avoidance of actions that the agent does not know about, a phenomenon we
call pessimistic underexploration. Second, current algorithms are directionally
uninformed, sampling actions with equal probability in opposite directions from
the current mean. This is wasteful, since we typically need actions taken along
certain directions much more than others. To address both of these phenomena,
we introduce a new algorithm, Optimistic Actor Critic, which approximates a
lower and upper confidence bound on the state-action value function. This
allows us to apply the principle of optimism in the face of uncertainty to
perform directed exploration using the upper bound while still using the lower
bound to avoid overestimation. We evaluate OAC in several challenging
continuous control tasks, achieving state-of the art sample efficiency.
| stat |
Approximate selective inference via maximum likelihood | This article considers a conditional approach to selective inference via
approximate maximum likelihood for data described by Gaussian models. There are
two important considerations in adopting a post-selection inferential
perspective. While one of them concerns the effective use of information in
data, the other aspect deals with the computational cost of adjusting for
selection. Our approximate proposal serves both these purposes-- (i) exploits
the use of randomness for efficient utilization of left-over information from
selection; (ii) enables us to bypass potentially expensive MCMC sampling from
conditional distributions. At the core of our method is the solution to a
convex optimization problem which assumes a separable form across multiple
selection queries. This allows us to address the problem of tractable and
efficient inference in many practical scenarios, where more than one learning
query is conducted to define and perhaps redefine models and their
corresponding parameters. Through an in-depth analysis, we illustrate the
potential of our proposal and provide extensive comparisons with other
post-selective schemes in both randomized and non-randomized paradigms of
inference.
| 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.