title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
A Systematic Comparison of Bayesian Deep Learning Robustness in Diabetic
Retinopathy Tasks | Evaluation of Bayesian deep learning (BDL) methods is challenging. We often
seek to evaluate the methods' robustness and scalability, assessing whether new
tools give `better' uncertainty estimates than old ones. These evaluations are
paramount for practitioners when choosing BDL tools on-top of which they build
their applications. Current popular evaluations of BDL methods, such as the UCI
experiments, are lacking: Methods that excel with these experiments often fail
when used in application such as medical or automotive, suggesting a pertinent
need for new benchmarks in the field. We propose a new BDL benchmark with a
diverse set of tasks, inspired by a real-world medical imaging application on
\emph{diabetic retinopathy diagnosis}. Visual inputs (512x512 RGB images of
retinas) are considered, where model uncertainty is used for medical
pre-screening---i.e. to refer patients to an expert when model diagnosis is
uncertain. Methods are then ranked according to metrics derived from
expert-domain to reflect real-world use of model uncertainty in automated
diagnosis. We develop multiple tasks that fall under this application,
including out-of-distribution detection and robustness to distribution shift.
We then perform a systematic comparison of well-tuned BDL techniques on the
various tasks. From our comparison we conclude that some current techniques
which solve benchmarks such as UCI `overfit' their uncertainty to the
dataset---when evaluated on our benchmark these underperform in comparison to
simpler baselines. The code for the benchmark, its baselines, and a simple API
for evaluating new BDL tools are made available at
https://github.com/oatml/bdl-benchmarks.
| stat |
Manufacturing Process Optimization using Statistical Methodologies | Response Surface Methodology (RSM) introduced in the paper (Box & Wilson,
1951) explores the relationships between explanatory and response variables in
complex settings and provides a framework to identify correct settings for the
explanatory variables to yield the desired response. RSM involves setting up
sequential experimental designs followed by application of elementary
optimization methods to identify direction of improvement in response. In this
paper, an application of RSM using a two-factor two-level Central Composite
Design (CCD) is explained for a diesel engine nozzle manufacturing sub-process.
The analysis shows that one of the factors has a significant influence in
improving desired values of the response. The implementation of RSM is done
using the DoE plug-in available in R software.
| stat |
Directed Acyclic Graphs and causal thinking in clinical risk prediction
modeling | Background: In epidemiology, causal inference and prediction modeling
methodologies have been historically distinct. Directed Acyclic Graphs (DAGs)
are used to model a priori causal assumptions and inform variable selection
strategies for causal questions. Although tools originally designed for
prediction are finding applications in causal inference, the counterpart has
remained largely unexplored. The aim of this theoretical and simulation-based
study is to assess the potential benefit of using DAGs in clinical risk
prediction modeling.
Methods and Findings: We explore how incorporating knowledge about the
underlying causal structure can provide insights about the transportability of
diagnostic clinical risk prediction models to different settings. A
single-predictor model in the causal direction is likely to have better
transportability than one in the anticausal direction. We further probe whether
causal knowledge can be used to improve predictor selection. We empirically
show that the Markov Blanket, the set of variables including the parents,
children, and parents of the children of the outcome node in a DAG, is the
optimal set of predictors for that outcome.
Conclusions: Our findings challenge the generally accepted notion that a
change in the distribution of the predictors does not affect diagnostic
clinical risk prediction model calibration if the predictors are properly
included in the model. Furthermore, using DAGs to identify Markov Blanket
variables may be a useful, efficient strategy to select predictors in clinical
risk prediction models if strong knowledge of the underlying causal structure
exists or can be learned.
| stat |
Heterogeneity Learning for SIRS model: an Application to the COVID-19 | We propose a Bayesian Heterogeneity Learning approach for
Susceptible-Infected-Removal-Susceptible (SIRS) model that allows underlying
clustering patterns for transmission rate, recovery rate, and loss of immunity
rate for the latest coronavirus (COVID-19) among different regions. Our
proposed method provides simultaneously inference on parameter estimation and
clustering information which contains both number of clusters and cluster
configurations. Specifically, our key idea is to formulates the SIRS model into
a hierarchical form and assign the Mixture of Finite mixtures priors for
heterogeneity learning. The properties of the proposed models are examined and
a Markov chain Monte Carlo sampling algorithm is used to sample from the
posterior distribution. Extensive simulation studies are carried out to examine
empirical performance of the proposed methods. We further apply the proposed
methodology to analyze the state level COVID-19 data in U.S.
| stat |
A cautionary note on the Hanurav-Vijayan sampling algorithm | We consider the Hanurav-Vijayan sampling design, which is the default method
programmed in the SURVEYSELECT procedure of the SAS software. We prove that it
is equivalent to the Sunter procedure, but is capable of handling any set of
inclusion probabilities. We prove that the Horvitz-Thompson estimator is not
generally consistent under this sampling design. We propose a conditional
Horvitz-Thompson estimator, and prove its consistency under a non-standard
assumption on the first-order inclusion probabilities. Since this assumption
seems difficult to control in practice, we recommend not to use the
Hanurav-Vijayan sampling design.
| stat |
Ridge Regularized Estimation of VAR Models for Inference and Sieve
Approximation | Developments in statistical learning have fueled the analysis of
high-dimensional time series. However, even in low-dimensional contexts the
issues arising from ill-conditioned regression problems are well-known. Because
linear time series modeling is naturally prone to such issues, I propose to
apply ridge regression to the estimation of dense VAR models. Theoretical
non-asymptotic results concerning the addition of a ridge-type penalty to the
least squares estimator are discussed, while standard asymptotic and inference
techniques are proven to be valid under mild conditions on the regularizer. The
proposed estimator is then applied to the problem of sieve approximation of
VAR($\infty$) processes under moderately harsh sample sizes. Simulation
evidence is used to discuss the small sample properties of the ridge estimator
(RLS) when compared to least squares and local projection approaches: I use a
Monte Carlo exercise to argue that RLS with a lag-adapted cross-validated
regularizer achieve meaningfully better performance in recovering impulse
response functions and asymptotic confidence intervals than other common
approaches.
| stat |
LazyIter: A Fast Algorithm for Counting Markov Equivalent DAGs and
Designing Experiments | The causal relationships among a set of random variables are commonly
represented by a Directed Acyclic Graph (DAG), where there is a directed edge
from variable $X$ to variable $Y$ if $X$ is a direct cause of $Y$. From the
purely observational data, the true causal graph can be identified up to a
Markov Equivalence Class (MEC), which is a set of DAGs with the same
conditional independencies between the variables. The size of an MEC is a
measure of complexity for recovering the true causal graph by performing
interventions. We propose a method for efficient iteration over possible MECs
given intervention results. We utilize the proposed method for computing MEC
sizes and experiment design in active and passive learning settings. Compared
to previous work for computing the size of MEC, our proposed algorithm reduces
the time complexity by a factor of $O(n)$ for sparse graphs where $n$ is the
number of variables in the system. Additionally, integrating our approach with
dynamic programming, we design an optimal algorithm for passive experiment
design. Experimental results show that our proposed algorithms for both
computing the size of MEC and experiment design outperform the state of the
art.
| stat |
Multi-Node EM Algorithm for Finite Mixture Models | Finite mixture models are powerful tools for modelling and analyzing
heterogeneous data. Parameter estimation is typically carried out using maximum
likelihood estimation via the Expectation-Maximization (EM) algorithm.
Recently, the adoption of flexible distributions as component densities has
become increasingly popular. Often, the EM algorithm for these models involves
complicated expressions that are time-consuming to evaluate numerically. In
this paper, we describe a parallel implementation of the EM-algorithm suitable
for both single-threaded and multi-threaded processors and for both single
machine and multiple-node systems. Numerical experiments are performed to
demonstrate the potential performance gain n different settings. Comparison is
also made across two commonly used platforms - R and MATLAB. For illustration,
a fairly general mixture model is used in the comparison.
| stat |
Promise and Challenges of a Data-Driven Approach for Battery Lifetime
Prognostics | Recent data-driven approaches have shown great potential in early prediction
of battery cycle life by utilizing features from the discharge voltage curve.
However, these studies caution that data-driven approaches must be combined
with specific design of experiments in order to limit the range of aging
conditions, since the expected life of Li-ion batteries is a complex function
of various aging factors. In this work, we investigate the performance of the
data-driven approach for battery lifetime prognostics with Li-ion batteries
cycled under a variety of aging conditions, in order to determine when the
data-driven approach can successfully be applied. Results show a correlation
between the variance of the discharge capacity difference and the end-of-life
for cells aged under a wide range of charge/discharge C-rates and operating
temperatures. This holds despite the different conditions being used not only
to cycle the batteries but also to obtain the features: the features are
calculated directly from cycling data without separate slow characterization
cycles at a controlled temperature. However, the correlation weakens
considerably when the voltage data window for feature extraction is reduced, or
when features from the charge voltage curve instead of discharge are used. As
deep constant-current discharges rarely happen in practice, this imposes new
challenges for applying this method in a real-world system.
| stat |
Estimation and inference in metabolomics with non-random missing data
and latent factors | High throughput metabolomics data are fraught with both non-ignorable missing
observations and unobserved factors that influence a metabolite's measured
concentration, and it is well known that ignoring either of these complications
can compromise estimators. However, current methods to analyze these data can
only account for the missing data or unobserved factors, but not both. We
therefore developed MetabMiss, a statistically rigorous method to account for
both non-random missing data and latent factors in high throughput metabolomics
data. Our methodology does not require the practitioner specify a probability
model for the missing data, and makes investigating the relationship between
the metabolome and tens, or even hundreds, of phenotypes computationally
tractable. We demonstrate the fidelity of MetabMiss's estimates using both
simulated and real metabolomics data. An R package that implements our method
is available from https://github.com/chrismckennan/MetabMiss.
| stat |
Edge-promoting adaptive Bayesian experimental design for X-ray imaging | This work considers sequential edge-promoting Bayesian experimental design
for (discretized) linear inverse problems, exemplified by X-ray tomography. The
process of computing a total variation type reconstruction of the absorption
inside the imaged body via lagged diffusivity iteration is interpreted in the
Bayesian framework. Assuming a Gaussian additive noise model, this leads to an
approximate Gaussian posterior with a covariance structure that contains
information on the location of edges in the posterior mean. The next projection
geometry is then chosen through A-optimal Bayesian design, which corresponds to
minimizing the trace of the updated posterior covariance matrix that accounts
for the new projection. Two and three-dimensional numerical examples based on
simulated data demonstrate the functionality of the introduced approach.
| stat |
Dataset Inference: Ownership Resolution in Machine Learning | With increasingly more data and computation involved in their training,
machine learning models constitute valuable intellectual property. This has
spurred interest in model stealing, which is made more practical by advances in
learning with partial, little, or no supervision. Existing defenses focus on
inserting unique watermarks in a model's decision surface, but this is
insufficient: the watermarks are not sampled from the training distribution and
thus are not always preserved during model stealing. In this paper, we make the
key observation that knowledge contained in the stolen model's training set is
what is common to all stolen copies. The adversary's goal, irrespective of the
attack employed, is always to extract this knowledge or its by-products. This
gives the original model's owner a strong advantage over the adversary: model
owners have access to the original training data. We thus introduce $dataset$
$inference$, the process of identifying whether a suspected model copy has
private knowledge from the original model's dataset, as a defense against model
stealing. We develop an approach for dataset inference that combines
statistical testing with the ability to estimate the distance of multiple data
points to the decision boundary. Our experiments on CIFAR10, SVHN, CIFAR100 and
ImageNet show that model owners can claim with confidence greater than 99% that
their model (or dataset as a matter of fact) was stolen, despite only exposing
50 of the stolen model's training points. Dataset inference defends against
state-of-the-art attacks even when the adversary is adaptive. Unlike prior
work, it does not require retraining or overfitting the defended model.
| stat |
Hypothesis Testing for Shapes using Vectorized Persistence Diagrams | Topological data analysis involves the statistical characterization of the
shape of data. Persistent homology is a primary tool of topological data
analysis, which can be used to analyze those topological features and perform
statistical inference. In this paper, we present a two-stage hypothesis test
for vectorized persistence diagrams. The first stage filters elements in the
vectorized persistence diagrams to reduce false positives. The second stage
consists of multiple hypothesis tests, with false positives controlled by false
discovery rates. We demonstrate applications of the proposed procedure on
simulated point clouds and three-dimensional rock image data. Our results show
that the proposed hypothesis tests can provide flexible and informative
inferences on the shape of data with lower computational cost compared to the
permutation test.
| stat |
Inference for extreme values under threshold-based stopping rules | There is a propensity for an extreme value analyses to be conducted as a
consequence of the occurrence of a large flooding event. This timing of the
analysis introduces bias and poor coverage probabilities into the associated
risk assessments and leads subsequently to inefficient flood protection
schemes. We explore these problems through studying stochastic stopping
criteria and propose new likelihood-based inferences that mitigate against
these difficulties. Our methods are illustrated through the analysis of the
river Lune, following it experiencing the UK's largest ever measured flow event
in 2015. We show that without accounting for this stopping feature there would
be substantial over-design in response to the event.
| stat |
Scalable Metropolis-Hastings for Exact Bayesian Inference with Large
Datasets | Bayesian inference via standard Markov Chain Monte Carlo (MCMC) methods is
too computationally intensive to handle large datasets, since the cost per step
usually scales like $\Theta(n)$ in the number of data points $n$. We propose
the Scalable Metropolis-Hastings (SMH) kernel that exploits Gaussian
concentration of the posterior to require processing on average only $O(1)$ or
even $O(1/\sqrt{n})$ data points per step. This scheme is based on a
combination of factorized acceptance probabilities, procedures for fast
simulation of Bernoulli processes, and control variate ideas. Contrary to many
MCMC subsampling schemes such as fixed step-size Stochastic Gradient Langevin
Dynamics, our approach is exact insofar as the invariant distribution is the
true posterior and not an approximation to it. We characterise the performance
of our algorithm theoretically, and give realistic and verifiable conditions
under which it is geometrically ergodic. This theory is borne out by empirical
results that demonstrate overall performance benefits over standard
Metropolis-Hastings and various subsampling algorithms.
| stat |
On the Likelihood of Local Projection Models | A local projection model is defined by a set of linear regressions that
account for the associations between exogenous variables and an endogenous
variable observed at different time points. While it is standard practice to
separately estimate individual regressions using the ordinary least squares
estimator, some recent studies treat a local projection model as a multivariate
regression with correlated errors, i.e., seemingly unrelated regressions, and
propose Bayesian and non-Bayesian methods to improve the estimation accuracy.
However, it is not clear how and when this way of treatment of local projection
models is justified. The primary purpose of this paper is to fill this gap by
showing that the likelihood of local projection models can be analytically
derived from a stationary vector moving average process. By means of numerical
experiments, we confirm that this treatment of local projections is tenable for
finite samples.
| stat |
A Fast deflation Method for Sparse Principal Component Analysis via
Subspace Projections | The implementation of conventional sparse principal component analysis (SPCA)
on high-dimensional data sets has become a time consuming work. In this paper,
a series of subspace projections are constructed efficiently by using Household
QR factorization. With the aid of these subspace projections, a fast deflation
method, called SPCA-SP, is developed for SPCA. This method keeps a good
tradeoff between various criteria, including sparsity, orthogonality, explained
variance, balance of sparsity, and computational cost. Comparative experiments
on the benchmark data sets confirm the effectiveness of the proposed method.
| stat |
Large-scale inference of correlation among mixed-type biological traits
with phylogenetic multivariate probit models | Inferring concerted changes among biological traits along an evolutionary
history remains an important yet challenging problem. Besides adjusting for
spurious correlation induced from the shared history, the task also requires
sufficient flexibility and computational efficiency to incorporate multiple
continuous and discrete traits as data size increases. To accomplish this, we
jointly model mixed-type traits by assuming latent parameters for binary
outcome dimensions at the tips of an unknown tree informed by molecular
sequences. This gives rise to a phylogenetic multivariate probit model. With
large sample sizes, posterior computation under this model is problematic, as
it requires repeated sampling from a high-dimensional truncated normal
distribution. Current best practices employ multiple-try rejection sampling
that suffers from slow-mixing and a computational cost that scales
quadratically in sample size. We develop a new inference approach that exploits
1) the bouncy particle sampler (BPS) based on piecewise deterministic Markov
processes to simultaneously sample all truncated normal dimensions, and 2)
novel dynamic programming that reduces the cost of likelihood and gradient
evaluations for BPS to linear in sample size. In an application with 535 HIV
viruses and 24 traits that necessitates sampling from a 12,840-dimensional
truncated normal, our method makes it possible to estimate the across-trait
correlation and detect factors that affect the pathogen's capacity to cause
disease. This inference framework is also applicable to a broader class of
covariance structures beyond comparative biology.
| stat |
anomaly : Detection of Anomalous Structure in Time Series Data | One of the contemporary challenges in anomaly detection is the ability to
detect, and differentiate between, both point and collective anomalies within a
data sequence or time series. The \pkg{anomaly} package has been developed to
provide users with a choice of anomaly detection methods and, in particular,
provides an implementation of the recently proposed CAPA family of anomaly
detection algorithms. This article describes the methods implemented whilst
also highlighting their application to simulated data as well as real data
examples contained in the package.
| stat |
Variational Bayes under Model Misspecification | Variational Bayes (VB) is a scalable alternative to Markov chain Monte Carlo
(MCMC) for Bayesian posterior inference. Though popular, VB comes with few
theoretical guarantees, most of which focus on well-specified models. However,
models are rarely well-specified in practice. In this work, we study VB under
model misspecification. We prove the VB posterior is asymptotically normal and
centers at the value that minimizes the Kullback-Leibler (KL) divergence to the
true data-generating distribution. Moreover, the VB posterior mean centers at
the same value and is also asymptotically normal. These results generalize the
variational Bernstein--von Mises theorem [29] to misspecified models. As a
consequence of these results, we find that the model misspecification error
dominates the variational approximation error in VB posterior predictive
distributions. It explains the widely observed phenomenon that VB achieves
comparable predictive accuracy with MCMC even though VB uses an approximating
family. As illustrations, we study VB under three forms of model
misspecification, ranging from model over-/under-dispersion to latent
dimensionality misspecification. We conduct two simulation studies that
demonstrate the theoretical results.
| stat |
Approximate Bayesian inference for a "steps and turns" continuous-time
random walk observed at regular time intervals | The study of animal movement is challenging because it is a process modulated
by many factors acting at different spatial and temporal scales. Several models
have been proposed which differ primarily in the temporal conceptualization,
namely continuous and discrete time formulations. Naturally, animal movement
occurs in continuous time but we tend to observe it at fixed time intervals. To
account for the temporal mismatch between observations and movement decisions,
we used a state-space model where movement decisions (steps and turns) are made
in continuous time. The movement process is then observed at regular time
intervals. As the likelihood function of this state-space model turned out to
be complex to calculate yet simulating data is straightforward, we conduct
inference using a few variations of Approximate Bayesian Computation (ABC). We
explore the applicability of these methods as a function of the discrepancy
between the temporal scale of the observations and that of the movement process
in a simulation study. We demonstrate the application of this model to a real
trajectory of a sheep that was reconstructed in high resolution using
information from magnetometer and GPS devices. Our results suggest that
accurate estimates can be obtained when the observations are less than 5 times
the average time between changes in movement direction. The state-space model
used here allowed us to connect the scales of the observations and movement
decisions in an intuitive and easy to interpret way. Our findings underscore
the idea that the time scale at which animal movement decisions are made needs
to be considered when designing data collection protocols, and that sometimes
high-frequency data may not be necessary to have good estimates of certain
movement processes.
| stat |
Personalized Policy Learning using Longitudinal Mobile Health Data | We address the personalized policy learning problem using longitudinal mobile
health application usage data. Personalized policy represents a paradigm shift
from developing a single policy that may prescribe personalized decisions by
tailoring. Specifically, we aim to develop the best policy, one per user, based
on estimating random effects under generalized linear mixed model. With many
random effects, we consider new estimation method and penalized objective to
circumvent high-dimension integrals for marginal likelihood approximation. We
establish consistency and optimality of our method with endogenous app usage.
We apply our method to develop personalized push ("prompt") schedules in 294
app users, with a goal to maximize the prompt response rate given past app
usage and other contextual factors. We found the best push schedule given the
same covariates varied among the users, thus calling for personalized policies.
Using the estimated personalized policies would have achieved a mean prompt
response rate of 23% in these users at 16 weeks or later: this is a remarkable
improvement on the observed rate (11%), while the literature suggests 3%-15%
user engagement at 3 months after download. The proposed method compares
favorably to existing estimation methods including using the R function "glmer"
in a simulation study.
| stat |
Variational Inference and Sparsity in High-Dimensional Deep Gaussian
Mixture Models | Gaussian mixture models are a popular tool for model-based clustering, and
mixtures of factor analyzers are Gaussian mixture models having parsimonious
factor covariance structure for mixture components. There are several recent
extensions of mixture of factor analyzers to deep mixtures, where the Gaussian
model for the latent factors is replaced by a mixture of factor analyzers. This
construction can be iterated to obtain a model with many layers. These deep
models are challenging to fit, and we consider Bayesian inference using
sparsity priors to further regularize the estimation. A scalable natural
gradient variational inference algorithm is developed for fitting the model,
and we suggest computationally efficient approaches to the architecture choice
using overfitted mixtures where unnecessary components drop out in the
estimation. In a number of simulated and two real examples, we demonstrate the
versatility of our approach for high-dimensional problems, and demonstrate that
the use of sparsity inducing priors can be helpful for obtaining improved
clustering results.
| stat |
Confidence Intervals for Selected Parameters | Practical or scientific considerations often lead to selecting a subset of
parameters as ``important.'' Inferences about those parameters often are based
on the same data used to select them in the first place. That can make the
reported uncertainties deceptively optimistic: confidence intervals that ignore
selection generally have less than their nominal coverage probability.
Controlling the probability that one or more intervals for selected parameters
do not cover---the ``simultaneous over the selected'' (SoS) error rate---is
crucial in many scientific problems. Intervals that control the SoS error rate
can be constructed in ways that take advantage of knowledge of the selection
rule. We construct SoS-controlling confidence intervals for parameters deemed
the most ``important'' $k$ of $m$ shift parameters because they are estimated
(by independent estimators) to be the largest. The new intervals improve
substantially over \v{S}id\'{a}k intervals when $k$ is small compared to $m$,
and approach the standard Bonferroni-corrected intervals when $k \approx m$.
Standard, unadjusted confidence intervals for location parameters have the
correct coverage probability for $k=1$, $m=2$ if, when the true parameters are
zero, the estimators are exchangeable and symmetric.
| stat |
Filtering and improved Uncertainty Quantification in the dynamic
estimation of effective reproduction numbers | The effective reproduction number $R_t$ measures an infectious disease's
transmissibility as the number of secondary infections in one reproduction time
in a population having both susceptible and non-susceptible hosts. Current
approaches do not quantify the uncertainty correctly in estimating $R_t$, as
expected by the observed variability in contagion patterns. We elaborate on the
Bayesian estimation of $R_t$ by improving on the Poisson sampling model of Cori
et al. (2013). By adding an autoregressive latent process, we build a Dynamic
Linear Model on the log of observed $R_t$s, resulting in a filtering type
Bayesian inference. We use a conjugate analysis, and all calculations are
explicit. Results show an improved uncertainty quantification on the estimation
of $R_t$'s, with a reliable method that could safely be used by non-experts and
within other forecasting systems. We illustrate our approach with recent data
from the current COVID19 epidemic in Mexico.
| stat |
Kernel Hypothesis Testing with Set-valued Data | We present a general framework for hypothesis testing on distributions of
sets of individual examples. Sets may represent many common data sources such
as groups of observations in time series, collections of words in text or a
batch of images of a given phenomenon. This observation pattern, however,
differs from the common assumptions required for hypothesis testing: each set
differs in size, may have differing levels of noise, and also may incorporate
nuisance variability, irrelevant for the analysis of the phenomenon of
interest; all features that bias test decisions if not accounted for. In this
paper, we propose to interpret sets as independent samples from a collection of
latent probability distributions, and introduce kernel two-sample and
independence tests in this latent space of distributions. We prove the
consistency of tests and observe them to outperform in a wide range of
synthetic experiments. Finally, we showcase their use in practice with
experiments of healthcare and climate data, where previously heuristics were
needed for feature extraction and testing.
| stat |
A study on information behavior of scholars for article keywords
selection | This project takes the factors of keyword selection behavior as the research
object. Qualitative analysis methods such as interview and grounded theory were
used to construct causal influence path model. Combined with computer
simulation technology such as multi-agent simulation experiment method was used
to study the factors of keyword selection from two dimensions of individual to
group. The research was carried out according to the path of factor analysis at
individual level macro situation simulation optimization of scientific research
data management. Based on the aforementioned review of existing researches and
explanations of keywords selection, this study adopts a qualitative research
design to expand the explanation, and macro simulation based on the results of
qualitative research. There are two steps in this study, one is do interview
with authors and then design macro simulation according the deductive and
qualitative content analysis results.
| stat |
The energy distance for ensemble and scenario reduction | Scenario reduction techniques are widely applied for solving sophisticated
dynamic and stochastic programs, especially in energy and power systems, but
also used in probabilistic forecasting, clustering and estimating generative
adversarial networks (GANs). We propose a new method for ensemble and scenario
reduction based on the energy distance which is a special case of the maximum
mean discrepancy (MMD). We discuss the choice of energy distance in detail,
especially in comparison to the popular Wasserstein distance which is
dominating the scenario reduction literature. The energy distance is a metric
between probability measures that allows for powerful tests for equality of
arbitrary multivariate distributions or independence. Thanks to the latter, it
is a suitable candidate for ensemble and scenario reduction problems. The
theoretical properties and considered examples indicate clearly that the
reduced scenario sets tend to exhibit better statistical properties for the
energy distance than a corresponding reduction with respect to the Wasserstein
distance. We show applications to a Bernoulli random walk and two real data
based examples for electricity demand profiles and day-ahead electricity
prices.
| stat |
First-order Adversarial Vulnerability of Neural Networks and Input
Dimension | Over the past few years, neural networks were proven vulnerable to
adversarial images: targeted but imperceptible image perturbations lead to
drastically different predictions. We show that adversarial vulnerability
increases with the gradients of the training objective when viewed as a
function of the inputs. Surprisingly, vulnerability does not depend on network
topology: for many standard network architectures, we prove that at
initialization, the $\ell_1$-norm of these gradients grows as the square root
of the input dimension, leaving the networks increasingly vulnerable with
growing image size. We empirically show that this dimension dependence persists
after either usual or robust training, but gets attenuated with higher
regularization.
| stat |
A Negative Correlation Strategy for Bracketing in
Difference-in-Differences | The method of difference-in-differences (DID) is widely used to study the
causal effect of policy interventions in observational studies. DID employs a
before and after comparison of the treated and control units to remove bias due
to time-invariant unmeasured confounders under the parallel trends assumption.
Estimates from DID, however, will be biased if the outcomes for the treated and
control units evolve differently in the absence of treatment, namely if the
parallel trends assumption is violated. We propose a general identification
strategy that leverages two groups of control units whose outcomes relative to
the treated units exhibit a negative correlation, and achieves partial
identification of the average treatment effect for the treated. The identified
set is of a union bounds form that involves the minimum and maximum operators,
which makes the canonical bootstrap generally inconsistent and naive methods
overly conservative. By utilizing the directional inconsistency of the
bootstrap distribution, we develop a novel bootstrap method to construct
uniformly valid confidence intervals for the identified set and parameter of
interest when the identified set is of a union bounds form, and we establish
the method's theoretical properties. We develop a simple falsification test and
sensitivity analysis. We apply the proposed strategy for bracketing to study
whether minimum wage laws affect employment levels.
| stat |
Minimizing post-shock forecasting error through aggregation of outside
information | We develop a forecasting methodology for providing credible forecasts for
time series that have recently undergone a shock. We achieve this by borrowing
knowledge from other time series that have undergone similar shocks for which
post-shock outcomes are observed. Three shock effect estimators are motivated
with the aim of minimizing average forecast risk. We propose risk-reduction
propositions that provide conditions that establish when our methodology works.
Bootstrap and leave-one-out cross validation procedures are provided to
prospectively assess the performance of our methodology. Several simulated data
examples, and a real data example of forecasting Conoco Phillips stock price
are provided for verification and illustration.
| stat |
Unlinked monotone regression | We consider so-called univariate unlinked (sometimes "decoupled," or
"shuffled") regression when the unknown regression curve is monotone. In
standard monotone regression, one observes a pair $(X,Y)$ where a response $Y$
is linked to a covariate $X$ through the model $Y= m_0(X) + \epsilon$, with
$m_0$ the (unknown) monotone regression function and $\epsilon$ the unobserved
error (assumed to be independent of $X$). In the unlinked regression setting
one gets only to observe a vector of realizations from both the response $Y$
and from the covariate $X$ where now $Y \stackrel{d}{=} m_0(X) + \epsilon$.
There is no (observed) pairing of $X$ and $Y$. Despite this, it is actually
still possible to derive a consistent non-parametric estimator of $m_0$ under
the assumption of monotonicity of $m_0$ and knowledge of the distribution of
the noise $\epsilon$. In this paper, we establish an upper bound on the rate of
convergence of such an estimator under minimal assumption on the distribution
of the covariate $X$. We discuss extensions to the case in which the
distribution of the noise is unknown. We develop a gradient-descent-based
algorithm for its computation, and we demonstrate its use on synthetic data.
Finally, we apply our method (in a fully data driven way, without knowledge of
the error distribution) on longitudinal data from the US Consumer Expenditure
Survey.
| stat |
Simulating Markov random fields with a conclique-based Gibbs sampler | For spatial and network data, we consider models formed from a Markov random
field (MRF) structure and the specification of a conditional distribution for
each observation. Fast simulation from such MRF models is often an important
consideration, particularly when repeated generation of large numbers of data
sets is required. However, a standard Gibbs strategy for simulating from MRF
models involves single-site updates, performed with the conditional univariate
distribution of each observation in a sequential manner, whereby a complete
Gibbs iteration may become computationally involved even for moderate samples.
As an alternative, we describe a general way to simulate from MRF models using
Gibbs sampling with "concliques" (i.e., groups of non-neighboring
observations). Compared to standard Gibbs sampling, this simulation scheme can
be much faster by reducing Gibbs steps and independently updating all
observations per conclique at once. The speed improvement depends on the number
of concliques relative to the sample size for simulation, and
order-of-magnitude speed increases are possible with many MRF models (e.g.,
having appropriately bounded neighborhoods). We detail the simulation method,
establish its validity, and assess its computational performance through
numerical studies, where speed advantages are shown for several spatial and
network examples.
| stat |
Towards A Unified Analysis of Random Fourier Features | Random Fourier features is a widely used, simple, and effective technique for
scaling up kernel methods. The existing theoretical analysis of the approach,
however, remains focused on specific learning tasks and typically gives
pessimistic bounds which are at odds with the empirical results. We tackle
these problems and provide the first unified risk analysis of learning with
random Fourier features using the squared error and Lipschitz continuous loss
functions. In our bounds, the trade-off between the computational cost and the
expected risk convergence rate is problem specific and expressed in terms of
the regularization parameter and the \emph{number of effective degrees of
freedom}. We study both the standard random Fourier features method for which
we improve the existing bounds on the number of features required to guarantee
the corresponding minimax risk convergence rate of kernel ridge regression, as
well as a data-dependent modification which samples features proportional to
\emph{ridge leverage scores} and further reduces the required number of
features. As ridge leverage scores are expensive to compute, we devise a simple
approximation scheme which provably reduces the computational cost without loss
of statistical efficiency.
| stat |
Impact of internal migration on population redistribution in Europe:
Urbanisation, counterurbanisation or spatial equilibrium? | The classical foundations of migration research date from the 1880s with
Ravenstein's Laws of migration, which represent the first comparative analyses
of internal migration. While his observations remain largely valid, the ensuing
century has seen considerable progress in data collection practices and methods
of analysis, which in turn has permitted theoretical advances in understanding
the role of migration in population redistribution. Coupling the extensive
range of migration data now available with these recent theoretical and
methodological advances, we endeavour to advance beyond Ravenstein's
understanding by examining the direction of population redistribution and
comparing the impact of internal migration on patterns of human settlement in
27 European countries. Results show that the overall redistributive impact of
internal migration is low in most European countries but the mechanisms differ
across the continent. In Southern and Eastern Europe migration effectiveness is
above average but is offset by low migration intensities, whereas in Northern
and Western Europe high intensities are absorbed in reciprocal flows resulting
in low migration effectiveness. About half the European countries are
experiencing a process of concentration toward urbanised regions, particularly
in Northern, Central and Eastern Europe, whereas countries in the West and
South are undergoing a process of population deconcentration. These results
suggest that population deconcentration is now more common than it was in the
1990s when counterurbanisation was limited to Western Europe. The results show
that 130 years on, Ravenstein's law of migration streams and counter-streams
remains a central facet of migration dynamics, while underlining the importance
of simple yet robust indices for the spatial analysis of migration.
| stat |
orthoDr: Semiparametric Dimension Reduction via Orthogonality
Constrained Optimization | orthoDr is a package in R that solves dimension reduction problems using
orthogonality constrained optimization approach. The package serves as a
unified framework for many regression and survival analysis dimension reduction
models that utilize semiparametric estimating equations. The main computational
machinery of orthoDr is a first-order algorithm developed by
\cite{wen2013feasible} for optimization within the Stiefel manifold. We
implement the algorithm through Rcpp and OpenMP for fast computation. In
addition, we developed a general-purpose solver for such constrained problems
with user-specified objective functions, which works as a drop-in version of
optim(). The package also serves as a platform for future methodology
developments along this line of work.
| stat |
A copula transformation in multivariate mixed discrete-continuous models | Copulas allow a flexible and simultaneous modeling of complicated dependence
structures together with various marginal distributions. Especially if the
density function can be represented as the product of the marginal density
functions and the copula density function, this leads to both an intuitive
interpretation of the conditional distribution and convenient estimation
procedures. However, this is no longer the case for copula models with mixed
discrete and continuous marginal distributions, because the corresponding
density function cannot be decomposed so nicely. In this paper, we introduce a
copula transformation method that allows to represent the density function of a
distribution with mixed discrete and continuous marginals as the product of the
marginal probability mass/density functions and the copula density function.
With the proposed method, conditional distributions can be described
analytically and the computational complexity in the estimation procedure can
be reduced depending on the type of copula used.
| stat |
Reliability study of a coherent system with single general standby
component | The properties of a coherent system with a single general standby component
is investigated. Here three different switch over viz. perfect switching,
imperfect switching and random worm up period of the standby component are
considered with some numerical examples.
| stat |
PIntMF: Penalized Integrative Matrix Factorization Method for
Multi-Omics Data | It is more and more common to explore the genome at diverse levels and not
only at a single omic level. Through integrative statistical methods, omics
data have the power to reveal new biological processes, potential biomarkers,
and subgroups of a cohort. The matrix factorization (MF) is a unsupervised
statistical method that allows giving a clustering of individuals, but also
revealing relevant omic variables from the various blocks. Here, we present
PIntMF (Penalized Integrative Matrix Factorization), a model of MF with
sparsity, positivity and equality constraints.To induce sparsity in the model,
we use a classical Lasso penalization on variable and individual matrices. For
the matrix of samples, sparsity helps for the clustering, and normalization
(matching an equality constraint) of inferred coefficients is added for a
better interpretation. Besides, we add an automatic tuning of the sparsity
parameters using the famous glmnet package. We also proposed three criteria to
help the user to choose the number of latent variables. PIntMF was compared to
other state-of-the-art integrative methods including feature selection
techniques in both synthetic and real data. PIntMF succeeds in finding relevant
clusters as well as variables in two types of simulated data (correlated and
uncorrelated). Then, PIntMF was applied to two real datasets (Diet and cancer),
and it reveals interpretable clusters linked to available clinical data. Our
method outperforms the existing ones on two criteria (clustering and variable
selection). We show that PIntMF is an easy, fast, and powerful tool to extract
patterns and cluster samples from multi-omics data.
| stat |
Ranking and Selection with Covariates for Personalized Decision Making | We consider a problem of ranking and selection via simulation in the context
of personalized decision making, where the best alternative is not universal
but varies as a function of some observable covariates. The goal of ranking and
selection with covariates (R&S-C) is to use simulation samples to obtain a
selection policy that specifies the best alternative with certain statistical
guarantee for subsequent individuals upon observing their covariates. A linear
model is proposed to capture the relationship between the mean performance of
an alternative and the covariates. Under the indifference-zone formulation, we
develop two-stage procedures for both homoscedastic and heteroscedastic
simulation errors, respectively, and prove their statistical validity in terms
of average probability of correct selection. We also generalize the well-known
slippage configuration, and prove that the generalized slippage configuration
is the least favorable configuration for our procedures. Extensive numerical
experiments are conducted to investigate the performance of the proposed
procedures, the experimental design issue, and the robustness to the linearity
assumption. Finally, we demonstrate the usefulness of R&S-C via a case study of
selecting the best treatment regimen in the prevention of esophageal cancer. We
find that by leveraging disease-related personal information, R&S-C can
substantially improve patients' expected quality-adjusted life years by
providing patient-specific treatment regimen.
| stat |
Modeling Multivariate Cyber Risks: Deep Learning Dating Extreme Value
Theory | Modeling cyber risks has been an important but challenging task in the domain
of cyber security. It is mainly because of the high dimensionality and heavy
tails of risk patterns. Those obstacles have hindered the development of
statistical modeling of the multivariate cyber risks. In this work, we propose
a novel approach for modeling the multivariate cyber risks which relies on the
deep learning and extreme value theory. The proposed model not only enjoys the
high accurate point predictions via deep learning but also can provide the
satisfactory high quantile prediction via extreme value theory. The simulation
study shows that the proposed model can model the multivariate cyber risks very
well and provide satisfactory prediction performances. The empirical evidence
based on real honeypot attack data also shows that the proposed model has very
satisfactory prediction performances.
| stat |
Minimax Quasi-Bayesian estimation in sparse canonical correlation
analysis via a Rayleigh quotient function | Canonical correlation analysis (CCA) is a popular statistical technique for
exploring the relationship between datasets. The estimation of sparse canonical
correlation vectors has emerged in recent years as an important but challenging
variation of the CCA problem, with widespread applications. Currently available
rate-optimal estimators for sparse canonical correlation vectors are expensive
to compute. We propose a quasi-Bayesian estimation procedure that achieves the
minimax estimation rate, and yet is easy to compute by Markov Chain Monte Carlo
(MCMC). The method builds on ([37]) and uses a re-scaled Rayleigh quotient
function as a quasi-log-likelihood. However unlike these authors, we adopt a
Bayesian framework that combines this quasi-log-likelihood with a
spike-and-slab prior that serves to regularize the inference and promote
sparsity. We investigated the empirical behavior of the proposed method on both
continuous and truncated data, and we noted that it outperforms several
state-of-the-art methods. As an application, we use the methodology to
maximally correlate clinical variables and proteomic data for a better
understanding of covid-19 disease.
| stat |
Confounder selection strategies targeting stable treatment effect
estimators | Inferring the causal effect of a treatment on an outcome in an observational
study requires adjusting for observed baseline confounders to avoid bias.
However, adjusting for all observed baseline covariates, when only a subset are
confounders of the effect of interest, is known to yield potentially
inefficient and unstable estimators of the treatment effect. Furthermore, it
raises the risk of finite-sample bias and bias due to model misspecification.
For these stated reasons, confounder (or covariate) selection is commonly used
to determine a subset of the available covariates that is sufficient for
confounding adjustment. In this article, we propose a confounder selection
strategy that focuses on stable estimation of the treatment effect. In
particular, when the propensity score model already includes covariates that
are sufficient to adjust for confounding, then the addition of covariates that
are associated with either treatment or outcome alone, but not both, should not
systematically change the effect estimator. The proposal, therefore, entails
first prioritizing covariates for inclusion in the propensity score model, then
using a change-in-estimate approach to select the smallest adjustment set that
yields a stable effect estimate. The ability of the proposal to correctly
select confounders, and to ensure valid inference of the treatment effect
following data-driven covariate selection, is assessed empirically and compared
with existing methods using simulation studies. We demonstrate the procedure
using three different publicly available datasets commonly used for causal
inference.
| stat |
Normalizing Constant Estimation with Gaussianized Bridge Sampling | Normalizing constant (also called partition function, Bayesian evidence, or
marginal likelihood) is one of the central goals of Bayesian inference, yet
most of the existing methods are both expensive and inaccurate. Here we develop
a new approach, starting from posterior samples obtained with a standard Markov
Chain Monte Carlo (MCMC). We apply a novel Normalizing Flow (NF) approach to
obtain an analytic density estimator from these samples, followed by Optimal
Bridge Sampling (OBS) to obtain the normalizing constant. We compare our method
which we call Gaussianized Bridge Sampling (GBS) to existing methods such as
Nested Sampling (NS) and Annealed Importance Sampling (AIS) on several
examples, showing our method is both significantly faster and substantially
more accurate than these methods, and comes with a reliable error estimation.
| stat |
Deep Machine Learning Approach to Develop a New Asphalt Pavement
Condition Index | Automated pavement distress detection via road images is still a challenging
issue among pavement researchers and computer-vision community. In recent
years, advancement in deep learning has enabled researchers to develop robust
tools for analyzing pavement images at unprecedented accuracies. Nevertheless,
deep learning models necessitate a big ground truth dataset, which is often not
readily accessible for pavement field. In this study, we reviewed our previous
study, which a labeled pavement dataset was presented as the first step towards
a more robust, easy-to-deploy pavement condition assessment system. In total,
7237 google street-view images were extracted, manually annotated for
classification (nine categories of distress classes). Afterward, YOLO (you look
only once) deep learning framework was implemented to train the model using the
labeled dataset. In the current study, a U-net based model is developed to
quantify the severity of the distresses, and finally, a hybrid model is
developed by integrating the YOLO and U-net model to classify the distresses
and quantify their severity simultaneously. Various pavement condition indices
are developed by implementing various machine learning algorithms using the
YOLO deep learning framework for distress classification and U-net for
segmentation and distress densification. The output of the distress
classification and segmentation models are used to develop a comprehensive
pavement condition tool which rates each pavement image according to the type
and severity of distress extracted.
| stat |
Robust Bayesian Cluster Enumeration Based on the $t$ Distribution | A major challenge in cluster analysis is that the number of data clusters is
mostly unknown and it must be estimated prior to clustering the observed data.
In real-world applications, the observed data is often subject to heavy tailed
noise and outliers which obscure the true underlying structure of the data.
Consequently, estimating the number of clusters becomes challenging. To this
end, we derive a robust cluster enumeration criterion by formulating the
problem of estimating the number of clusters as maximization of the posterior
probability of multivariate $t_\nu$ distributed candidate models. We utilize
Bayes' theorem and asymptotic approximations to come up with a robust criterion
that possesses a closed-form expression. Further, we refine the derivation and
provide a robust cluster enumeration criterion for data sets with finite sample
size. The robust criteria require an estimate of cluster parameters for each
candidate model as an input. Hence, we propose a two-step cluster enumeration
algorithm that uses the expectation maximization algorithm to partition the
data and estimate cluster parameters prior to the calculation of one of the
robust criteria. The performance of the proposed algorithm is tested and
compared to existing cluster enumeration methods using numerical and real data
experiments.
| stat |
Composition of kernel and acquisition functions for High Dimensional
Bayesian Optimization | Bayesian Optimization has become the reference method for the global
optimization of black box, expensive and possibly noisy functions. Bayesian
Op-timization learns a probabilistic model about the objective function,
usually a Gaussian Process, and builds, depending on its mean and variance, an
acquisition function whose optimizer yields the new evaluation point, leading
to update the probabilistic surrogate model. Despite its sample efficiency,
Bayesian Optimiza-tion does not scale well with the dimensions of the problem.
The optimization of the acquisition function has received less attention
because its computational cost is usually considered negligible compared to
that of the evaluation of the objec-tive function. Its efficient optimization
is often inhibited, particularly in high di-mensional problems, by multiple
extrema. In this paper we leverage the addition-ality of the objective function
into mapping both the kernel and the acquisition function of the Bayesian
Optimization in lower dimensional subspaces. This ap-proach makes more
efficient the learning/updating of the probabilistic surrogate model and allows
an efficient optimization of the acquisition function. Experi-mental results
are presented for real-life application, that is the control of pumps in urban
water distribution systems.
| stat |
A Latent Variable Approach to Gaussian Process Modeling with Qualitative
and Quantitative Factors | Computer simulations often involve both qualitative and numerical inputs.
Existing Gaussian process (GP) methods for handling this mainly assume a
different response surface for each combination of levels of the qualitative
factors and relate them via a multiresponse cross-covariance matrix. We
introduce a substantially different approach that maps each qualitative factor
to an underlying numerical latent variable (LV), with the mapped value for each
level estimated similarly to the correlation parameters. This provides a
parsimonious GP parameterization that treats qualitative factors the same as
numerical variables and views them as effecting the response via similar
physical mechanisms. This has strong physical justification, as the effects of
a qualitative factor in any physics-based simulation model must always be due
to some underlying numerical variables. Even when the underlying variables are
many, sufficient dimension reduction arguments imply that their effects can be
represented by a low-dimensional LV. This conjecture is supported by the
superior predictive performance observed across a variety of examples.
Moreover, the mapped LVs provide substantial insight into the nature and
effects of the qualitative factors.
| stat |
Assessing the global and local uncertainty in scientific evidence in the
presence of model misspecification | Scientists need to compare the support for models based on observed
phenomena. The main goal of the evidential paradigm is to quantify the strength
of evidence in the data for a reference model relative to an alternative model.
This is done via an evidence function, such as $\Delta SIC$, an estimator of
the sample size scaled difference of divergences between the generating
mechanism and the competing models. To use evidence, either for decision making
or as a guide to the accumulation of knowledge, an understanding of the
uncertainty in the evidence is needed. This uncertainty is well characterized
by the standard statistical theory of estimation. Unfortunately, the standard
theory breaks down if the models are misspecified, as it is normally the case
in scientific studies. We develop non-parametric bootstrap methodologies for
estimating the sampling distribution of the evidence estimator under model
misspecification. This sampling distribution allows us to determine how secure
we are in our evidential statement. We characterize this uncertainty in the
strength of evidence with two different types of confidence intervals, which we
term "global" and "local". We discuss how evidence uncertainty can be used to
improve scientific inference and illustrate this with a reanalysis of the model
identification problem in a prominent landscape ecology study (Grace and
Keeley, 2006) using structural equations.
| stat |
The Minecraft Kernel: Modelling correlated Gaussian Processes in the
Fourier domain | In the univariate setting, using the kernel spectral representation is an
appealing approach for generating stationary covariance functions. However,
performing the same task for multiple-output Gaussian processes is
substantially more challenging. We demonstrate that current approaches to
modelling cross-covariances with a spectral mixture kernel possess a critical
blind spot. For a given pair of processes, the cross-covariance is not
reproducible across the full range of permitted correlations, aside from the
special case where their spectral densities are of identical shape. We present
a solution to this issue by replacing the conventional Gaussian components of a
spectral mixture with block components of finite bandwidth (i.e. rectangular
step functions). The proposed family of kernel represents the first
multi-output generalisation of the spectral mixture kernel that can approximate
any stationary multi-output kernel to arbitrary precision.
| stat |
Variable selection with false discovery rate control in deep neural
networks | Deep neural networks (DNNs) are famous for their high prediction accuracy,
but they are also known for their black-box nature and poor interpretability.
We consider the problem of variable selection, that is, selecting the input
variables that have significant predictive power on the output, in DNNs. We
propose a backward elimination procedure called SurvNet, which is based on a
new measure of variable importance that applies to a wide variety of networks.
More importantly, SurvNet is able to estimate and control the false discovery
rate of selected variables, while no existing methods provide such a quality
control. Further, SurvNet adaptively determines how many variables to eliminate
at each step in order to maximize the selection efficiency. To study its
validity, SurvNet is applied to image data and gene expression data, as well as
various simulation datasets.
| stat |
Learning Beam Search Policies via Imitation Learning | Beam search is widely used for approximate decoding in structured prediction
problems. Models often use a beam at test time but ignore its existence at
train time, and therefore do not explicitly learn how to use the beam. We
develop an unifying meta-algorithm for learning beam search policies using
imitation learning. In our setting, the beam is part of the model, and not just
an artifact of approximate decoding. Our meta-algorithm captures existing
learning algorithms and suggests new ones. It also lets us show novel no-regret
guarantees for learning beam search policies.
| stat |
UltraNest -- a robust, general purpose Bayesian inference engine | UltraNest is a general-purpose Bayesian inference package for parameter
estimation and model comparison. It allows fitting arbitrary models specified
as likelihood functions written in Python, C, C++, Fortran, Julia or R. With a
focus on correctness and speed (in that order), UltraNest is especially useful
for multi-modal or non-Gaussian parameter spaces, computational expensive
models, in robust pipelines. Parallelisation to computing clusters and resuming
incomplete runs is available.
| stat |
Semi-Unsupervised Learning: Clustering and Classifying using
Ultra-Sparse Labels | In semi-supervised learning for classification, it is assumed that every
ground truth class of data is present in the small labelled dataset. Many
real-world sparsely-labelled datasets are plausibly not of this type. It could
easily be the case that some classes of data are found only in the unlabelled
dataset -- perhaps the labelling process was biased -- so we do not have any
labelled examples to train on for some classes. We call this learning regime
$\textit{semi-unsupervised learning}$, an extreme case of semi-supervised
learning, where some classes have no labelled exemplars in the training set.
First, we outline the pitfalls associated with trying to apply deep generative
model (DGM)-based semi-supervised learning algorithms to datasets of this type.
We then show how a combination of clustering and semi-supervised learning,
using DGMs, can be brought to bear on this problem. We study several different
datasets, showing how one can still learn effectively when half of the ground
truth classes are entirely unlabelled and the other half are sparsely labelled.
| stat |
Quantile Multi-Armed Bandits: Optimal Best-Arm Identification and a
Differentially Private Scheme | We study the best-arm identification problem in multi-armed bandits with
stochastic, potentially private rewards, when the goal is to identify the arm
with the highest quantile at a fixed, prescribed level. First, we propose a
(non-private) successive elimination algorithm for strictly optimal best-arm
identification, we show that our algorithm is $\delta$-PAC and we characterize
its sample complexity. Further, we provide a lower bound on the expected number
of pulls, showing that the proposed algorithm is essentially optimal up to
logarithmic factors. Both upper and lower complexity bounds depend on a special
definition of the associated suboptimality gap, designed in particular for the
quantile bandit problem, as we show when the gap approaches zero, best-arm
identification is impossible. Second, motivated by applications where the
rewards are private, we provide a differentially private successive elimination
algorithm whose sample complexity is finite even for distributions with
infinite support-size, and we characterize its sample complexity. Our
algorithms do not require prior knowledge of either the suboptimality gap or
other statistical information related to the bandit problem at hand.
| stat |
Implicit Normalizing Flows | Normalizing flows define a probability distribution by an explicit invertible
transformation $\boldsymbol{\mathbf{z}}=f(\boldsymbol{\mathbf{x}})$. In this
work, we present implicit normalizing flows (ImpFlows), which generalize
normalizing flows by allowing the mapping to be implicitly defined by the roots
of an equation $F(\boldsymbol{\mathbf{z}}, \boldsymbol{\mathbf{x}})=
\boldsymbol{\mathbf{0}}$. ImpFlows build on residual flows (ResFlows) with a
proper balance between expressiveness and tractability. Through theoretical
analysis, we show that the function space of ImpFlow is strictly richer than
that of ResFlows. Furthermore, for any ResFlow with a fixed number of blocks,
there exists some function that ResFlow has a non-negligible approximation
error. However, the function is exactly representable by a single-block
ImpFlow. We propose a scalable algorithm to train and draw samples from
ImpFlows. Empirically, we evaluate ImpFlow on several classification and
density modeling tasks, and ImpFlow outperforms ResFlow with a comparable
amount of parameters on all the benchmarks.
| stat |
Robust Tensor Principal Component Analysis: Exact Recovery via
Deterministic Model | Tensor, also known as multi-dimensional array, arises from many applications
in signal processing, manufacturing processes, healthcare, among others. As one
of the most popular methods in tensor literature, Robust tensor principal
component analysis (RTPCA) is a very effective tool to extract the low rank and
sparse components in tensors. In this paper, a new method to analyze RTPCA is
proposed based on the recently developed tensor-tensor product and tensor
singular value decomposition (t-SVD). Specifically, it aims to solve a convex
optimization problem whose objective function is a weighted combination of the
tensor nuclear norm and the l1-norm. In most of literature of RTPCA, the exact
recovery is built on the tensor incoherence conditions and the assumption of a
uniform model on the sparse support. Unlike this conventional way, in this
paper, without any assumption of randomness, the exact recovery can be achieved
in a completely deterministic fashion by characterizing the tensor
rank-sparsity incoherence, which is an uncertainty principle between the
low-rank tensor spaces and the pattern of sparse tensor.
| stat |
Modeling "Equitable and Sustainable Well-being" (BES) using Bayesian
Networks: A Case Study of the Italian regions | Measurement of well-being has been a highly debated topic since the end of
the last century. While some specific aspects are still open issues, a
multidimensional approach as well as the construction of shared and well-rooted
systems of indicators are now accepted as the main route to measure this
complex phenomenon. A meaningful effort, in this direction, is that of the
Italian "Equitable and Sustainable Well-being" (BES) system of indicators,
developed by the Italian National Institute of Statistics (ISTAT) and the
National Council for Economics and Labour (CNEL). The BES framework comprises a
number of atomic indicators measured yearly at the regional level and
reflecting the different domains of well-being (e.g. Health, Education, Work \&
Life Balance, Environment,...). In this work we aim at dealing with the
multidimensionality of the BES system of indicators and try to answer three
main research questions: I) What is the structure of the relationships among
the BES atomic indicators; II) What is the structure of the relationships among
the BES domains; III) To what extent the structure of the relationships
reflects the current BES theoretical framework. We address these questions by
implementing Bayesian Networks (BNs), a widely accepted class of multivariate
statistical models, particularly suitable for handling reasoning with
uncertainty. Implementation of a BN results in a set of nodes and a set of
conditional independence statements that provide an effective tool to explore
associations in a system of variables. In this work, we also suggest two
strategies for encoding prior knowledge in the BN estimating algorithm so that
the BES theoretical framework can be represented into the network.
| stat |
Manifold GPLVMs for discovering non-Euclidean latent structure in neural
data | A common problem in neuroscience is to elucidate the collective neural
representations of behaviorally important variables such as head direction,
spatial location, upcoming movements, or mental spatial transformations. Often,
these latent variables are internal constructs not directly accessible to the
experimenter. Here, we propose a new probabilistic latent variable model to
simultaneously identify the latent state and the way each neuron contributes to
its representation in an unsupervised way. In contrast to previous models which
assume Euclidean latent spaces, we embrace the fact that latent states often
belong to symmetric manifolds such as spheres, tori, or rotation groups of
various dimensions. We therefore propose the manifold Gaussian process latent
variable model (mGPLVM), where neural responses arise from (i) a shared latent
variable living on a specific manifold, and (ii) a set of non-parametric tuning
curves determining how each neuron contributes to the representation.
Cross-validated comparisons of models with different topologies can be used to
distinguish between candidate manifolds, and variational inference enables
quantification of uncertainty. We demonstrate the validity of the approach on
several synthetic datasets, as well as on calcium recordings from the ellipsoid
body of Drosophila melanogaster and extracellular recordings from the mouse
anterodorsal thalamic nucleus. These circuits are both known to encode head
direction, and mGPLVM correctly recovers the ring topology expected from neural
populations representing a single angular variable.
| stat |
Learning big Gaussian Bayesian networks: partition, estimation, and
fusion | Structure learning of Bayesian networks has always been a challenging
problem. Nowadays, massive-size networks with thousands or more of nodes but
fewer samples frequently appear in many areas. We develop a divide-and-conquer
framework, called partition-estimation-fusion (PEF), for structure learning of
such big networks. The proposed method first partitions nodes into clusters,
then learns a subgraph on each cluster of nodes, and finally fuses all learned
subgraphs into one Bayesian network. The PEF method is designed in a flexible
way so that any structure learning method may be used in the second step to
learn a subgraph structure as either a DAG or a CPDAG. In the clustering step,
we adapt the hierarchical clustering method to automatically choose a proper
number of clusters. In the fusion step, we propose a novel hybrid method that
sequentially add edges between subgraphs. Extensive numerical experiments
demonstrate the competitive performance of our PEF method, in terms of both
speed and accuracy compared to existing methods. Our method can improve the
accuracy of structure learning by 20% or more, while reducing running time up
to two orders-of-magnitude.
| stat |
Projected P\'olya Tree | One way of defining probability distributions for circular variables
(directions in two dimensions) is to radially project probability
distributions, originally defined on $\mathbb{R}^2$, to the unit circle.
Projected distributions have proved to be useful in the study of circular and
directional data. Although any bivariate distribution can be used to produce a
projected circular model, these distributions are typically parametric. In this
article we consider a bivariate P\'olya tree on $\mathbb{R}^2$ and project it
to the unit circle to define a new Bayesian nonparametric model for circular
data. We study the properties of the proposed model, obtain its posterior
characterisation and show its performance with simulated and real datasets.
| stat |
Geostatistical Learning: Challenges and Opportunities | Statistical learning theory provides the foundation to applied machine
learning, and its various successful applications in computer vision, natural
language processing and other scientific domains. The theory, however, does not
take into account the unique challenges of performing statistical learning in
geospatial settings. For instance, it is well known that model errors cannot be
assumed to be independent and identically distributed in geospatial (a.k.a.
regionalized) variables due to spatial correlation; and trends caused by
geophysical processes lead to covariate shifts between the domain where the
model was trained and the domain where it will be applied, which in turn harm
the use of classical learning methodologies that rely on random samples of the
data. In this work, we introduce the geostatistical (transfer) learning
problem, and illustrate the challenges of learning from geospatial data by
assessing widely-used methods for estimating generalization error of learning
models, under covariate shift and spatial correlation. Experiments with
synthetic Gaussian process data as well as with real data from geophysical
surveys in New Zealand indicate that none of the methods are adequate for model
selection in a geospatial context. We provide general guidelines regarding the
choice of these methods in practice while new methods are being actively
researched.
| stat |
Uniform Deconvolution for Poisson Point Processes | We focus on the estimation of the intensity of a Poisson process in the
presence of a uniform noise. We propose a kernel-based procedure fully
calibrated in theory and practice. We show that our adaptive estimator is
optimal from the oracle and minimax points of view, and provide new lower
bounds when the intensity belongs to a Sobolev ball. By developing the
Goldenshluger-Lepski methodology in the case of deconvolution for Poisson
processes, we propose an optimal data-driven selection of the kernel's
bandwidth, and we provide a heuristic framework to calibrate the estimator in
practice. Our method is illustrated on the spatial repartition of replication
origins along the human genome.
| stat |
Hamiltonian zigzag sampler got more momentum than its Markovian
counterpart: Equivalence of two zigzags under a momentum refreshment limit | Zigzag and other piecewise deterministic Markov process samplers have
attracted significant interest for their non-reversibility and other appealing
properties for Bayesian posterior computation. Hamiltonian Monte Carlo is
another state-of-the-art sampler, exploiting fictitious momentum to guide
Markov chains through complex target distributions. In this article, we uncover
a remarkable connection between the zigzag sampler and a variant of Hamiltonian
Monte Carlo exploiting Laplace-distributed momentum. The position and velocity
component of the corresponding Hamiltonian dynamics travels along a zigzag path
paralleling the Markovian zigzag process; however, the dynamics is
non-Markovian as the momentum component encodes non-immediate pasts. This
information is partially lost during a momentum refreshment step, in which we
preserve its direction but re-sample magnitude. In the limit of increasingly
frequent momentum refreshments, we prove that this Hamiltonian zigzag converges
to its Markovian counterpart. This theoretical insight suggests that, by
retaining full momentum information, Hamiltonian zigzag can better explore
target distributions with highly correlated parameters. We corroborate this
intuition by comparing performance of the two zigzag cousins on
high-dimensional truncated multivariate Gaussians, including a
11,235-dimensional target arising from a Bayesian phylogenetic multivariate
probit model applied to HIV virus data.
| stat |
Sequential Bayesian Risk Set Inference for Robust Discrete Optimization
via Simulation | Optimization via simulation (OvS) procedures that assume the simulation
inputs are generated from the real-world distributions are subject to the risk
of selecting a suboptimal solution when the distributions are substituted with
input models estimated from finite real-world data -- known as input model
risk. Focusing on discrete OvS, this paper proposes a new Bayesian framework
for analyzing input model risk of implementing an arbitrary solution, $x$,
where uncertainty about the input models is captured by a posterior
distribution. We define the $\alpha$-level risk set of solution $x$ as the set
of solutions whose expected performance is better than $x$ by a practically
meaningful margin $(>\delta)$ given common input models with significant
probability ($>\alpha$) under the posterior distribution. The user-specified
parameters, $\delta$ and $\alpha$, control robustness of the procedure to the
desired level as well as guards against unnecessary conservatism. An empty risk
set implies that there is no practically better solution than $x$ with
significant probability even though the real-world input distributions are
unknown. For efficient estimation of the risk set, the conditional mean
performance of a solution given a set of input distributions is modeled as a
Gaussian process (GP) that takes the solution-distributions pair as an input.
In particular, our GP model allows both parametric and nonparametric input
models. We propose the sequential risk set inference procedure that estimates
the risk set and selects the next solution-distributions pair to simulate using
the posterior GP at each iteration. We show that simulating the pair expected
to change the risk set estimate the most in the next iteration is the
asymptotic one-step optimal sampling rule that minimizes the number of
incorrectly classified solutions, if the procedure runs without stopping.
| stat |
Exploring the Uncertainty Properties of Neural Networks' Implicit Priors
in the Infinite-Width Limit | Modern deep learning models have achieved great success in predictive
accuracy for many data modalities. However, their application to many
real-world tasks is restricted by poor uncertainty estimates, such as
overconfidence on out-of-distribution (OOD) data and ungraceful failing under
distributional shift. Previous benchmarks have found that ensembles of neural
networks (NNs) are typically the best calibrated models on OOD data. Inspired
by this, we leverage recent theoretical advances that characterize the
function-space prior of an ensemble of infinitely-wide NNs as a Gaussian
process, termed the neural network Gaussian process (NNGP). We use the NNGP
with a softmax link function to build a probabilistic model for multi-class
classification and marginalize over the latent Gaussian outputs to sample from
the posterior. This gives us a better understanding of the implicit prior NNs
place on function space and allows a direct comparison of the calibration of
the NNGP and its finite-width analogue. We also examine the calibration of
previous approaches to classification with the NNGP, which treat classification
problems as regression to the one-hot labels. In this case the Bayesian
posterior is exact, and we compare several heuristics to generate a categorical
distribution over classes. We find these methods are well calibrated under
distributional shift. Finally, we consider an infinite-width final layer in
conjunction with a pre-trained embedding. This replicates the important
practical use case of transfer learning and allows scaling to significantly
larger datasets. As well as achieving competitive predictive accuracy, this
approach is better calibrated than its finite width analogue.
| stat |
The FMRIB Variational Bayesian Inference Tutorial II: Stochastic
Variational Bayes | Bayesian methods have proved powerful in many applications for the inference
of model parameters from data. These methods are based on Bayes' theorem, which
itself is deceptively simple. However, in practice the computations required
are intractable even for simple cases. Hence methods for Bayesian inference
have historically either been significantly approximate, e.g., the Laplace
approximation, or achieve samples from the exact solution at significant
computational expense, e.g., Markov Chain Monte Carlo methods. Since around the
year 2000 so-called Variational approaches to Bayesian inference have been
increasingly deployed. In its most general form Variational Bayes (VB) involves
approximating the true posterior probability distribution via another more
'manageable' distribution, the aim being to achieve as good an approximation as
possible. In the original FMRIB Variational Bayes tutorial we documented an
approach to VB based that took a 'mean field' approach to forming the
approximate posterior, required the conjugacy of prior and likelihood, and
exploited the Calculus of Variations, to derive an iterative series of update
equations, akin to Expectation Maximisation. In this tutorial we revisit VB,
but now take a stochastic approach to the problem that potentially circumvents
some of the limitations imposed by the earlier methodology. This new approach
bears a lot of similarity to, and has benefited from, computational methods
applied to machine learning algorithms. Although, what we document here is
still recognisably Bayesian inference in the classic sense, and not an attempt
to use machine learning as a black-box to solve the inference problem.
| stat |
Fast Exact Computation of Expected HyperVolume Improvement | In multi-objective Bayesian optimization and surrogate-based evolutionary
algorithms, Expected HyperVolume Improvement (EHVI) is widely used as the
acquisition function to guide the search approaching the Pareto front. This
paper focuses on the exact calculation of EHVI given a nondominated set, for
which the existing exact algorithms are complex and can be inefficient for
problems with more than three objectives. Integrating with different
decomposition algorithms, we propose a new method for calculating the integral
in each decomposed high-dimensional box in constant time. We develop three new
exact EHVI calculation algorithms based on three region decomposition methods.
The first grid-based algorithm has a complexity of $O(m\cdot n^m)$ with $n$
denoting the size of the nondominated set and $m$ the number of objectives. The
Walking Fish Group (WFG)-based algorithm has a worst-case complexity of
$O(m\cdot 2^n)$ but has a better average performance. These two can be applied
for problems with any $m$. The third CLM-based algorithm is only for $m=3$ and
asymptotically optimal with complexity $\Theta(n\log{n})$. Performance
comparison results show that all our three algorithms are at least twice faster
than the state-of-the-art algorithms with the same decomposition methods. When
$m>3$, our WFG-based algorithm can be over $10^2$ faster than the corresponding
existing algorithms. Our algorithm is demonstrated in an example involving
efficient multi-objective material design with Bayesian optimization.
| stat |
Efficient estimation of optimal regimes under a no direct effect
assumption | We derive new estimators of an optimal joint testing and treatment regime
under the no direct effect (NDE) assumption that a given laboratory,
diagnostic, or screening test has no effect on a patient's clinical outcomes
except through the effect of the test results on the choice of treatment. We
model the optimal joint strategy using an optimal regime structural nested mean
model (opt-SNMM). The proposed estimators are more efficient than previous
estimators of the parameters of an opt-SNMM because they efficiently leverage
the `no direct effect (NDE) of testing' assumption. Our methods will be of
importance to decision scientists who either perform cost-benefit analyses or
are tasked with the estimation of the `value of information' supplied by an
expensive diagnostic test (such as an MRI to screen for lung cancer).
| stat |
Sparse Reduced-Rank Regression for Simultaneous Rank and Variable
Selection via Manifold Optimization | We consider the problem of constructing a reduced-rank regression model whose
coefficient parameter is represented as a singular value decomposition with
sparse singular vectors. The traditional estimation procedure for the
coefficient parameter often fails when the true rank of the parameter is high.
To overcome this issue, we develop an estimation algorithm with rank and
variable selection via sparse regularization and manifold optimization, which
enables us to obtain an accurate estimation of the coefficient parameter even
if the true rank of the coefficient parameter is high. Using sparse
regularization, we can also select an optimal value of the rank. We conduct
Monte Carlo experiments and real data analysis to illustrate the effectiveness
of our proposed method.
| stat |
Random Persistence Diagram Generation | Topological data analysis (TDA) studies the shape patterns of data.
Persistent homology (PH) is a widely used method in TDA that summarizes
homological features of data at multiple scales and stores this in persistence
diagrams (PDs). As TDA is commonly used in the analysis of high dimensional
data sets, a sufficiently large amount of PDs that allow performing statistical
analysis is typically unavailable or requires inordinate computational
resources. In this paper, we propose random persistence diagram generation
(RPDG), a method that generates a sequence of random PDs from the ones produced
by the data. RPDG is underpinned (i) by a parametric model based on pairwise
interacting point processes for inference of persistence diagrams and (ii) by a
reversible jump Markov chain Monte Carlo (RJ-MCMC) algorithm for generating
samples of PDs. The parametric model combines a Dirichlet partition to capture
spatial homogeneity of the location of points in a PD and a step function to
capture the pairwise interaction between them. The RJ-MCMC algorithm
incorporates trans-dimensional addition and removal of points and
same-dimensional relocation of points across samples of PDs. The efficacy of
RPDG is demonstrated via an example and a detailed comparison with other
existing methods is presented.
| stat |
A New Distribution-Free Concept for Representing, Comparing, and
Propagating Uncertainty in Dynamical Systems with Kernel Probabilistic
Programming | This work presents the concept of kernel mean embedding and kernel
probabilistic programming in the context of stochastic systems. We propose
formulations to represent, compare, and propagate uncertainties for fairly
general stochastic dynamics in a distribution-free manner. The new tools enjoy
sound theory rooted in functional analysis and wide applicability as
demonstrated in distinct numerical examples. The implication of this new
concept is a new mode of thinking about the statistical nature of uncertainty
in dynamical systems.
| stat |
Robust Inference for Mediated Effects in Partially Linear Models | We consider mediated effects of an exposure, X on an outcome, Y, via a
mediator, M, under no unmeasured confounding assumptions in the setting where
models for the conditional expectation of the mediator and outcome are
partially linear. We propose G-estimators for the direct and indirect effect
and demonstrate consistent asymptotic normality for indirect effects when
models for the conditional means of M, or X and Y are correctly specified, and
for direct effects, when models for the conditional means of Y, or X and M are
correct. This marks an improvement, in this particular setting, over previous
`triple' robust methods, which do not assume partially linear mean models.
Testing of the no-mediation hypothesis is inherently problematic due to the
composite nature of the test (either X has no effect on M or M no effect on Y),
leading to low power when both effect sizes are small. We use Generalized
Methods of Moments (GMM) results to construct a new score testing framework,
which includes as special cases the no-mediation and the no-direct-effect
hypotheses. The proposed tests rely on an orthogonal estimation strategy for
estimating nuisance parameters. Simulations show that the GMM based tests
perform better in terms of power and small sample performance compared with
traditional tests in the partially linear setting, with drastic improvement
under model misspecification. New methods are illustrated in a mediation
analysis of data from the COPERS trial, a randomized trial investigating the
effect of a non-pharmacological intervention of patients suffering from chronic
pain. An accompanying R package implementing these methods can be found at
github.com/ohines/plmed.
| stat |
The Spatially-Conscious Machine Learning Model | Successfully predicting gentrification could have many social and commercial
applications; however, real estate sales are difficult to predict because they
belong to a chaotic system comprised of intrinsic and extrinsic
characteristics, perceived value, and market speculation. Using New York City
real estate as our subject, we combine modern techniques of data science and
machine learning with traditional spatial analysis to create robust real estate
prediction models for both classification and regression tasks. We compare
several cutting edge machine learning algorithms across spatial, semi-spatial
and non-spatial feature engineering techniques, and we empirically show that
spatially-conscious machine learning models outperform non-spatial models when
married with advanced prediction techniques such as feed-forward artificial
neural networks and gradient boosting machine models.
| stat |
Posterior Consistency of Semi-Supervised Regression on Graphs | Graph-based semi-supervised regression (SSR) is the problem of estimating the
value of a function on a weighted graph from its values (labels) on a small
subset of the vertices. This paper is concerned with the consistency of SSR in
the context of classification, in the setting where the labels have small noise
and the underlying graph weighting is consistent with well-clustered nodes. We
present a Bayesian formulation of SSR in which the weighted graph defines a
Gaussian prior, using a graph Laplacian, and the labeled data defines a
likelihood. We analyze the rate of contraction of the posterior measure around
the ground truth in terms of parameters that quantify the small label error and
inherent clustering in the graph. We obtain bounds on the rates of contraction
and illustrate their sharpness through numerical experiments. The analysis also
gives insight into the choice of hyperparameters that enter the definition of
the prior.
| stat |
On cointegration for modeling and forecasting wind power production | This study evaluates the performance of cointegrated vector autoregressive
(VAR) models for very short- and short-term wind power forecasting. Preliminary
results for a German data set comprising six wind power production time series
indicate that taking into account potential cointegrating relations between the
individual series can improve forecasts at short-term time horizons.
| stat |
Looking Deeper into Tabular LIME | Interpretability of machine learning algorithms is an urgent need. Numerous
methods appeared in recent years, but do their explanations make sense? In this
paper, we present a thorough theoretical analysis of one of these methods,
LIME, in the case of tabular data. We prove that in the large sample limit, the
interpretable coefficients provided by Tabular LIME can be computed in an
explicit way as a function of the algorithm parameters and some expectation
computations related to the black-box model. When the function to explain has
some nice algebraic structure (linear, multiplicative, or sparsely depending on
a subset of the coordinates), our analysis provides interesting insights into
the explanations provided by LIME. These can be applied to a range of machine
learning models including Gaussian kernels or CART random forests. As an
example, for linear functions we show that LIME has the desirable property to
provide explanations that are proportional to the coefficients of the function
to explain and to ignore coordinates that are not used by the function to
explain. For partition-based regressors, on the other side, we show that LIME
produces undesired artifacts that may provide misleading explanations.
| stat |
Tutorial: Effective visual communication for the quantitative scientist | Effective visual communication is a core competency for pharmacometricians,
statisticians, and more generally any quantitative scientist. It is essential
in every step of a quantitative workflow, from scoping to execution and
communicating results and conclusions. With this competency, we can better
understand data and influence decisions towards appropriate actions. Without
it, we can fool ourselves and others and pave the way to wrong conclusions and
actions. The goal of this tutorial is to convey this competency. We posit three
laws of effective visual communication for the quantitative scientist: have a
clear purpose, show the data clearly, and make the message obvious. A concise
"Cheat Sheet", available on https://graphicsprinciples.github.io, distills more
granular recommendations for everyday practical use. Finally, these laws and
recommendations are illustrated in four case studies.
| stat |
Model Bridging: Connection between Simulation Model and Neural Network | The interpretability of machine learning, particularly for deep neural
networks, is crucial for decision making in real-world applications. One
approach is replacing the un-interpretable machine learning model with a
surrogate model, which has a simple structure for interpretation. Another
approach is understanding the target system by using a simulation modeled by
human knowledge with interpretable simulation parameters. Recently, simulator
calibration has been developed based on kernel mean embedding to estimate the
simulation parameters as posterior distributions. Our idea is to use a
simulation model as an interpretable surrogate model. However, the
computational cost of simulator calibration is high owing to the complexity of
the simulation model. Thus, we propose a ''model-bridging'' framework to bridge
machine learning models with simulation models by a series of kernel mean
embeddings to address these difficulties. The proposed framework enables us to
obtain predictions and interpretable simulation parameters simultaneously
without the computationally expensive calculations of the simulations. In this
study, we apply the proposed framework to essential simulations in the
manufacturing industry, such as production simulation and fluid dynamics
simulation.
| stat |
LOCUS: A Novel Decomposition Method for Brain Network Connectivity
Matrices using Low-rank Structure with Uniform Sparsity | Network-oriented research has been increasingly popular in many scientific
areas. In neuroscience research, imaging-based network connectivity measures
have become the key for understanding brain organizations, potentially serving
as individual neural fingerprints. There are major challenges in analyzing
connectivity matrices including the high dimensionality of brain networks,
unknown latent sources underlying the observed connectivity, and the large
number of brain connections leading to spurious findings. In this paper, we
propose a novel blind source separation method with low-rank structure and
uniform sparsity (LOCUS) as a fully data-driven decomposition method for
network measures. Compared with the existing method that vectorizes
connectivity matrices ignoring brain network topology, LOCUS achieves more
efficient and accurate source separation for connectivity matrices using
low-rank structure. We propose a novel angle-based uniform sparsity
regularization that demonstrates better performance than the existing sparsity
controls for low-rank tensor methods. We propose a highly efficient iterative
Node-Rotation algorithm that exploits the block multi-convexity of the
objective function to solve the non-convex optimization problem for learning
LOCUS. We illustrate the advantage of LOCUS through extensive simulation
studies. Application of LOCUS to Philadelphia Neurodevelopmental Cohort
neuroimaging study reveals biologically insightful connectivity traits which
are not found using the existing method.
| stat |
Kernel Selection for Modal Linear Regression: Optimal Kernel and IRLS
Algorithm | Modal linear regression (MLR) is a method for obtaining a conditional mode
predictor as a linear model. We study kernel selection for MLR from two
perspectives: "which kernel achieves smaller error?" and "which kernel is
computationally efficient?". First, we show that a Biweight kernel is optimal
in the sense of minimizing an asymptotic mean squared error of a resulting MLR
parameter. This result is derived from our refined analysis of an asymptotic
statistical behavior of MLR. Secondly, we provide a kernel class for which
iteratively reweighted least-squares algorithm (IRLS) is guaranteed to
converge, and especially prove that IRLS with an Epanechnikov kernel terminates
in a finite number of iterations. Simulation studies empirically verified that
using a Biweight kernel provides good estimation accuracy and that using an
Epanechnikov kernel is computationally efficient. Our results improve MLR of
which existing studies often stick to a Gaussian kernel and modal EM algorithm
specialized for it, by providing guidelines of kernel selection.
| stat |
Approximating predictive probabilities of Gibbs-type priors | Gibbs-type random probability measures, or Gibbs-type priors, are arguably
the most "natural" generalization of the celebrated Dirichlet prior. Among them
the two parameter Poisson-Dirichlet prior certainly stands out for the
mathematical tractability and interpretability of its predictive probabilities,
which made it the natural candidate in several applications. Given a sample of
size $n$, in this paper we show that the predictive probabilities of any
Gibbs-type prior admit a large $n$ approximation, with an error term vanishing
as $o(1/n)$, which maintains the same desirable features as the predictive
probabilities of the two parameter Poisson-Dirichlet prior.
| stat |
Machine learning for total cloud cover prediction | Accurate and reliable forecasting of total cloud cover (TCC) is vital for
many areas such as astronomy, energy demand and production, or agriculture.
Most meteorological centres issue ensemble forecasts of TCC, however, these
forecasts are often uncalibrated and exhibit worse forecast skill than ensemble
forecasts of other weather variables. Hence, some form of post-processing is
strongly required to improve predictive performance. As TCC observations are
usually reported on a discrete scale taking just nine different values called
oktas, statistical calibration of TCC ensemble forecasts can be considered a
classification problem with outputs given by the probabilities of the oktas.
This is a classical area where machine learning methods are applied. We
investigate the performance of post-processing using multilayer perceptron
(MLP) neural networks, gradient boosting machines (GBM) and random forest (RF)
methods. Based on the European Centre for Medium-Range Weather Forecasts global
TCC ensemble forecasts for 2002-2014 we compare these approaches with the
proportional odds logistic regression (POLR) and multiclass logistic regression
(MLR) models, as well as the raw TCC ensemble forecasts. We further assess
whether improvements in forecast skill can be obtained by incorporating
ensemble forecasts of precipitation as additional predictor. Compared to the
raw ensemble, all calibration methods result in a significant improvement in
forecast skill. RF models provide the smallest increase in predictive
performance, while MLP, POLR and GBM approaches perform best. The use of
precipitation forecast data leads to further improvements in forecast skill and
except for very short lead times the extended MLP model shows the best overall
performance.
| stat |
Contrastive Learning of Structured World Models | A structured understanding of our world in terms of objects, relations, and
hierarchies is an important component of human cognition. Learning such a
structured world model from raw sensory data remains a challenge. As a step
towards this goal, we introduce Contrastively-trained Structured World Models
(C-SWMs). C-SWMs utilize a contrastive approach for representation learning in
environments with compositional structure. We structure each state embedding as
a set of object representations and their relations, modeled by a graph neural
network. This allows objects to be discovered from raw pixel observations
without direct supervision as part of the learning process. We evaluate C-SWMs
on compositional environments involving multiple interacting objects that can
be manipulated independently by an agent, simple Atari games, and a
multi-object physics simulation. Our experiments demonstrate that C-SWMs can
overcome limitations of models based on pixel reconstruction and outperform
typical representatives of this model class in highly structured environments,
while learning interpretable object-based representations.
| stat |
Layer-wise Learning of Kernel Dependence Networks | Due to recent debate over the biological plausibility of backpropagation
(BP), finding an alternative network optimization strategy has become an active
area of interest. We design a new type of kernel network, that is solved
greedily, to theoretically answer several questions of interest. First, if BP
is difficult to simulate in the brain, are there instead "trivial network
weights" (requiring minimum computation) that allow a greedily trained network
to classify any pattern. Perhaps a simple repetition of some basic rule can
yield a network equally powerful as ones trained by BP with Stochastic Gradient
Descent (SGD). Second, can a greedily trained network converge to a kernel?
What kernel will it converge to? Third, is this trivial solution optimal? How
is the optimal solution related to generalization? Lastly, can we theoretically
identify the network width and depth without a grid search? We prove that the
kernel embedding is the trivial solution that compels the greedy procedure to
converge to a kernel with Universal property. Yet, this trivial solution is not
even optimal. By obtaining the optimal solution spectrally, it provides insight
into the generalization of the network while informing us of the network width
and depth.
| stat |
Statistical Modeling for Spatio-Temporal Data from Stochastic
Convection-Diffusion Processes | This paper proposes a physical-statistical modeling approach for
spatio-temporal data arising from a class of stochastic convection-diffusion
processes. Such processes are widely found in scientific and engineering
applications where fundamental physics imposes critical constraints on how data
can be modeled and how models should be interpreted. The idea of spectrum
decomposition is employed to approximate a physical spatio-temporal process by
the linear combination of spatial basis functions and a multivariate random
process of spectral coefficients. Unlike existing approaches assuming
spatially- and temporally-invariant convection-diffusion, this paper considers
a more general scenario with spatially-varying convection-diffusion and
nonzero-mean source-sink. As a result, the temporal dynamics of spectral
coefficients is coupled with each other, which can be interpreted as the
non-linear energy redistribution across multiple scales from the perspective of
physics. Because of the spatially-varying convection-diffusion, the space-time
covariance is non-stationary in space. The theoretical results are integrated
into a hierarchical dynamical spatio-temporal model. The connection is
established between the proposed model and the existing models based on
Integro-Difference Equations. Computational efficiency and scalability are also
investigated to make the proposed approach practical. The advantages of the
proposed methodology are demonstrated by numerical examples, a case study, and
comprehensive comparison studies. Computer code is available on GitHub.
| stat |
Convergence of Sparse Variational Inference in Gaussian Processes
Regression | Gaussian processes are distributions over functions that are versatile and
mathematically convenient priors in Bayesian modelling. However, their use is
often impeded for data with large numbers of observations, $N$, due to the
cubic (in $N$) cost of matrix operations used in exact inference. Many
solutions have been proposed that rely on $M \ll N$ inducing variables to form
an approximation at a cost of $\mathcal{O}(NM^2)$. While the computational cost
appears linear in $N$, the true complexity depends on how $M$ must scale with
$N$ to ensure a certain quality of the approximation. In this work, we
investigate upper and lower bounds on how $M$ needs to grow with $N$ to ensure
high quality approximations. We show that we can make the KL-divergence between
the approximate model and the exact posterior arbitrarily small for a
Gaussian-noise regression model with $M\ll N$. Specifically, for the popular
squared exponential kernel and $D$-dimensional Gaussian distributed covariates,
$M=\mathcal{O}((\log N)^D)$ suffice and a method with an overall computational
cost of $\mathcal{O}(N(\log N)^{2D}(\log\log N)^2)$ can be used to perform
inference.
| stat |
Adversarial Self-Paced Learning for Mixture Models of Hawkes Processes | We propose a novel adversarial learning strategy for mixture models of Hawkes
processes, leveraging data augmentation techniques of Hawkes process in the
framework of self-paced learning. Instead of learning a mixture model directly
from a set of event sequences drawn from different Hawkes processes, the
proposed method learns the target model iteratively, which generates "easy"
sequences and uses them in an adversarial and self-paced manner. In each
iteration, we first generate a set of augmented sequences from original
observed sequences. Based on the fact that an easy sample of the target model
can be an adversarial sample of a misspecified model, we apply a maximum
likelihood estimation with an adversarial self-paced mechanism. In this manner
the target model is updated, and the augmented sequences that obey it are
employed for the next learning iteration. Experimental results show that the
proposed method outperforms traditional methods consistently.
| stat |
Good Initializations of Variational Bayes for Deep Models | Stochastic variational inference is an established way to carry out
approximate Bayesian inference for deep models. While there have been effective
proposals for good initializations for loss minimization in deep learning, far
less attention has been devoted to the issue of initialization of stochastic
variational inference. We address this by proposing a novel layer-wise
initialization strategy based on Bayesian linear models. The proposed method is
extensively validated on regression and classification tasks, including
Bayesian DeepNets and ConvNets, showing faster and better convergence compared
to alternatives inspired by the literature on initializations for loss
minimization.
| stat |
XGBoostLSS -- An extension of XGBoost to probabilistic forecasting | We propose a new framework of XGBoost that predicts the entire conditional
distribution of a univariate response variable. In particular, XGBoostLSS
models all moments of a parametric distribution (i.e., mean, location, scale
and shape [LSS]) instead of the conditional mean only. Choosing from a wide
range of continuous, discrete and mixed discrete-continuous distribution,
modelling and predicting the entire conditional distribution greatly enhances
the flexibility of XGBoost, as it allows to gain additional insight into the
data generating process, as well as to create probabilistic forecasts from
which prediction intervals and quantiles of interest can be derived. We present
both a simulation study and real world examples that demonstrate the virtues of
our approach.
| stat |
Nearly Minimax-Optimal Regret for Linearly Parameterized Bandits | We study the linear contextual bandit problem with finite action sets. When
the problem dimension is $d$, the time horizon is $T$, and there are $n \leq
2^{d/2}$ candidate actions per time period, we (1) show that the minimax
expected regret is $\Omega(\sqrt{dT (\log T) (\log n)})$ for every algorithm,
and (2) introduce a Variable-Confidence-Level (VCL) SupLinUCB algorithm whose
regret matches the lower bound up to iterated logarithmic factors. Our
algorithmic result saves two $\sqrt{\log T}$ factors from previous analysis,
and our information-theoretical lower bound also improves previous results by
one $\sqrt{\log T}$ factor, revealing a regret scaling quite different from
classical multi-armed bandits in which no logarithmic $T$ term is present in
minimax regret. Our proof techniques include variable confidence levels and a
careful analysis of layer sizes of SupLinUCB on the upper bound side, and
delicately constructed adversarial sequences showing the tightness of
elliptical potential lemmas on the lower bound side.
| stat |
Estimating Mixed Memberships with Sharp Eigenvector Deviations | We consider the problem of estimating community memberships of nodes in a
network, where every node is associated with a vector determining its degree of
membership in each community. Existing provably consistent algorithms often
require strong assumptions about the population, are computationally expensive,
and only provide an overall error bound for the whole community membership
matrix. This paper provides uniform rates of convergence for the inferred
community membership vector of each node in a network generated from the Mixed
Membership Stochastic Blockmodel (MMSB); to our knowledge, this is the first
work to establish per-node rates for overlapping community detection in
networks. We achieve this by establishing sharp row-wise eigenvector deviation
bounds for MMSB. Based on the simplex structure inherent in the
eigen-decomposition of the population matrix, we build on established
corner-finding algorithms from the optimization community to infer the
community membership vectors. Our results hold over a broad parameter regime
where the average degree only grows poly-logarithmically with the number of
nodes. Using experiments with simulated and real datasets, we show that our
method achieves better error with lower variability over competing methods, and
processes real world networks of up to 100,000 nodes within tens of seconds.
| stat |
Real-time prediction of severe influenza epidemics using Extreme Value
Statistics | Each year, seasonal influenza epidemics cause hundreds of thousands of deaths
worldwide and put high loads on health care systems. A main concern for
resource planning is the risk of exceptionnally severe epidemics. Taking
advantage of the weekly influenza cases reporting in France, we use recent
results on multivariate GP models in Extreme Value Statistics to develop
methods for real-time prediction of the risk that an ongoing epidemic will be
exceptionally severe and for real-time detection of anomalous epidemics.
Quality of predictions is assessed on observed and simulated data.
| stat |
Double-robust and efficient methods for estimating the causal effects of
a binary treatment | We consider the problem of estimating the effects of a binary treatment on a
continuous outcome of interest from observational data in the absence of
confounding by unmeasured factors. We provide a new estimator of the population
average treatment effect (ATE) based on the difference of novel double-robust
(DR) estimators of the treatment-specific outcome means. We compare our new
estimator with previously estimators both theoretically and via simulation.
DR-difference estimators may have poor finite sample behavior when the
estimated propensity scores in the treated and untreated do not overlap. We
therefore propose an alternative approach, which can be used even in this
unfavorable setting, based on locally efficient double-robust estimation of a
semiparametric regression model for the modification on an additive scale of
the magnitude of the treatment effect by the baseline covariates $X$. In
contrast with existing methods, our approach simultaneously provides estimates
of: i) the average treatment effect in the total study population, ii) the
average treatment effect in the random subset of the population with
overlapping estimated propensity scores, and iii) the treatment effect at each
level of the baseline covariates $X$.
When the covariate vector $X$ is high dimensional, one cannot be certain,
owing to lack of power, that given models for the propensity score and for the
regression of the outcome on treatment and $X$ used in constructing our DR
estimators are nearly correct, even if they pass standard goodness of fit
tests. Therefore to select among candidate models, we propose a novel approach
to model selection that leverages the DR-nature of our treatment effect
estimator and that outperforms cross-validation in a small simulation study.
| stat |
Hi3+3: A Model-Assisted Dose-Finding Design Borrowing Historical Data | Background -- In phase I clinical trials, historical data may be available
through multi-regional programs, reformulation of the same drug, or previous
trials for a drug under the same class. Statistical designs that borrow
information from historical data can reduce cost, speed up drug development,
and maintain safety. Purpose -- Based on a hybrid design that partly uses
probability models and partly uses algorithmic rules for decision making, we
aim to improve the efficiency of the dose-finding trials in the presence of
historical data, maintain safety for patients, and achieve a level of
simplicity for practical applications. Methods -- We propose the Hi3+3 design,
in which the letter "H" represents "historical data". We apply the idea in
power prior to borrow historical data and define the effective sample size
(ESS) of the prior. Dose-finding decision rules follow the idea in the i3+3
design while incorporating the historical data via the power prior and ESS. The
proposed Hi3+3 design pretabulates the dosing decisions before the trial
starts, a desirable feature for ease of application in practice. Results -- The
Hi3+3 design is superior than the i3+3 design due to information borrow from
historical data. It is capable of maintaining a high level of safety for trial
patients without sacrificing the ability to identify the correct MTD.
Illustration of this feature are found in the simulation results. Conclusion --
With the demonstrated safety, efficiency, and simplicity, the Hi3+3 design
could be a desirable choice for dose-finding trials borrowing historical data.
| stat |
On Racial Disparities in Recent Fatal Police Shootings | Fatal police shootings in the United States continue to be a polarizing
social and political issue. Clear disagreement between racial proportions of
victims and nationwide racial demographics together with graphic video footage
has created fertile ground for controversy. However, simple population level
summary statistics fail to take into account fundamental local characteristics
such as county-level racial demography, local arrest demography, and law
enforcement density. Utilizing data on fatal police shootings between January
2015 and July 2016, we implement a number of straightforward resampling
procedures designed to carefully examine how unlikely the victim totals from
each race are with respect to these local population characteristics if no
racial bias were present in the decision to shoot by police. We present several
approaches considering the shooting locations both as fixed and also as a
random sample. In both cases, we find overwhelming evidence of a racial
disparity in shooting victims with respect to local population demographics but
substantially less disparity after accounting for local arrest demographics. We
conclude our analyses by examining the effect of police-worn body cameras and
find no evidence that the presence of such cameras impacts the racial
distribution of victims.
| stat |
Global and Local Two-Sample Tests via Regression | Two-sample testing is a fundamental problem in statistics. Despite its long
history, there has been renewed interest in this problem with the advent of
high-dimensional and complex data. Specifically, in the machine learning
literature, there have been recent methodological developments such as
classification accuracy tests. The goal of this work is to present a regression
approach to comparing multivariate distributions of complex data. Depending on
the chosen regression model, our framework can efficiently handle different
types of variables and various structures in the data, with competitive power
under many practical scenarios. Whereas previous work has been largely limited
to global tests which conceal much of the local information, our approach
naturally leads to a local two-sample testing framework in which we identify
local differences between multivariate distributions with statistical
confidence. We demonstrate the efficacy of our approach both theoretically and
empirically, under some well-known parametric and nonparametric regression
methods. Our proposed methods are applied to simulated data as well as a
challenging astronomy data set to assess their practical usefulness.
| stat |
On the relationship between a Gamma distributed precision parameter and
the associated standard deviation in the context of Bayesian parameter
inference | In Bayesian inference, an unknown measurement uncertainty is often quantified
in terms of a Gamma distributed precision parameter, which is impractical when
prior information on the standard deviation of the measurement uncertainty
shall be utilised during inference. This paper thus introduces a method for
transforming between a gamma distributed precision parameter and the
distribution of the associated standard deviation. The proposed method is based
on numerical optimisation and shows adequate results for a wide range of
scenarios.
| stat |
Taking Advantage of Multitask Learning for Fair Classification | A central goal of algorithmic fairness is to reduce bias in automated
decision making. An unavoidable tension exists between accuracy gains obtained
by using sensitive information (e.g., gender or ethnic group) as part of a
statistical model, and any commitment to protect these characteristics. Often,
due to biases present in the data, using the sensitive information in the
functional form of a classifier improves classification accuracy. In this paper
we show how it is possible to get the best of both worlds: optimize model
accuracy and fairness without explicitly using the sensitive feature in the
functional form of the model, thereby treating different individuals equally.
Our method is based on two key ideas. On the one hand, we propose to use
Multitask Learning (MTL), enhanced with fairness constraints, to jointly learn
group specific classifiers that leverage information between sensitive groups.
On the other hand, since learning group specific models might not be permitted,
we propose to first predict the sensitive features by any learning method and
then to use the predicted sensitive feature to train MTL with fairness
constraints. This enables us to tackle fairness with a three-pronged approach,
that is, by increasing accuracy on each group, enforcing measures of fairness
during training, and protecting sensitive information during testing.
Experimental results on two real datasets support our proposal, showing
substantial improvements in both accuracy and fairness.
| stat |
RADIOHEAD: Radiogenomic Analysis Incorporating Tumor Heterogeneity in
Imaging Through Densities | Recent technological advancements have enabled detailed investigation of
associations between the molecular architecture and tumor heterogeneity,
through multi-source integration of radiological imaging and genomic
(radiogenomic) data. In this paper, we integrate and harness radiogenomic data
in patients with lower grade gliomas (LGG), a type of brain cancer, in order to
develop a regression framework called RADIOHEAD (RADIOgenomic analysis
incorporating tumor HEterogeneity in imAging through Densities) to identify
radiogenomic associations. Imaging data is represented through voxel intensity
probability density functions of tumor sub-regions obtained from multimodal
magnetic resonance imaging, and genomic data through molecular signatures in
the form of pathway enrichment scores corresponding to their gene expression
profiles. Employing a Riemannian-geometric framework for principal component
analysis on the set of probability densities functions, we map each probability
density to a vector of principal component scores, which are then included as
predictors in a Bayesian regression model with the pathway enrichment scores as
the response. Variable selection compatible with the grouping structure amongst
the predictors induced through the tumor sub-regions is carried out under a
group spike-and-slab prior. A Bayesian false discovery rate mechanism is then
used to infer significant associations based on the posterior distribution of
the regression coefficients. Our analyses reveal several pathways relevant to
LGG etiology (such as synaptic transmission, nerve impulse and neurotransmitter
pathways), to have significant associations with the corresponding
imaging-based predictors.
| 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.