title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Customer Price Sensitivities in Competitive Automobile Insurance Markets | Insurers are increasingly adopting more demand-based strategies to
incorporate the indirect effect of premium changes on their policyholders'
willingness to stay. However, since in practice both insurers' renewal premia
and customers' responses to these premia typically depend on the customer's
level of risk, it remains challenging in these strategies to determine how to
properly control for this confounding. We therefore consider a causal inference
approach in this paper to account for customer price sensitivities and to
deduce optimal, multi-period profit maximizing premium renewal offers. More
specifically, we extend the discrete treatment framework of Guelman and
Guill\'en (2014) by Extreme Gradient Boosting, or XGBoost, and by multiple
imputation to better account for the uncertainty in the counterfactual
responses. We additionally introduce the continuous treatment framework with
XGBoost to the insurance literature to allow identification of the exact
optimal renewal offers and account for any competition in the market by
including competitor offers. The application of the two treatment frameworks to
a Dutch automobile insurance portfolio suggests that a policy's competitiveness
in the market is crucial for a customer's price sensitivity and that XGBoost is
more appropriate to describe this than the traditional logistic regression.
Moreover, an efficient frontier of both frameworks indicates that substantially
more profit can be gained on the portfolio than realized, also already with
less churn and in particular if we allow for continuous rate changes. A
multi-period renewal optimization confirms these findings and demonstrates that
the competitiveness enables temporal feedback of previous rate changes on
future demand.
| stat |
Goodness (of fit) of Imputation Accuracy: The GoodImpact Analysis | In statistical survey analysis, (partial) non-responders are integral
elements during data acquisition. Treating missing values during data
preparation and data analysis is therefore a non-trivial underpinning. Focusing
on different data sets from the Federal Statistical Office of Germany
(DESTATIS), we investigate various imputation methods regarding their
imputation accuracy. Since the latter is not uniquely determined in theory and
practice, we study different measures for assessing imputation accuracy: Beyond
the most common measures, the normalized-root mean squared error (NRMSE) and
the proportion of false classification (PFC), we put a special focus on
(distribution) distance- and association measures for assessing imputation
accuracy. The aim is to deliver guidelines for correctly assessing
distributional accuracy after imputation. Our empirical findings indicate a
discrepancy between the NRMSE resp. PFC and distance measures. While the latter
measure distributional similarities, NRMSE and PFC focus on data
reproducibility. We realize that a low NRMSE or PFC seem not to imply lower
distributional discrepancies. Although several measures for assessing
distributional discrepancies exist, our results indicate that not all of them
are suitable for evaluating imputation-induced differences.
| stat |
Safety-margin-based design and redesign considering mixed epistemic
model uncertainty and aleatory parameter uncertainty | At the initial design stage engineers often rely on low-fidelity models that
have high epistemic uncertainty. Traditional safety-margin-based deterministic
design resorts to testing (e.g. prototype experiment, evaluation of
high-fidelity simulation, etc.) to reduce epistemic uncertainty and achieve
targeted levels of safety. Testing is used to calibrate models and prescribe
redesign when tests are not passed. After calibration, reduced epistemic model
uncertainty can be leveraged through redesign to restore safety or improve
design performance; however, redesign may be associated with substantial costs
or delays. In this paper, a methodology is described for optimizing the
safety-margin-based design, testing, and redesign process to allow the designer
to tradeoff between the risk of future redesign and the possible performance
and reliability benefits. The proposed methodology represents the epistemic
model uncertainty with a Kriging surrogate and is applicable in a wide range of
design problems. The method is illustrated on a cantilever beam bending example
and then a sounding rocket example. It is shown that redesign acts as a type of
quality control measure to prevent an undesirable initial design from being
accepted as the final design. It is found that the optimal design/redesign
strategy for maximizing expected design performance includes not only redesign
to correct an initial design that is later revealed to be unsafe, but also
redesign to improve performance when an initial design is later revealed to be
too conservative (e.g. too heavy).
| stat |
Quantile Tracking in Dynamically Varying Data Streams Using a
Generalized Exponentially Weighted Average of Observations | The Exponentially Weighted Average (EWA) of observations is known to be
state-of-art estimator for tracking expectations of dynamically varying data
stream distributions. However, how to devise an EWA estimator to rather track
quantiles of data stream distributions is not obvious. In this paper, we
present a lightweight quantile estimator using a generalized form of the EWA.
To the best of our knowledge, this work represents the first reported quantile
estimator of this form in the literature. An appealing property of the
estimator is that the update step size is adjusted online proportionally to the
difference between current observation and the current quantile estimate. Thus,
if the estimator is off-track compared to the data stream, large steps will be
taken to promptly get the estimator back on-track. The convergence of the
estimator to the true quantile is proven using the theory of stochastic
learning.
Extensive experimental results using both synthetic and real-life data show
that our estimator clearly outperforms legacy state-of-the-art quantile
tracking estimators and achieves faster adaptivity in dynamic environments. The
quantile estimator was further tested on real-life data where the objective is
efficient online control of indoor climate. We show that the estimator can be
incorporated into a concept drift detector for efficiently decide when a
machine learning model used to predict future indoor temperature should be
retrained/updated.
| stat |
Non-Gaussian processes and neural networks at finite widths | Gaussian processes are ubiquitous in nature and engineering. A case in point
is a class of neural networks in the infinite-width limit, whose priors
correspond to Gaussian processes. Here we perturbatively extend this
correspondence to finite-width neural networks, yielding non-Gaussian processes
as priors. The methodology developed herein allows us to track the flow of
preactivation distributions by progressively integrating out random variables
from lower to higher layers, reminiscent of renormalization-group flow. We
further develop a perturbative procedure to perform Bayesian inference with
weakly non-Gaussian priors.
| stat |
Instrumental Variables: To Strengthen or Not to Strengthen? | Instrumental variables (IVs) are extensively used to estimate treatment
effects when the treatment and outcome are confounded by unmeasured
confounders; however, weak IVs are often encountered in empirical studies and
may cause problems. Many studies have considered building a stronger IV from
the original, possibly weak, IV in the design stage of a matched study at the
cost of not using some of the samples in the analysis. It is widely accepted
that strengthening an IV tends to render nonparametric tests more powerful and
will increase the power of sensitivity analyses in large samples. In this
article, we re-evaluate this conventional wisdom to bring new insights into
this topic. We consider matched observational studies from three perspectives.
First, we evaluate the trade-off between IV strength and sample size on
nonparametric tests assuming the IV is valid and exhibit conditions under which
strengthening an IV increases power and conversely conditions under which it
decreases power. Second, we derive a necessary condition for a valid
sensitivity analysis model with continuous doses. We show that the $\Gamma$
sensitivity analysis model, which has been previously used to come to the
conclusion that strengthening an IV increases the power of sensitivity analyses
in large samples, does not apply to the continuous IV setting and thus this
previously reached conclusion may be invalid. Third, we quantify the bias of
the Wald estimator with a possibly invalid IV under an oracle and leverage it
to develop a valid sensitivity analysis framework; under this framework, we
show that strengthening an IV may amplify or mitigate the bias of the
estimator, and may or may not increase the power of sensitivity analyses. We
also discuss how to better adjust for the observed covariates when building an
IV in matched studies.
| stat |
Variational inference for neural network matrix factorization and its
application to stochastic blockmodeling | We consider the probabilistic analogue to neural network matrix factorization
(Dziugaite & Roy, 2015), which we construct with Bayesian neural networks and
fit with variational inference. We find that a linear model fit with
variational inference can attain equivalent predictive performance to the
regular neural network variants on the Movielens data sets. We discuss the
implications of this result, which include some suggestions on the pros and
cons of using the neural network construction, as well as the variational
approach to inference. Such a probabilistic approach is required, however, when
considering the important class of stochastic block models. We describe a
variational inference algorithm for a neural network matrix factorization model
with nonparametric block structure and evaluate its performance on the NIPS
co-authorship data set.
| stat |
Measure Inducing Classification and Regression Trees for Functional Data | We propose a tree-based algorithm for classification and regression problems
in the context of functional data analysis, which allows to leverage
representation learning and multiple splitting rules at the node level,
reducing generalization error while retaining the interpretability of a tree.
This is achieved by learning a weighted functional $L^{2}$ space by means of
constrained convex optimization, which is then used to extract multiple
weighted integral features from the input functions, in order to determine the
binary split for each internal node of the tree. The approach is designed to
manage multiple functional inputs and/or outputs, by defining suitable
splitting rules and loss functions that can depend on the specific problem and
can also be combined with scalar and categorical data, as the tree is grown
with the original greedy CART algorithm. We focus on the case of scalar-valued
functional inputs defined on unidimensional domains and illustrate the
effectiveness of our method in both classification and regression tasks,
through a simulation study and four real world applications.
| stat |
A New Standard for the Analysis and Design of Replication Studies | A new standard is proposed for the evidential assessment of replication
studies. The approach combines a specific reverse-Bayes technique with
prior-predictive tail probabilities to define replication success. The method
gives rise to a quantitative measure for replication success, called the
sceptical p-value. The sceptical p-value integrates traditional significance of
both the original and replication study with a comparison of the respective
effect sizes. It incorporates the uncertainty of both the original and
replication effect estimates and reduces to the ordinary p-value of the
replication study if the uncertainty of the original effect estimate is
ignored. The proposed framework can also be used to determine the power or the
required replication sample size to achieve replication success. Numerical
calculations highlight the difficulty to achieve replication success if the
evidence from the original study is only suggestive. An application to data
from the Open Science Collaboration project on the replicability of
psychological science illustrates the proposed methodology.
| stat |
Neural Tangents: Fast and Easy Infinite Neural Networks in Python | Neural Tangents is a library designed to enable research into infinite-width
neural networks. It provides a high-level API for specifying complex and
hierarchical neural network architectures. These networks can then be trained
and evaluated either at finite-width as usual or in their infinite-width limit.
Infinite-width networks can be trained analytically using exact Bayesian
inference or using gradient descent via the Neural Tangent Kernel.
Additionally, Neural Tangents provides tools to study gradient descent training
dynamics of wide but finite networks in either function space or weight space.
The entire library runs out-of-the-box on CPU, GPU, or TPU. All computations
can be automatically distributed over multiple accelerators with near-linear
scaling in the number of devices. Neural Tangents is available at
www.github.com/google/neural-tangents. We also provide an accompanying
interactive Colab notebook.
| stat |
The DeCAMFounder: Non-Linear Causal Discovery in the Presence of Hidden
Variables | Many real-world decision-making tasks require learning casual relationships
between a set of variables. Typical causal discovery methods, however, require
that all variables are observed, which might not be realistic in practice.
Unfortunately, in the presence of latent confounding, recovering casual
relationships from observational data without making additional assumptions is
an ill-posed problem. Fortunately, in practice, additional structure among the
confounders can be expected, one such example being pervasive confounding,
which has been exploited for consistent causal estimation in the special case
of linear causal models. In this paper, we provide a proof and method to
estimate causal relationships in the non-linear, pervasive confounding setting.
The heart of our procedure relies on the ability to estimate the pervasive
confounding variation through a simple spectral decomposition of the observed
data matrix. We derive a DAG score function based on this insight, and
empirically compare our method to existing procedures. We show improved
performance on both simulated and real datasets by explicitly accounting for
both confounders and non-linear effects.
| stat |
Approximating Posterior Predictive Distributions by Averaging Output
From Many Particle Filters | This paper introduces the {\it particle swarm filter} (not to be confused
with particle swarm optimization): a recursive and embarrassingly parallel
algorithm that targets an approximation to the sequence of posterior predictive
distributions by averaging expectation approximations from many particle
filters. A law of large numbers and a central limit theorem are provided, as
well as an numerical study of simulated data from a stochastic volatility
model.
| stat |
Graph-Fused Multivariate Regression via Total Variation Regularization | In this paper, we propose the Graph-Fused Multivariate Regression (GFMR) via
Total Variation regularization, a novel method for estimating the association
between a one-dimensional or multidimensional array outcome and scalar
predictors. While we were motivated by data from neuroimaging and physical
activity tracking, the methodology is designed and presented in a generalizable
format and is applicable to many other areas of scientific research. The
estimator is the solution of a penalized regression problem where the objective
is the sum of square error plus a total variation (TV) regularization on the
predicted mean across all subjects. We propose an algorithm for parameter
estimation, which is efficient and scalable in a distributed computing
platform. Proof of the algorithm convergence is provided, and the statistical
consistency of the estimator is presented via an oracle inequality. We present
1D and 2D simulation results and demonstrate that GFMR outperforms existing
methods in most cases. We also demonstrate the general applicability of the
method by two real data examples, including the analysis of the 1D
accelerometry subsample of a large community-based study for mood disorders and
the analysis of the 3D MRI data from the attention-deficient/hyperactive
deficient (ADHD) 200 consortium.
| stat |
Heterogeneity Pursuit for Spatial Point Pattern with Application to Tree
Locations: A Bayesian Semiparametric Recourse | Spatial point pattern data are routinely encountered. A flexible regression
model for the underlying intensity is essential to characterizing the spatial
point pattern and understanding the impacts of potential risk factors on such
pattern. We propose a Bayesian semiparametric regression model where the
observed spatial points follow a spatial Poisson process with an intensity
function which adjusts a nonparametric baseline intensity with multiplicative
covariate effects. The baseline intensity is piecewise constant, approached
with a powered Chinese restaurant process prior which prevents an unnecessarily
large number of pieces. The parametric regression part allows for variable
selection through the spike-slab prior on the regression coefficients. An
efficient Markov chain Monte Carlo (MCMC) algorithm is developed for the
proposed methods. The performance of the methods is validated in an extensive
simulation study. In application to the locations of Beilschmiedia pendula
trees in the Barro Colorado Island forest dynamics research plot in central
Panama, the spatial heterogeneity is attributed to a subset of soil
measurements in addition to geographic measurements with a spatially varying
baseline intensity.
| stat |
Sharp bounds on the relative treatment effect for ordinal outcomes | For ordinal outcomes, the average treatment effect is often ill-defined and
hard to interpret. Echoing Agresti and Kateri (2017), we argue that the
relative treatment effect can be a useful measure especially for ordinal
outcomes, which is defined as $\gamma = \mathrm{pr}\{ Y_i(1) > Y_i(0) \} -
\mathrm{pr}\{ Y_i(1) < Y_i(0) \}$, with $Y_i(1)$ and $Y_i(0)$ being the
potential outcomes of unit $i$ under treatment and control, respectively. Given
the marginal distributions of the potential outcomes, we derive the sharp
bounds on $\gamma,$ which are identifiable parameters based on the observed
data. Agresti and Kateri (2017) focused on modeling strategies under the
assumption of independent potential outcomes, but we allow for arbitrary
dependence.
| stat |
Inference for BART with Multinomial Outcomes | The multinomial probit Bayesian additive regression trees (MPBART) framework
was proposed by Kindo et al. (KD), approximating the latent utilities in the
multinomial probit (MNP) model with BART (Chipman et al. 2010). Compared to
multinomial logistic models, MNP does not assume independent alternatives and
the correlation structure among alternatives can be specified through
multivariate Gaussian distributed latent utilities. We introduce two new
algorithms for fitting the MPBART and show that the theoretical mixing rates of
our proposals are equal or superior to the existing algorithm in KD. Through
simulations, we explore the robustness of the methods to the choice of
reference level, imbalance in outcome frequencies, and the specifications of
prior hyperparameters for the utility error term. The work is motivated by the
application of generating posterior predictive distributions for mortality and
engagement in care among HIV-positive patients based on electronic health
records (EHRs) from the Academic Model Providing Access to Healthcare (AMPATH)
in Kenya. In both the application and simulations, we observe better
performance using our proposals as compared to KD in terms of MCMC convergence
rate and posterior predictive accuracy.
| stat |
Assessing Time-Varying Causal Effect Moderation in the Presence of
Cluster-Level Treatment Effect Heterogeneity | Micro-randomized trial (MRT) is a sequential randomized experimental design
to empirically evaluate the effectiveness of mobile health (mHealth)
intervention components that may be delivered at hundreds or thousands of
decision points. The MRT context has motivated a new class of causal estimands,
termed "causal excursion effects", for which inference can be made by a
weighted, centered least squares approach (Boruvka et al., 2017). Existing
methods assume between-subject independence and non-interference. Deviations
from these assumptions often occur which, if unaccounted for, may result in
bias and overconfident variance estimates. In this paper, causal excursion
effects are considered under potential cluster-level correlation and
interference and when the treatment effect of interest depends on cluster-level
moderators. The utility of our proposed methods is shown by analyzing data from
a multi-institution cohort of first year medical residents in the United
States. The approach paves the way for construction of mHealth interventions
that account for observed social network information.
| stat |
Parameter-wise co-clustering for high-dimensional data | In recent years, data dimensionality has increasingly become a concern,
leading to many parameter and dimension reduction techniques being proposed in
the literature. A parameter-wise co-clustering model, for data modelled via
continuous random variables, is presented. The proposed model, although
allowing more flexibility, still maintains the very high degree of parsimony
achieved by traditional co-clustering. A stochastic expectation-maximization
(SEM) algorithm along with a Gibbs sampler is used for parameter estimation and
an integrated complete log-likelihood criterion is used for model selection.
Simulated and real datasets are used for illustration and comparison with
traditional co-clustering.
| stat |
Augmenting and Tuning Knowledge Graph Embeddings | Knowledge graph embeddings rank among the most successful methods for link
prediction in knowledge graphs, i.e., the task of completing an incomplete
collection of relational facts. A downside of these models is their strong
sensitivity to model hyperparameters, in particular regularizers, which have to
be extensively tuned to reach good performance [Kadlec et al., 2017]. We
propose an efficient method for large scale hyperparameter tuning by
interpreting these models in a probabilistic framework. After a model
augmentation that introduces per-entity hyperparameters, we use a variational
expectation-maximization approach to tune thousands of such hyperparameters
with minimal additional cost. Our approach is agnostic to details of the model
and results in a new state of the art in link prediction on standard benchmark
data.
| stat |
Robust Estimation of Propensity Score Weights via Subclassification | Weighting estimators based on propensity scores are widely used for causal
estimation in a variety of contexts, such as observational studies, marginal
structural models and interference. They enjoy appealing theoretical properties
such as consistency and possible efficiency under correct model specification.
However, this theoretical appeal may be diminished in practice by sensitivity
to misspecification of the propensity score model. To improve on this, we
borrow an idea from an alternative approach to causal effect estimation in
observational studies, namely subclassification estimators. It is well known
that compared to weighting estimators, subclassification methods are usually
more robust to model misspecification. In this paper, we first discuss an
intrinsic connection between the seemingly unrelated weighting and
subclassification estimators, and then use this connection to construct robust
propensity score weights via subclassification. We illustrate this idea by
proposing so-called full-classification weights and accompanying estimators for
causal effect estimation in observational studies. Our novel estimators are
both consistent and robust to model misspecification, thereby combining the
strengths of traditional weighting and subclassification estimators for causal
effect estimation from observational studies. Numerical studies show that the
proposed estimators perform favorably compared to existing methods.
| stat |
A Fast Proximal Point Method for Computing Exact Wasserstein Distance | Wasserstein distance plays increasingly important roles in machine learning,
stochastic programming and image processing. Major efforts have been under way
to address its high computational complexity, some leading to approximate or
regularized variations such as Sinkhorn distance. However, as we will
demonstrate, regularized variations with large regularization parameter will
degradate the performance in several important machine learning applications,
and small regularization parameter will fail due to numerical stability issues
with existing algorithms. We address this challenge by developing an Inexact
Proximal point method for exact Optimal Transport problem (IPOT) with the
proximal operator approximately evaluated at each iteration using projections
to the probability simplex. The algorithm (a) converges to exact Wasserstein
distance with theoretical guarantee and robust regularization parameter
selection, (b) alleviates numerical stability issue, (c) has similar
computational complexity to Sinkhorn, and (d) avoids the shrinking problem when
apply to generative models. Furthermore, a new algorithm is proposed based on
IPOT to obtain sharper Wasserstein barycenter.
| stat |
Total positivity in exponential families with application to binary
variables | We study exponential families of distributions that are multivariate totally
positive of order 2 (MTP2), show that these are convex exponential families,
and derive conditions for existence of the MLE. Quadratic exponential familes
of MTP2 distributions contain attractive Gaussian graphical models and
ferromagnetic Ising models as special examples. We show that these are defined
by intersecting the space of canonical parameters with a polyhedral cone whose
faces correspond to conditional independence relations. Hence MTP2 serves as an
implicit regularizer for quadratic exponential families and leads to sparsity
in the estimated graphical model. We prove that the maximum likelihood
estimator (MLE) in an MTP2 binary exponential family exists if and only if both
of the sign patterns $(1,-1)$ and $(-1,1)$ are represented in the sample for
every pair of variables; in particular, this implies that the MLE may exist
with $n=d$ observations, in stark contrast to unrestricted binary exponential
families where $2^d$ observations are required. Finally, we provide a novel and
globally convergent algorithm for computing the MLE for MTP2 Ising models
similar to iterative proportional scaling and apply it to the analysis of data
from two psychological disorders.
| stat |
The Relationship between Major Depression Symptom Severity and Sleep
Collected Using a Wristband Wearable Device: Multi-centre Longitudinal
Observational Study | Research in mental health has implicated sleep pathologies with depression.
However, the gold standard for sleep assessment, polysomnography, is not
suitable for long-term, continuous, monitoring of daily sleep, and methods such
as sleep diaries rely on subjective recall, which is qualitative and
inaccurate. Wearable devices, on the other hand, provide a low-cost and
convenient means to monitor sleep in home settings. The main aim of this study
was to devise and extract sleep features, from data collected using a wearable
device, and analyse their correlation with depressive symptom severity and
sleep quality, as measured by the self-assessed Patient Health Questionnaire
8-item. Daily sleep data were collected passively by Fitbit wristband devices,
and depressive symptom severity was self-reported every two weeks by the PHQ-8.
The data used in this paper included 2,812 PHQ-8 records from 368 participants
recruited from three study sites in the Netherlands, Spain, and the UK.We
extracted 21 sleep features from Fitbit data which describe sleep in the
following five aspects: sleep architecture, sleep stability, sleep quality,
insomnia, and hypersomnia. Linear mixed regression models were used to explore
associations between sleep features and depressive symptom severity. The z-test
was used to evaluate the significance of the coefficient of each feature. We
tested our models on the entire dataset and individually on the data of three
different study sites. We identified 16 sleep features that were significantly
correlated with the PHQ-8 score on the entire dataset. Associations between
sleep features and the PHQ-8 score varied across different sites, possibly due
to the difference in the populations.
| stat |
Tree based credible set estimation | Estimating a joint Highest Posterior Density credible set for a multivariate
posterior density is challenging as dimension gets larger. Credible intervals
for univariate marginals are usually presented for ease of computation and
visualisation. There are often two layers of approximation, as we may need to
compute a credible set for a target density which is itself only an
approximation to the true posterior density. We obtain joint Highest Posterior
Density credible sets for density estimation trees given by Li et al. (2016)
approximating a density truncated to a compact subset of R^d as this is
preferred to a copula construction. These trees approximate a joint posterior
distribution from posterior samples using a piecewise constant function defined
by sequential binary splits. We use a consistent estimator to measure of the
symmetric difference between our credible set estimate and the true HPD set of
the target density samples. This quality measure can be computed without the
need to know the true set. We show how the true-posterior-coverage of an
approximate credible set estimated for an approximate target density may be
estimated in doubly intractable cases where posterior samples are not
available. We illustrate our methods with simulation studies and find that our
estimator is competitive with existing methods.
| stat |
Clustering Data with Nonignorable Missingness using Semi-Parametric
Mixture Models | We are concerned in clustering continuous data sets subject to non-ignorable
missingness. We perform clustering with a specific semi-parametric mixture,
under the assumption of conditional independence given the component. The
mixture model isused for clustering and not for estimating the density of the
full variables (observed and unobserved), thus we do not need other assumptions
on the component distribution neither to specify the missingness mechanism.
Estimation is performed by maximizing an extension of smoothed likelihood
allowing missingness. This optimization is achieved by a
Majorization-Minorization algorithm. We illustrate the relevance of our
approach by numerical experiments. Under mild assumptions, we show the
identifiability of the model defining the distribution of the observed data and
the monotony of the algorithm. We also propose an extension of this new method
to the case of mixed-type data that we illustrate on a real data set.
| stat |
Who Gets the Job and How are They Paid? Machine Learning Application on
H-1B Case Data | In this paper, we use machine learning techniques to explore the H-1B
application dataset disclosed by the Department of Labor (DOL), from 2008 to
2018, in order to provide more stylized facts of the international workers in
US labor market. We train a LASSO Regression model to analyze the impact of
different features on the applicant's wage, and a Logistic Regression with
L1-Penalty as a classifier to study the feature's impact on the likelihood of
the case being certified. Our analysis shows that working in the healthcare
industry, working in California, higher job level contribute to higher
salaries. In the meantime, lower job level, working in the education services
industry and nationality of Philippines are negatively correlated with the
salaries. In terms of application status, a Ph.D. degree, working in retail or
finance, majoring in computer science will give the applicants a better chance
of being certified. Applicants with no or an associate degree, working in the
education services industry, or majoring in education are more likely to be
rejected.
| stat |
Information Theoretic Lower Bounds for Feed-Forward Fully-Connected Deep
Networks | In this paper, we study the sample complexity lower bounds for the exact
recovery of parameters and for a positive excess risk of a feed-forward,
fully-connected neural network for binary classification, using
information-theoretic tools. We prove these lower bounds by the existence of a
generative network characterized by a backwards data generating process, where
the input is generated based on the binary output, and the network is
parametrized by weight parameters for the hidden layers. The sample complexity
lower bound for the exact recovery of parameters is $\Omega(d r \log(r) + p )$
and for a positive excess risk is $\Omega(r \log(r) + p )$, where $p$ is the
dimension of the input, $r$ reflects the rank of the weight matrices and $d$ is
the number of hidden layers. To the best of our knowledge, our results are the
first information theoretic lower bounds.
| stat |
Revisiting Reweighted Wake-Sleep for Models with Stochastic Control Flow | Stochastic control-flow models (SCFMs) are a class of generative models that
involve branching on choices from discrete random variables. Amortized
gradient-based learning of SCFMs is challenging as most approaches targeting
discrete variables rely on their continuous relaxations---which can be
intractable in SCFMs, as branching on relaxations requires evaluating all
(exponentially many) branching paths. Tractable alternatives mainly combine
REINFORCE with complex control-variate schemes to improve the variance of naive
estimators. Here, we revisit the reweighted wake-sleep (RWS) (Bornschein and
Bengio, 2015) algorithm, and through extensive evaluations, show that it
outperforms current state-of-the-art methods in learning SCFMs. Further, in
contrast to the importance weighted autoencoder, we observe that RWS learns
better models and inference networks with increasing numbers of particles. Our
results suggest that RWS is a competitive, often preferable, alternative for
learning SCFMs.
| stat |
Cycle-to-Cycle Queue Length Estimation from Connected Vehicles with
Filtering on Primary Parameters | Estimation models from connected vehicles often assume low level parameters
such as arrival rates and market penetration rates as known or estimate them in
real-time. At low market penetration rates, such parameter estimators produce
large errors making estimated queue lengths inefficient for control or
operations applications. In order to improve accuracy of low level parameter
estimations, this study investigates the impact of connected vehicles
information filtering on queue length estimation models. Filters are used as
multilevel real-time estimators. Accuracy is tested against known arrival rate
and market penetration rate scenarios using microsimulations. To understand the
effectiveness for short-term or for dynamic processes, arrival rates, and
market penetration rates are changed every 15 minutes. The results show that
with Kalman and Particle filters, parameter estimators are able to find the
true values within 15 minutes and meet and surpass the accuracy of known
parameter scenarios especially for low market penetration rates. In addition,
using last known estimated queue lengths when no connected vehicle is present
performs better than inputting average estimated values. Moreover, the study
shows that both filtering algorithms are suitable for real-time applications
that require less than 0.1 second computational time.
| stat |
A Notion of Individual Fairness for Clustering | A common distinction in fair machine learning, in particular in fair
classification, is between group fairness and individual fairness. In the
context of clustering, group fairness has been studied extensively in recent
years; however, individual fairness for clustering has hardly been explored. In
this paper, we propose a natural notion of individual fairness for clustering.
Our notion asks that every data point, on average, is closer to the points in
its own cluster than to the points in any other cluster. We study several
questions related to our proposed notion of individual fairness. On the
negative side, we show that deciding whether a given data set allows for such
an individually fair clustering in general is NP-hard. On the positive side,
for the special case of a data set lying on the real line, we propose an
efficient dynamic programming approach to find an individually fair clustering.
For general data sets, we investigate heuristics aimed at minimizing the number
of individual fairness violations and compare them to standard clustering
approaches on real data sets.
| stat |
Algorithms for slate bandits with non-separable reward functions | In this paper, we study a slate bandit problem where the function that
determines the slate-level reward is non-separable: the optimal value of the
function cannot be determined by learning the optimal action for each slot. We
are mainly concerned with cases where the number of slates is large relative to
the time horizon, so that trying each slate as a separate arm in a traditional
multi-armed bandit, would not be feasible. Our main contribution is the design
of algorithms that still have sub-linear regret with respect to the time
horizon, despite the large number of slates. Experimental results on simulated
data and real-world data show that our proposed method outperforms popular
benchmark bandit algorithms.
| stat |
Tensor Restricted Isometry Property Analysis For a Large Class of Random
Measurement Ensembles | In previous work, theoretical analysis based on the tensor Restricted
Isometry Property (t-RIP) established the robust recovery guarantees of a
low-tubal-rank tensor. The obtained sufficient conditions depend strongly on
the assumption that the linear measurement maps satisfy the t-RIP. In this
paper, by exploiting the probabilistic arguments, we prove that such linear
measurement maps exist under suitable conditions on the number of measurements
in terms of the tubal rank r and the size of third-order tensor n1, n2, n3. And
the obtained minimal possible number of linear measurements is nearly optimal
compared with the degrees of freedom of a tensor with tubal rank r. Specially,
we consider a random sub-Gaussian distribution that includes Gaussian,
Bernoulli and all bounded distributions and construct a large class of linear
maps that satisfy a t-RIP with high probability. Moreover, the validity of the
required number of measurements is verified by numerical experiments.
| stat |
Fitting ARMA Time Series Models without Identification: A Proximal
Approach | Fitting autoregressive moving average (ARMA) time series models requires
model identification before parameter estimation. Model identification involves
determining the order of the autoregressive and moving average components which
is generally performed by inspection of the autocorrelation and partial
autocorrelation functions or other offline methods. In this work, we regularize
the parameter estimation optimization problem with a nonsmooth hierarchical
sparsity-inducing penalty based on two path graphs that allows performing model
identification and parameter estimation simultaneously. A proximal block
coordinate descent algorithm is then proposed to solve the underlying
optimization problem efficiently. The resulting model satisfies the required
stationarity and invertibility conditions for ARMA models. Numerical studies
supporting the performance of the proposed method and comparing it with other
schemes are presented.
| stat |
Combining Representation Learning with Tensor Factorization for Risk
Factor Analysis - an application to Epilepsy and Alzheimer's disease | Existing studies consider Alzheimer's disease (AD) a comorbidity of epilepsy,
but also recognize epilepsy to occur more frequently in patients with AD than
those without. The goal of this paper is to understand the relationship between
epilepsy and AD by studying causal relations among subgroups of epilepsy
patients. We develop an approach combining representation learning with tensor
factorization to provide an in-depth analysis of the risk factors among
epilepsy patients for AD. An epilepsy-AD cohort of ~600,000 patients were
extracted from Cerner Health Facts data (50M patients). Our experimental
results not only suggested a causal relationship between epilepsy and later
onset of AD ( p = 1.92e-51), but also identified five epilepsy subgroups with
distinct phenotypic patterns leading to AD. While such findings are
preliminary, the proposed method combining representation learning with tensor
factorization seems to be an effective approach for risk factor analysis.
| stat |
Using mobility data in the design of optimal lockdown strategies for the
COVID-19 pandemic | A mathematical model for the COVID-19 pandemic spread, which integrates
age-structured Susceptible-Exposed-Infected-Recovered-Deceased dynamics with
real mobile phone data accounting for the population mobility, is presented.
The dynamical model adjustment is performed via Approximate Bayesian
Computation. Optimal lockdown and exit strategies are determined based on
nonlinear model predictive control, constrained to public-health and
socio-economic factors. Through an extensive computational validation of the
methodology, it is shown that it is possible to compute robust exit strategies
with realistic reduced mobility values to inform public policy making, and we
exemplify the applicability of the methodology using datasets from England and
France. Code implementing the described experiments is available at
https://github.com/OptimalLockdown.
| stat |
Hierarchical Models for Independence Structures of Networks | We introduce a new family of network models, called hierarchical network
models, that allow us to represent in an explicit manner the stochastic
dependence among the dyads (random ties) of the network. In particular, each
member of this family can be associated with a graphical model defining
conditional independence clauses among the dyads of the network, called the
dependency graph. Every network model with dyadic independence assumption can
be generalized to construct members of this new family. Using this new
framework, we generalize the Erd\"os-R\'enyi and beta-models to create
hierarchical Erd\"os-R\'enyi and beta-models. We describe various methods for
parameter estimation as well as simulation studies for models with sparse
dependency graphs.
| stat |
Equal Opportunity and Affirmative Action via Counterfactual Predictions | Machine learning (ML) can automate decision-making by learning to predict
decisions from historical data. However, these predictors may inherit
discriminatory policies from past decisions and reproduce unfair decisions. In
this paper, we propose two algorithms that adjust fitted ML predictors to make
them fair. We focus on two legal notions of fairness: (a) providing equal
opportunity (EO) to individuals regardless of sensitive attributes and (b)
repairing historical disadvantages through affirmative action (AA). More
technically, we produce fair EO and AA predictors by positing a causal model
and considering counterfactual decisions. We prove that the resulting
predictors are theoretically optimal in predictive performance while satisfying
fairness. We evaluate the algorithms, and the trade-offs between accuracy and
fairness, on datasets about admissions, income, credit and recidivism.
| stat |
Query-Based Selection of Optimal Candidates under the Mallows Model | We introduce the problem of query-based selection of the optimal candidate in
rank-ordered lists generated by the Mallows model. In this setting, one is
presented with a list of candidates drawn according to a Gaussian-like
distribution for permutations and the goal is to identify the highest ranked
candidate through a sequential interview process that does not allow rejected
candidates to be revisited. The new modeling assumption consists in being able
to query a Genie at the time a selection is to be made. The query provides an
answer indicating if the candidate in question is the highest-ranked or not. If
the Genie confirms the query, the selection process terminates. Otherwise, the
sequential examinations continue until a new potentially optimal candidate is
identified. Our results include optimal interview strategies for a bounded
number of queries that can be exactly determined through numerical evaluations
as well as the expected number of interviews until the optimal candidate is
identified or the interview process completed. Although the problem exhibits
similarities with the Dowry problem with multiple choices of Gilbert and
Mosteller, the proposed Genie-based model substantially differs from it as it
allows for early stopping and addresses nonuniform candidate interview
distributions.
| stat |
An essay on copula modelling for discrete random vectors; or how to pour
new wine into old bottles | Copulas have now become ubiquitous statistical tools for describing,
analysing and modelling dependence between random variables. Sklar's theorem,
"the fundamental theorem of copulas", makes a clear distinction between the
continuous case and the discrete case, though. In particular, the copula of a
discrete random vector is not identifiable, which causes serious
inconsistencies. In spite of this, downplaying statements are widespread in the
related literature, and copula methods are used for modelling dependence
between discrete variables. This paper calls to reconsidering the soundness of
copula modelling for discrete data. It suggests a more fundamental construction
which allows copula ideas to smoothly carry over to the discrete case. Actually
it is an attempt at rejuvenating some century-old ideas of Udny Yule, who
mentioned a similar construction a long time before copulas got in fashion.
| stat |
The Role of Shopping Mission in Retail Customer Segmentation | In retailing, it is important to understand customer behavior and determine
customer value. A useful tool to achieve such goals is the cluster analysis of
transaction data. Typically, a customer segmentation is based on the recency,
frequency and monetary value of shopping or the structure of purchased
products. We take a different approach and base our segmentation on the
shopping mission - a reason why a customer visits the shop. Shopping missions
include focused purchases of specific product categories and general purchases
of various sizes. In an application to a Czech drugstore chain, we show that
the proposed segmentation brings unique information about customers and should
be used alongside the traditional methods.
| stat |
Group level MEG/EEG source imaging via optimal transport: minimum
Wasserstein estimates | Magnetoencephalography (MEG) and electroencephalogra-phy (EEG) are
non-invasive modalities that measure the weak electromagnetic fields generated
by neural activity. Inferring the location of the current sources that
generated these magnetic fields is an ill-posed inverse problem known as source
imaging. When considering a group study, a baseline approach consists in
carrying out the estimation of these sources independently for each subject.
The ill-posedness of each problem is typically addressed using sparsity
promoting regularizations. A straightforward way to define a common pattern for
these sources is then to average them. A more advanced alternative relies on a
joint localization of sources for all subjects taken together, by enforcing
some similarity across all estimated sources. An important advantage of this
approach is that it consists in a single estimation in which all measurements
are pooled together, making the inverse problem better posed. Such a joint
estimation poses however a few challenges, notably the selection of a valid
regularizer that can quantify such spatial similarities. We propose in this
work a new procedure that can do so while taking into account the geometrical
structure of the cortex. We call this procedure Minimum Wasserstein Estimates
(MWE). The benefits of this model are twofold. First, joint inference allows to
pool together the data of different brain geometries, accumulating more spatial
information. Second, MWE are defined through Optimal Transport (OT) metrics
which provide a tool to model spatial proximity between cortical sources of
different subjects, hence not enforcing identical source location in the group.
These benefits allow MWE to be more accurate than standard MEG source
localization techniques. To support these claims, we perform source
localization on realistic MEG simulations based on forward operators derived
from MRI scans. On a visual task dataset, we demonstrate how MWE infer neural
patterns similar to functional Magnetic Resonance Imaging (fMRI) maps.
| stat |
Fused graphical lasso for brain networks with symmetries | Neuroimaging is the growing area of neuroscience devoted to produce data with
the goal of capturing processes and dynamics of the human brain. We consider
the problem of inferring the brain connectivity network from time dependent
functional magnetic resonance imaging (fMRI) scans. To this aim we propose the
symmetric graphical lasso, a penalized likelihood method with a fused type
penalty function that takes into explicit account the natural symmetrical
structure of the brain. Symmetric graphical lasso allows one to learn
simultaneously both the network structure and a set of symmetries across the
two hemispheres. We implement an alternating directions method of multipliers
algorithm to solve the corresponding convex optimization problem. Furthermore,
we apply our methods to estimate the brain networks of two subjects, one
healthy and the other affected by a mental disorder, and to compare them with
respect to their symmetric structure. The method applies once the temporal
dependence characterising fMRI data has been accounted for and we compare the
impact on the analysis of different detrending techniques on the estimated
brain networks.
| stat |
A Random Matrix Analysis of Random Fourier Features: Beyond the Gaussian
Kernel, a Precise Phase Transition, and the Corresponding Double Descent | This article characterizes the exact asymptotics of random Fourier feature
(RFF) regression, in the realistic setting where the number of data samples
$n$, their dimension $p$, and the dimension of feature space $N$ are all large
and comparable. In this regime, the random RFF Gram matrix no longer converges
to the well-known limiting Gaussian kernel matrix (as it does when $N \to
\infty$ alone), but it still has a tractable behavior that is captured by our
analysis. This analysis also provides accurate estimates of training and test
regression errors for large $n,p,N$. Based on these estimates, a precise
characterization of two qualitatively different phases of learning, including
the phase transition between them, is provided; and the corresponding double
descent test error curve is derived from this phase transition behavior. These
results do not depend on strong assumptions on the data distribution, and they
perfectly match empirical results on real-world data sets.
| stat |
Cox regression analysis for distorted covariates with an unknown
distortion function | We study inference for censored survival data where some covariates are
distorted by some unknown functions of an observable confounding variable in a
multiplicative form. Example of this kind of data in medical studies is the
common practice to normalizing some important observed exposure variables by
patients' body mass index (BMI), weight or age. Such phenomenon also appears
frequently in environmental studies where ambient measure is used for
normalization, and in genomic studies where library size needs to be normalized
for next generation sequencing data. We propose a new covariate-adjusted Cox
proportional hazards regression model and utilize the kernel smoothing method
to estimate the distorting function, then employ an estimated maximum
likelihood method to derive estimator for the regression parameters. We
establish the large sample properties of the proposed estimator. Extensive
simulation studies demonstrate that the proposed estimator performs well in
correcting the bias arising from distortion. A real data set from the National
Wilms' Tumor Study (NWTS) is used to illustrate the proposed approach.
| stat |
Model selection for high-dimensional linear regression with dependent
observations | We investigate the prediction capability of the orthogonal greedy algorithm
(OGA) in high-dimensional regression models with dependent observations. The
rates of convergence of the prediction error of OGA are obtained under a
variety of sparsity conditions. To prevent OGA from overfitting, we introduce a
high-dimensional Akaike's information criterion (HDAIC) to determine the number
of OGA iterations. A key contribution of this work is to show that OGA, used in
conjunction with HDAIC, can achieve the optimal convergence rate without
knowledge of how sparse the underlying high-dimensional model is.
| stat |
Bayesian calibration and sensitivity analysis for a karst aquifer model
using active subspaces | In this article, we perform a parameter study for a recently developed karst
hydrological model. The study consists of a high-dimensional Bayesian inverse
problem and a global sensitivity analysis. For the first time in karst
hydrology, we use the active subspace method to find directions in the
parameter space that dominate the Bayesian update from the prior to the
posterior distribution in order to effectively reduce the dimension of the
problem and for computational efficiency. Additionally, the calculated active
subspace can be exploited to construct sensitivity metrics on each of the
individual parameters and be used to construct a natural model surrogate. The
model consists of 21 parameters to reproduce the hydrological behavior of
spring discharge in a karst aquifer located in the Kerschbaum spring recharge
area at Waidhofen a.d. Ybbs in Austria. The experimental spatial and time
series data for the inference process were collected by the water works in
Waidhofen. We show that this case study has implicit low-dimensionality, and we
run an adjusted Markov chain Monte Carlo algorithm in a low-dimensional
subspace to construct samples of the posterior distribution. The results are
visualized and verified by plots of the posterior's push-forward distribution
displaying the uncertainty in predicting discharge values due to the
experimental noise in the data. Finally, a discussion provides hydrological
interpretation of these results for the Kerschbaum area.
| stat |
Predictive distributions that mimic frequencies over a restricted
subdomain (expanded preprint version) | A predictive distribution over a sequence of $N+1$ events is said to be
"frequency mimicking" whenever the probability for the final event conditioned
on the outcome of the first $N$ events equals the relative frequency of
successes among them. Infinitely extendible exchangeable distributions that
universally inhere this property are known to have several annoying concomitant
properties. We motivate frequency mimicking assertions over a limited subdomain
in practical problems of finite inference, and we identify their computable
coherent implications. We provide some computed examples using reference
distributions, and we introduce computational software to generate any
specification. The software derives from an inversion of the finite form of the
exchangeability representation theorem. Three new theorems delineate the extent
of the usefulness of such distributions, and we show why it may not be
appropriate to extend the frequency mimicking assertions for a specified value
of $N$ to any arbitrary larger size of $N$. The constructive results identify
the source and structure of "adherent masses" in the limit of a sequence of
finitely additive distributions. Appendices develop a novel geometrical
representation of conditional probabilities which illuminate the analysis.
| stat |
TEAM: A Multiple Testing Algorithm on the Aggregation Tree for Flow
Cytometry Analysis | In immunology studies, flow cytometry is a commonly used multivariate
single-cell assay. One key goal in flow cytometry analysis is to pinpoint the
immune cells responsive to certain stimuli. Statistically, this problem can be
translated into comparing two protein expression probability density functions
(PDFs) before and after the stimulus; the goal is to pinpoint the regions where
these two pdfs differ. In this paper, we model this comparison as a multiple
testing problem. First, we partition the sample space into small bins. In each
bin we form a hypothesis to test the existence of differential pdfs. Second, we
develop a novel multiple testing method, called TEAM (Testing on the
Aggregation tree Method), to identify those bins that harbor differential pdfs
while controlling the false discovery rate (FDR) under the desired level. TEAM
embeds the testing procedure into an aggregation tree to test from fine- to
coarse-resolution. The procedure achieves the statistical goal of pinpointing
differential pdfs to the smallest possible regions. TEAM is computationally
efficient, capable of analyzing large flow cytometry data sets in much shorter
time compared with competing methods. We applied TEAM and competing methods on
a flow cytometry data set to identify T cells responsive to the cytomeglovirus
(CMV)-pp65 antigen stimulation. TEAM successfully identified the
monofunctional, bifunctional, and polyfunctional T cells while the competing
methods either did not finish in a reasonable time frame or provided less
interpretable results. Numerical simulations and theoretical justifications
demonstrate that TEAM has asymptotically valid, powerful, and robust
performance. Overall, TEAM is a computationally efficient and statistically
powerful algorithm that can yield meaningful biological insights in flow
cytometry studies.
| stat |
Analysis of zero inflated dichotomous variables from a Bayesian
perspective: Application to occupational health | This work proposes a new methodology to fit zero inflated Bernoulli data from
a Bayesian approach, able to distinguish between two potential sources of zeros
(structurals and non-structurals). Its usage is illustrated by means of a real
example from the field of occupational health as the phenomenon of sickness
presenteeism, in which it is reasonable to think that some individuals will
never be at risk of suffering it because they have not been sick in the period
of study (structural zeros). Without separating structural and non-structural
zeros one would one would be studying jointly the general health status and the
presenteeism itself, and therefore obtaining potentially biased estimates as
the phenomenon is being implicitly underestimated by diluting it into the
general health status. The proposed methodology performance has been evaluated
through a comprehensive simulation study, and it has been compiled as an R
package freely available to the community.
| stat |
Evaluating probabilistic forecasts of football matches: The case against
the Ranked Probability Score | A scoring rule is a function of a probabilistic forecast and a corresponding
outcome that is used to evaluate forecast performance. A wide range of scoring
rules have been defined over time and there is some debate as to which are the
most appropriate for evaluating the performance of forecasts of sporting
events. This paper focuses on forecasts of the outcomes of football matches.
The ranked probability score (RPS) is often recommended since it is `sensitive
to distance', that is it takes into account the ordering in the outcomes (a
home win is `closer' to a draw than it is to an away win, for example). In this
paper, this reasoning is disputed on the basis that it adds nothing in terms of
the actual aims of using scoring rules. A related property of scoring rules is
locality. A scoring rule is local if it only takes the probability placed on
the outcome into consideration. Two simulation experiments are carried out in
the context of football matches to compare the performance of the RPS, which is
non-local and sensitive to distance, the Brier score, which is non-local and
insensitive to distance, and the ignorance score, which is local and
insensitive to distance. The ignorance score is found to outperform both the
RPS and the Brier score, casting doubt on the value of non-locality and
sensitivity to distance as properties of scoring rules in this context.
| stat |
Automatic Discovery of Privacy-Utility Pareto Fronts | Differential privacy is a mathematical framework for privacy-preserving data
analysis. Changing the hyperparameters of a differentially private algorithm
allows one to trade off privacy and utility in a principled way. Quantifying
this trade-off in advance is essential to decision-makers tasked with deciding
how much privacy can be provided in a particular application while maintaining
acceptable utility. Analytical utility guarantees offer a rigorous tool to
reason about this trade-off, but are generally only available for relatively
simple problems. For more complex tasks, such as training neural networks under
differential privacy, the utility achieved by a given algorithm can only be
measured empirically. This paper presents a Bayesian optimization methodology
for efficiently characterizing the privacy--utility trade-off of any
differentially private algorithm using only empirical measurements of its
utility. The versatility of our method is illustrated on a number of machine
learning tasks involving multiple models, optimizers, and datasets.
| stat |
Distributionally Robust Local Non-parametric Conditional Estimation | Conditional estimation given specific covariate values (i.e., local
conditional estimation or functional estimation) is ubiquitously useful with
applications in engineering, social and natural sciences. Existing data-driven
non-parametric estimators mostly focus on structured homogeneous data (e.g.,
weakly independent and stationary data), thus they are sensitive to adversarial
noise and may perform poorly under a low sample size. To alleviate these
issues, we propose a new distributionally robust estimator that generates
non-parametric local estimates by minimizing the worst-case conditional
expected loss over all adversarial distributions in a Wasserstein ambiguity
set. We show that despite being generally intractable, the local estimator can
be efficiently found via convex optimization under broadly applicable settings,
and it is robust to the corruption and heterogeneity of the data. Experiments
with synthetic and MNIST datasets show the competitive performance of this new
class of estimators.
| stat |
Synthesized Difference in Differences | Randomized clinical trials (RCTs) eliminate confounding but impose strict
exclusion criteria that prevent sampling of the entire clinical population.
Observational datasets are more inclusive but suffer from confounding.
Difference in Differences (DD) eliminates confounding from observational data
by comparing outcomes before and after treatment administration. However, the
algorithm requires a parallel slopes assumption that may not apply in practice
when confounding shifts across time. In this paper, we propose Synthesized
Difference in Differences (SDD) that infers the correct (possibly non-parallel)
slopes by linearly adjusting a conditional version of DD using additional RCT
data. The algorithm achieves state of the art performance across multiple
synthetic and real datasets even when the RCT excludes the majority of
patients.
| stat |
Stationary vine copula models for multivariate time series | Multivariate time series exhibit two types of dependence: across variables
and across time points. Vine copulas are graphical models for the dependence
and can conveniently capture both types of dependence in the same model. We
derive the maximal class of graph structures that guarantees stationarity under
a condition called translation invariance. Translation invariance is not only a
necessary condition for stationarity, but also the only condition we can
reasonably check in practice. In this sense, the new model class characterizes
all practically relevant vine structures for modeling stationary time series.
We propose computationally efficient methods for estimation, simulation,
prediction, and uncertainty quantification and show their validity by
asymptotic results and simulations. The theoretical results allow for
misspecified models and, even when specialized to the \emph{iid} case, go
beyond what is available in the literature. The new model class is illustrated
by an application to forecasting returns of a portolio of 20 stocks, where they
show excellent forecast performance. The paper is accompanied by an open source
software implementation.
| stat |
Two-step method for assessing dissimilarity of random sets | The paper concerns a new statistical method for assessing dissimilarity of
two random sets based on one realisation of each of them. The method focuses on
shapes of the components of the random sets, namely on the curvature of their
boundaries together with the ratios of their perimeters and areas. Theoretical
background is introduced and then, the method is described, justified by a
simulation study and applied to real data of two different types of tissue -
mammary cancer and mastopathy.
| stat |
The role of mobility and sanitary measures on Covid-19 in Costa Rica,
March through July 2020 | The aim of this paper is to infer the effects that changes on human mobility
had on the transmission dynamics during the first four months of the SARS-CoV-2
outbreak in Costa Rica, before community transmission was established in the
country. By using parametric and non parametric detection change-point
techniques we were able to identify two different periods where at least the
trend and variability of new daily cases significantly changed. In order to
combine this information with population movement, we use data from Google
Mobility Trends that allow us to estimate the lag between the rate of new daily
cases and each of the categories established by Google. The information is then
used to establish an association between changes in population mobility and the
sanitary measures taken during the study period.
| stat |
Sparse Unit-Sum Regression | This paper considers sparsity in linear regression under the restriction that
the regression weights sum to one. We propose an approach that combines
$\ell_0$- and $\ell_1$-regularization. We compute its solution by adapting a
recent methodological innovation made by Bertsimas et al. (2016) for
$\ell_0$-regularization in standard linear regression. In a simulation
experiment we compare our approach to $\ell_0$-regularization and
$\ell_1$-regularization and find that it performs favorably in terms of
predictive performance and sparsity. In an application to index tracking we
show that our approach can obtain substantially sparser portfolios compared to
$\ell_1$-regularization while maintaining a similar tracking performance.
| stat |
Causal Clustering for 1-Factor Measurement Models on Data with Various
Types | The tetrad constraint is a condition of which the satisfaction signals a rank
reduction of a covariance submatrix and is used to design causal discovery
algorithms that detects the existence of latent (unmeasured) variables, such as
FOFC. Initially such algorithms only work for cases where the measured and
latent variables are all Gaussian and have linear relations (Gaussian-Gaussian
Case). It has been shown that a unidimentional latent variable model implies
tetrad constraints when the measured and latent variables are all binary
(Binary-Binary case). This paper proves that the tetrad constraint can also be
entailed when the measured variables are of mixed data types and when the
measured variables are discrete and the latent common causes are continuous,
which implies that any clustering algorithm relying on this constraint can work
on those cases. Each case is shown with an example and a proof. The performance
of FOFC on mixed data is shown by simulation studies and is compared with some
algorithms with similar functions.
| stat |
Empirical Likelihood-Based Estimation and Inference in Randomized
Controlled Trials with High-Dimensional Covariates | In this paper, we propose a data-adaptive empirical likelihood-based approach
for treatment effect estimation and inference, which overcomes the obstacle of
the traditional empirical likelihood-based approaches in the high-dimensional
setting by adopting penalized regression and machine learning methods to model
the covariate-outcome relationship. In particular, we show that our procedure
successfully recovers the true variance of Zhang's treatment effect estimator
(Zhang, 2018) by utilizing a data-splitting technique. Our proposed estimator
is proved to be asymptotically normal and semiparametric efficient under mild
regularity conditions. Simulation studies indicate that our estimator is more
efficient than the estimator proposed by Wager et al. (2016) when random
forests are employed to model the covariate-outcome relationship. Moreover,
when multiple machine learning models are imposed, our estimator is at least as
efficient as any regular estimator with a single machine learning model. We
compare our method to existing ones using the ACTG175 data and the GSE118657
data, and confirm the outstanding performance of our approach.
| stat |
Geometrical Insights for Implicit Generative Modeling | Learning algorithms for implicit generative models can optimize a variety of
criteria that measure how the data distribution differs from the implicit model
distribution, including the Wasserstein distance, the Energy distance, and the
Maximum Mean Discrepancy criterion. A careful look at the geometries induced by
these distances on the space of probability measures reveals interesting
differences. In particular, we can establish surprising approximate global
convergence guarantees for the $1$-Wasserstein distance,even when the
parametric generator has a nonconvex parametrization.
| stat |
Introducing Super Pseudo Panels: Application to Transport Preference
Dynamics | We propose a new approach for constructing synthetic pseudo-panel data from
cross-sectional data. The pseudo panel and the preferences it intends to
describe is constructed at the individual level and is not affected by
aggregation bias across cohorts. This is accomplished by creating a
high-dimensional probabilistic model representation of the entire data set,
which allows sampling from the probabilistic model in such a way that all of
the intrinsic correlation properties of the original data are preserved. The
key to this is the use of deep learning algorithms based on the Conditional
Variational Autoencoder (CVAE) framework. From a modelling perspective, the
concept of a model-based resampling creates a number of opportunities in that
data can be organized and constructed to serve very specific needs of which the
forming of heterogeneous pseudo panels represents one. The advantage, in that
respect, is the ability to trade a serious aggregation bias (when aggregating
into cohorts) for an unsystematic noise disturbance. Moreover, the approach
makes it possible to explore high-dimensional sparse preference distributions
and their linkage to individual specific characteristics, which is not possible
if applying traditional pseudo-panel methods. We use the presented approach to
reveal the dynamics of transport preferences for a fixed pseudo panel of
individuals based on a large Danish cross-sectional data set covering the
period from 2006 to 2016. The model is also utilized to classify individuals
into 'slow' and 'fast' movers with respect to the speed at which their
preferences change over time. It is found that the prototypical fast mover is a
young woman who lives as a single in a large city whereas the typical slow
mover is a middle-aged man with high income from a nuclear family who lives in
a detached house outside a city.
| stat |
Bayesian filtering unifies adaptive and non-adaptive neural network
optimization methods | We formulate the problem of neural network optimization as Bayesian
filtering, where the observations are the backpropagated gradients. While
neural network optimization has previously been studied using natural gradient
methods which are closely related to Bayesian inference, they were unable to
recover standard optimizers such as Adam and RMSprop with a root-mean-square
gradient normalizer, instead getting a mean-square normalizer. To recover the
root-mean-square normalizer, we find it necessary to account for the temporal
dynamics of all the other parameters as they are geing optimized. The resulting
optimizer, AdaBayes, adaptively transitions between SGD-like and Adam-like
behaviour, automatically recovers AdamW, a state of the art variant of Adam
with decoupled weight decay, and has generalisation performance competitive
with SGD.
| stat |
Loss based prior for the degrees of freedom of the Wishart distribution | In this paper we propose a novel method to deal with Vector Autoregressive
models, when the Normal-Wishart prior is considered. In particular, we depart
from the current approach of setting $\nu=m+1$ by setting a loss-based prior on
$\nu$. Doing so, we have been able to exploit any information about $\nu$ in
the data and achieve better predictive performances than the method currently
used in the literature. We show how this works both on simulated and real data
sets where, in the latter case, we used data of macroeconometric fashion as
well as viral data. In addition, we show the reason why we believe we achieve a
better performance by showing that the data appears to suggest a value of $\nu$
far from the canonical $m+1$ value.
| stat |
Capacitated spatial clustering with multiple constraints and attributes | Capacitated spatial clustering, a type of unsupervised machine learning
method, is often used to tackle problems in compressing, classifying, logistic
optimization and infrastructure optimization. Depending on the application at
hand, a wide set of extensions may be necessary in clustering.
In this article we propose a number of novel extensions to PACK that is a
novel capacitated spatial clustering method. These extensions are relocation
and location preference of cluster centers, outliers, and non-spatial
attributes. The strength of PACK is that it can consider all of these
extensions jointly. We demonstrate the usefulness PACK with a real world
example in edge computing server placement for a city region with various
different set ups, where we take into consideration outliers, center placement,
and non-spatial attributes. Different setups are evaluated with summary
statistics on spatial proximity and attribute similarity. As a result, the
similarity of the clusters was improved at best by 53%, while simultaneously
the proximity degraded only 18%. In alternate scenarios, both proximity and
similarity were improved. The different extensions proved to provide a valuable
way to include non-spatial information into the cluster analysis, and attain
better overall proximity and similarity. Furthermore, we provide easy-to-use
software tools (rpack) for conducting clustering analyses.
| stat |
Measurement error and precision medicine: error-prone tailoring
covariates in dynamic treatment regimes | Precision medicine incorporates patient-level covariates to tailor treatment
decisions, seeking to improve outcomes. In longitudinal studies with
time-varying covariates and sequential treatment decisions, precision medicine
can be formalized with dynamic treatment regimes (DTRs): sequences of
covariate-dependent treatment rules. To date, the precision medicine literature
has not addressed a ubiquitous concern in health research - measurement error -
where observed data deviate from the truth. We discuss the consequences of
ignoring measurement error in the context of DTRs, focusing on challenges
unique to precision medicine. We show - through simulation and theoretical
results - that relatively simple measurement error correction techniques can
lead to substantial improvements over uncorrected analyses, and apply these
findings to the Sequenced Treatment Alternatives to Relieve Depression (STAR*D)
study.
| stat |
A review of possible effects of cognitive biases on the interpretation
of rule-based machine learning models | While the interpretability of machine learning models is often equated with
their mere syntactic comprehensibility, we think that interpretability goes
beyond that, and that human interpretability should also be investigated from
the point of view of cognitive science. In particular, the goal of this paper
is to discuss to what extent cognitive biases may affect human understanding of
interpretable machine learning models, in particular of logical rules
discovered from data. Twenty cognitive biases are covered, as are possible
debiasing techniques that can be adopted by designers of machine learning
algorithms and software. Our review transfers results obtained in cognitive
psychology to the domain of machine learning, aiming to bridge the current gap
between these two areas. It needs to be followed by empirical studies
specifically focused on the machine learning domain.
| stat |
Joint models as latent Gaussian models - not reinventing the wheel | Joint models have received increasing attention during recent years with
extensions into various directions; numerous hazard functions, different
association structures, linear and non-linear longitudinal trajectories amongst
others. Many of these resulted in new R packages and new formulations of the
joint model. However, a joint model with a linear bivariate Gaussian
association structure is still a latent Gaussian model (LGM) and thus can be
implemented using most existing packages for LGM's. In this paper, we will show
that these joint models can be implemented from a LGM viewpoint using the
R-INLA package. As a particular example, we will focus on the joint model with
a non-linear longitudinal trajectory, recently developed and termed the
partially linear joint model. Instead of the usual spline approach, we argue
for using a Bayesian smoothing spline framework for the joint model that is
stable with respect to knot selection and hence less cumbersome for
practitioners.
| stat |
Amortized Probabilistic Detection of Communities in Graphs | Graph neural networks (GNNs) have reshaped the landscape of models and
algorithms for analyzing data on graphs. Yet despite the many architectures
proposed to encode graph information and perform different tasks, a proper
probabilistic formulation of neural models for community discovery is still
lacking. In this work we fill this gap by combining GNNs with recent
architectures for amortized probabilistic clustering, which we extend with
attention layers. The resulting model is trained with labeled network data
containing varying numbers of communities K. At test time, the model yields
samples from the posterior distribution of community labels for networks of any
size using only O(K) forward evaluations. Apart from quantifying uncertainty in
a principled way, this approach accommodates Bayesian nonparametric priors,
thus bypassing the common requirement to hard code the (maximum) number of
communities in the neural architecture. We illustrate the method on several
synthetic and real datasets, showing superior performance both to previous
approaches to supervised neural community detection and to similar
architectures built using other amortized clustering models.
| stat |
Behavior of linear L2-boosting algorithms in the vanishing learning rate
asymptotic | We investigate the asymptotic behaviour of gradient boosting algorithms when
the learning rate converges to zero and the number of iterations is rescaled
accordingly. We mostly consider L2-boosting for regression with linear base
learner as studied in B{\"u}hlmann and Yu (2003) and analyze also a stochastic
version of the model where subsampling is used at each step (Friedman 2002). We
prove a deterministic limit in the vanishing learning rate asymptotic and
characterize the limit as the unique solution of a linear differential equation
in an infinite dimensional function space. Besides, the training and test error
of the limiting procedure are thoroughly analyzed. We finally illustrate and
discuss our result on a simple numerical experiment where the linear
L2-boosting operator is interpreted as a smoothed projection and time is
related to its number of degrees of freedom.
| stat |
Validation of cluster analysis results on validation data: A systematic
framework | Cluster analysis refers to a wide range of data analytic techniques for class
discovery and is popular in many application fields. To judge the quality of a
clustering result, different cluster validation procedures have been proposed
in the literature. While there is extensive work on classical validation
techniques, such as internal and external validation, less attention has been
given to validating and replicating a clustering result using a validation
dataset. Such a dataset may be part of the original dataset, which is separated
before analysis begins, or it could be an independently collected dataset. We
present a systematic structured framework for validating clustering results on
validation data that includes most existing validation approaches. In
particular, we review classical validation techniques such as internal and
external validation, stability analysis, hypothesis testing, and visual
validation, and show how they can be interpreted in terms of our framework. We
precisely define and formalise different types of validation of clustering
results on a validation dataset and explain how each type can be implemented in
practice. Furthermore, we give examples of how clustering studies from the
applied literature that used a validation dataset can be classified into the
framework.
| stat |
M-type penalized splines for functional linear regression | Functional data analysis is a fast evolving branch of modern statistics.
Despite the popularity of the functional linear model in recent years, current
estimation procedures either suffer from lack of robustness or are
computationally burdensome. To address these shortcomings, we propose a
flexible family of lower-rank smoothers that combines penalized splines and
M-estimation. Under suitable conditions on the design, these estimators exhibit
the same asymptotic properties as the corresponding least-squares estimators,
while being considerably more reliable in the presence of outliers. The
proposed methods can easily be generalized to functional models with additional
functional predictors, scalar covariates or nonparametric components, thus
providing a wide framework of estimation. Empirical investigation shows that
the proposed estimators can combine high efficiency with protection against
outliers, and produce smooth estimates that compare favourably with existing
approaches, robust and non-robust alike.
| stat |
Sparse Principal Component Analysis via Variable Projection | Sparse principal component analysis (SPCA) has emerged as a powerful
technique for modern data analysis, providing improved interpretation of
low-rank structures by identifying localized spatial structures in the data and
disambiguating between distinct time scales. We demonstrate a robust and
scalable SPCA algorithm by formulating it as a value-function optimization
problem. This viewpoint leads to a flexible and computationally efficient
algorithm. Further, we can leverage randomized methods from linear algebra to
extend the approach to the large-scale (big data) setting. Our proposed
innovation also allows for a robust SPCA formulation which obtains meaningful
sparse principal components in spite of grossly corrupted input data. The
proposed algorithms are demonstrated using both synthetic and real world data,
and show exceptional computational efficiency and diagnostic performance.
| stat |
Leave-One-Out Cross-Validation for Bayesian Model Comparison in Large
Data | Recently, new methods for model assessment, based on subsampling and
posterior approximations, have been proposed for scaling leave-one-out
cross-validation (LOO) to large datasets. Although these methods work well for
estimating predictive performance for individual models, they are less powerful
in model comparison. We propose an efficient method for estimating differences
in predictive performance by combining fast approximate LOO surrogates with
exact LOO subsampling using the difference estimator and supply proofs with
regards to scaling characteristics. The resulting approach can be orders of
magnitude more efficient than previous approaches, as well as being better
suited to model comparison.
| stat |
A different perspective of cross-world independence assumption and the
utility of natural effects versus controlled effects | The pure effects described by Robins and Greenland, and later called natural
effects by Pearl, have been criticized because they require a cross-world
independence assumption. In this paper, we use potential outcomes and
sufficient causal sets to present a conceptual perspective of the cross-world
independence assumption that explains why the clinical utility of natural
effects is sometimes greater than that of controlled effects. Our perspective
is consistent with recent work on mediation of natural effects, path specific
effects and separable effects.
| stat |
A Neural Network MCMC sampler that maximizes Proposal Entropy | Markov Chain Monte Carlo (MCMC) methods sample from unnormalized probability
distributions and offer guarantees of exact sampling. However, in the
continuous case, unfavorable geometry of the target distribution can greatly
limit the efficiency of MCMC methods. Augmenting samplers with neural networks
can potentially improve their efficiency. Previous neural network based
samplers were trained with objectives that either did not explicitly encourage
exploration, or used a L2 jump objective which could only be applied to well
structured distributions. Thus it seems promising to instead maximize the
proposal entropy for adapting the proposal to distributions of any shape. To
allow direct optimization of the proposal entropy, we propose a neural network
MCMC sampler that has a flexible and tractable proposal distribution.
Specifically, our network architecture utilizes the gradient of the target
distribution for generating proposals. Our model achieves significantly higher
efficiency than previous neural network MCMC techniques in a variety of
sampling tasks. Further, the sampler is applied on training of a convergent
energy-based model of natural images. The adaptive sampler achieves unbiased
sampling with significantly higher proposal entropy than Langevin dynamics
sampler.
| stat |
Enhancing Explainability of Neural Networks through Architecture
Constraints | Prediction accuracy and model explainability are the two most important
objectives when developing machine learning algorithms to solve real-world
problems. The neural networks are known to possess good prediction performance,
but lack of sufficient model interpretability. In this paper, we propose to
enhance the explainability of neural networks through the following
architecture constraints: a) sparse additive subnetworks; b) projection pursuit
with orthogonality constraint; and c) smooth function approximation. It leads
to an explainable neural network (xNN) with the superior balance between
prediction performance and model interpretability. We derive the necessary and
sufficient identifiability conditions for the proposed xNN model. The multiple
parameters are simultaneously estimated by a modified mini-batch gradient
descent method based on the backpropagation algorithm for calculating the
derivatives and the Cayley transform for preserving the projection
orthogonality. Through simulation study under six different scenarios, we
compare the proposed method to several benchmarks including least absolute
shrinkage and selection operator, support vector machine, random forest,
extreme learning machine, and multi-layer perceptron. It is shown that the
proposed xNN model keeps the flexibility of pursuing high prediction accuracy
while attaining improved interpretability. Finally, a real data example is
employed as a showcase application.
| stat |
Byzantine-Resilient SGD in High Dimensions on Heterogeneous Data | We study distributed stochastic gradient descent (SGD) in the master-worker
architecture under Byzantine attacks. We consider the heterogeneous data model,
where different workers may have different local datasets, and we do not make
any probabilistic assumptions on data generation. At the core of our algorithm,
we use the polynomial-time outlier-filtering procedure for robust mean
estimation proposed by Steinhardt et al. (ITCS 2018) to filter-out corrupt
gradients. In order to be able to apply their filtering procedure in our {\em
heterogeneous} data setting where workers compute {\em stochastic} gradients,
we derive a new matrix concentration result, which may be of independent
interest.
We provide convergence analyses for smooth strongly-convex and non-convex
objectives. We derive our results under the bounded variance assumption on
local stochastic gradients and a {\em deterministic} condition on datasets,
namely, gradient dissimilarity; and for both these quantities, we provide
concrete bounds in the statistical heterogeneous data model. We give a
trade-off between the mini-batch size for stochastic gradients and the
approximation error. Our algorithm can tolerate up to $\frac{1}{4}$ fraction
Byzantine workers. It can find approximate optimal parameters in the
strongly-convex setting exponentially fast and reach to an approximate
stationary point in the non-convex setting with a linear speed, thus, matching
the convergence rates of vanilla SGD in the Byzantine-free setting.
We also propose and analyze a Byzantine-resilient SGD algorithm with gradient
compression, where workers send $k$ random coordinates of their gradients.
Under mild conditions, we show a $\frac{d}{k}$-factor saving in communication
bits as well as decoding complexity over our compression-free algorithm without
affecting its convergence rate (order-wise) and the approximation error.
| stat |
Deep Generative Pattern-Set Mixture Models for Nonignorable Missingness | We propose a variational autoencoder architecture to model both ignorable and
nonignorable missing data using pattern-set mixtures as proposed by Little
(1993). Our model explicitly learns to cluster the missing data into
missingness pattern sets based on the observed data and missingness masks.
Underpinning our approach is the assumption that the data distribution under
missingness is probabilistically semi-supervised by samples from the observed
data distribution. Our setup trades off the characteristics of ignorable and
nonignorable missingness and can thus be applied to data of both types. We
evaluate our method on a wide range of data sets with different types of
missingness and achieve state-of-the-art imputation performance. Our model
outperforms many common imputation algorithms, especially when the amount of
missing data is high and the missingness mechanism is nonignorable.
| stat |
Dynamic modeling of mortality via mixtures of skewed distribution
functions | There has been growing interest on forecasting mortality. In this article, we
propose a novel dynamic Bayesian approach for modeling and forecasting the
age-at-death distribution, focusing on a three-components mixture of a Dirac
mass, a Gaussian distribution and a Skew-Normal distribution. According to the
specified model, the age-at-death distribution is characterized via seven
parameters corresponding to the main aspects of infant, adult and old-age
mortality. The proposed approach focuses on coherent modeling of multiple
countries, and following a Bayesian approach to inference we allow to borrow
information across populations and to shrink parameters towards a common mean
level, implicitly penalizing diverging scenarios. Dynamic modeling across years
is induced trough an hierarchical dynamic prior distribution that allows to
characterize the temporal evolution of each mortality component and to forecast
the age-at-death distribution. Empirical results on multiple countries indicate
that the proposed approach outperforms popular methods for forecasting
mortality, providing interpretable insights on the evolution of mortality.
| stat |
On Finite-Sample Analysis of Offline Reinforcement Learning with Deep
ReLU Networks | This paper studies the statistical theory of offline reinforcement learning
with deep ReLU networks. We consider the off-policy evaluation (OPE) problem
where the goal is to estimate the expected discounted reward of a target policy
given the logged data generated by unknown behaviour policies. We study a
regression-based fitted Q evaluation (FQE) method using deep ReLU networks and
characterize a finite-sample bound on the estimation error of this method under
mild assumptions. The prior works in OPE with either general function
approximation or deep ReLU networks ignore the data-dependent structure in the
algorithm, dodging the technical bottleneck of OPE, while requiring a rather
restricted regularity assumption. In this work, we overcome these limitations
and provide a comprehensive analysis of OPE with deep ReLU networks. In
particular, we precisely quantify how the distribution shift of the offline
data, the dimension of the input space, and the regularity of the system
control the OPE estimation error. Consequently, we provide insights into the
interplay between offline reinforcement learning and deep learning.
| stat |
Semi-supervised learning with an open augmenting unknown class for
cost-effective training and reliable classifications | The ability to (a) train off partially labelled datasets and (b) ensure
resulting networks separate data outside the domain of interest hugely expands
the practical and cost-effective applicability of neural network classifiers.
We design a classifier based off generative adversarial networks (GANs) that
trains off a practical and cost-saving semi-supervised criteria which,
specifically, allows novel classes within the unlabelled training set.
Furthermore, we ensure the resulting classifier is capable of absolute novel
class detection, be these from the semi-supervised unlabelled training set or a
so-called open set. Results are both state-of-the-art and a first of its kind.
We argue this technique greatly decreases training cost in respect to labelling
while greatly improving the reliability of classifications.
| stat |
Re-evaluation of the comparative effectiveness of bootstrap-based
optimism correction methods in the development of multivariable clinical
prediction models | Multivariable predictive models are important statistical tools for providing
synthetic diagnosis and prognostic algorithms based on multiple patients'
characteristics. Their apparent discriminant and calibration measures usually
have overestimation biases (known as 'optimism') relative to the actual
performances for external populations. Existing statistical evidence and
guidelines suggest that three bootstrap-based bias correction methods are
preferable in practice, namely Harrell's bias correction and the .632 and .632+
estimators. Although Harrell's method has been widely adopted in clinical
studies, simulation-based evidence indicates that the .632+ estimator may
perform better than the other two methods. However, there is limited evidence
and these methods' actual comparative effectiveness is still unclear. In this
article, we conducted extensive simulations to compare the effectiveness of
these methods, particularly using the following modern regression models:
conventional logistic regression, stepwise variable selections, Firth's
penalized likelihood method, ridge, lasso, and elastic-net. Under relatively
large sample settings, the three bootstrap-based methods were comparable and
performed well. However, all three methods had biases under small sample
settings, and the directions and sizes of the biases were inconsistent. In
general, the .632+ estimator is recommended, but we provide several notes
concerning the operating characteristics of each method.
| stat |
Computer code validation via mixture model estimation | When computer codes are used for modeling complex physical systems, their
unknown parameters are tuned by calibration techniques. A discrepancy function
may be added to the computer code in order to capture its discrepancy with the
real physical process. By considering the validation question of a computer
code as a Bayesian selection model problem, Damblin et al. (2016) have
highlighted a possible confounding effect in certain configurations between the
code discrepancy and a linear computer code by using a Bayesian testing
procedure based on the intrinsic Bayes factor. In this paper, we investigate
the issue of code error identifiability by applying another Bayesian model
selection technique which has been recently developed by Kamary et al. (2014).
By embedding the competing models within an encompassing mixture model, Kamary
et al. (2014)'s method allows each observation to belong to a different mixing
component, providing a more flexible inference, while remaining competitive in
terms of computational cost with the intrinsic Bayesian approach. By using the
technique of sharing parameters mentioned in Kamary et al. (2014), an improper
non-informative prior can be used for some computer code parameters and we
demonstrate that the resulting posterior distribution is proper. We then check
the sensitivity of our posterior estimates to the choice of the parameter prior
distributions. We illustrate that the value of the correlation length of the
discrepancy Gaussian process prior impacts the Bayesian inference of the
mixture model parameters and that the model discrepancy can be identified by
applying the Kamary et al. (2014) method when the correlation length is not too
small. Eventually, the proposed method is applied on a hydraulic code in an
industrial context.
| stat |
Feature relevance quantification in explainable AI: A causal problem | We discuss promising recent contributions on quantifying feature relevance
using Shapley values, where we observed some confusion on which probability
distribution is the right one for dropped features. We argue that the confusion
is based on not carefully distinguishing between observational and
interventional conditional probabilities and try a clarification based on
Pearl's seminal work on causality. We conclude that unconditional rather than
conditional expectations provide the right notion of dropping features in
contradiction to the theoretical justification of the software package SHAP.
Parts of SHAP are unaffected because unconditional expectations (which we argue
to be conceptually right) are used as approximation for the conditional ones,
which encouraged others to `improve' SHAP in a way that we believe to be
flawed.
| stat |
DCMD: Distance-based Classification Using Mixture Distributions on
Microbiome Data | Current advances in next generation sequencing techniques have allowed
researchers to conduct comprehensive research on microbiome and human diseases,
with recent studies identifying associations between human microbiome and
health outcomes for a number of chronic conditions. However, microbiome data
structure, characterized by sparsity and skewness, presents challenges to
building effective classifiers. To address this, we present an innovative
approach for distance-based classification using mixture distributions (DCMD).
The method aims to improve classification performance when using microbiome
community data, where the predictors are composed of sparse and heterogeneous
count data. This approach models the inherent uncertainty in sparse counts by
estimating a mixture distribution for the sample data, and representing each
observation as a distribution, conditional on observed counts and the estimated
mixture, which are then used as inputs for distance-based classification. The
method is implemented into a k-means and k-nearest neighbours framework and we
identify two distance metrics that produce optimal results. The performance of
the model is assessed using simulations and applied to a human microbiome
study, with results compared against a number of existing machine learning and
distance-based approaches. The proposed method is competitive when compared to
the machine learning approaches and showed a clear improvement over commonly
used distance-based classifiers. The range of applicability and robustness make
the proposed method a viable alternative for classification using sparse
microbiome count data.
| stat |
Handling missing data in large healthcare dataset: a case study of
unknown trauma outcomes | Handling of missed data is one of the main tasks in data preprocessing
especially in large public service datasets. We have analysed data from the
Trauma Audit and Research Network (TARN) database, the largest trauma database
in Europe. For the analysis we used 165,559 trauma cases. Among them, there are
19,289 cases (13.19\%) with unknown outcome. We have demonstrated that these
outcomes are not missed `completely at random' and, hence, it is impossible
just to exclude these cases from analysis despite the large amount of available
data. We have developed a system of non-stationary Markov models for the
handling of missed outcomes and validated these models on the data of 15,437
patients which arrived into TARN hospitals later than 24 hours but within 30
days from injury. We used these Markov models for the analysis of mortality. In
particular, we corrected the observed fraction of death. Two na\"ive approaches
give 7.20\% (available case study) or 6.36\% (if we assume that all unknown
outcomes are `alive'). The corrected value is 6.78\%. Following the seminal
paper of Trunkey (1983) the multimodality of mortality curves has become a much
discussed idea. For the whole analysed TARN dataset the coefficient of
mortality monotonically decreases in time but the stratified analysis of the
mortality gives a different result: for lower severities the coefficient of
mortality is a non-monotonic function of the time after injury and may have
maxima at the second and third weeks. The approach developed here can be
applied to various healthcare datasets which experience the problem of lost
patients and missed outcomes.
| stat |
Two-Sample Testing on Ranked Preference Data and the Role of Modeling
Assumptions | A number of applications require two-sample testing on ranked preference
data. For instance, in crowdsourcing, there is a long-standing question of
whether pairwise comparison data provided by people is distributed similar to
ratings-converted-to-comparisons. Other examples include sports data analysis
and peer grading. In this paper, we design two-sample tests for pairwise
comparison data and ranking data. For our two-sample test for pairwise
comparison data, we establish an upper bound on the sample complexity required
to correctly distinguish between the distributions of the two sets of samples.
Our test requires essentially no assumptions on the distributions. We then
prove complementary lower bounds showing that our results are tight (in the
minimax sense) up to constant factors. We investigate the role of modeling
assumptions by proving lower bounds for a range of pairwise comparison models
(WST, MST,SST, parameter-based such as BTL and Thurstone). We also provide
testing algorithms and associated sample complexity bounds for the problem of
two-sample testing with partial (or total) ranking data.Furthermore, we
empirically evaluate our results via extensive simulations as well as two
real-world datasets consisting of pairwise comparisons. By applying our
two-sample test on real-world pairwise comparison data, we conclude that
ratings and rankings provided by people are indeed distributed differently. On
the other hand, our test recognizes no significant difference in the relative
performance of European football teams across two seasons. Finally, we apply
our two-sample test on a real-world partial and total ranking dataset and find
a statistically significant difference in Sushi preferences across demographic
divisions based on gender, age and region of residence.
| stat |
Evaluation of Granger causality measures for constructing networks from
multivariate time series | Granger causality and variants of this concept allow the study of complex
dynamical systems as networks constructed from multivariate time series. In
this work, a large number of Granger causality measures used to form causality
networks from multivariate time series are assessed. These measures are in the
time domain, such as model-based and information measures, the frequency domain
and the phase domain. The study aims also to compare bivariate and multivariate
measures, linear and nonlinear measures, as well as the use of dimension
reduction in linear model-based measures and information measures. The latter
is particular relevant in the study of high-dimensional time series. For the
performance of the multivariate causality measures, low and high dimensional
coupled dynamical systems are considered in discrete and continuous time, as
well as deterministic and stochastic. The measures are evaluated and ranked
according to their ability to provide causality networks that match the
original coupling structure. The simulation study concludes that the Granger
causality measures using dimension reduction are superior and should be
preferred particularly in studies involving many observed variables, such as
multi-channel electroencephalograms and financial markets.
| stat |
Hierarchical Bayesian Nearest Neighbor Co-Kriging Gaussian Process
Models; An Application to Intersatellite Calibration | Recent advancements in remote sensing technology and the increasing size of
satellite constellations allows massive geophysical information to be gathered
daily on a global scale by numerous platforms of different fidelity. The
auto-regressive co-kriging model is a suitable framework to analyse such data
sets because it accounts for cross-dependencies among different fidelity
satellite outputs. However, its implementation in multifidelity large spatial
data-sets is practically infeasible because its computational complexity
increases cubically with the total number of observations. In this paper, we
propose a nearest neighbour co-kriging Gaussian process that couples the
auto-regressive model and nearest neighbour GP by using augmentation ideas;
reducing the computational complexity to be linear with the total number of
spatial observed locations. The latent process of the nearest neighbour GP is
augmented in a manner which allows the specification of semi-conjugate priors.
This facilitates the design of an efficient MCMC sampler involving mostly
direct sampling updates which can be implemented in parallel computational
environments. The good predictive performance of the proposed method is
demonstrated in a simulation study. We use the proposed method to analyze
High-resolution Infrared Radiation Sounder data gathered from two NOAA polar
orbiting satellites.
| stat |
Certified Robustness to Adversarial Examples with Differential Privacy | Adversarial examples that fool machine learning models, particularly deep
neural networks, have been a topic of intense research interest, with attacks
and defenses being developed in a tight back-and-forth. Most past defenses are
best effort and have been shown to be vulnerable to sophisticated attacks.
Recently a set of certified defenses have been introduced, which provide
guarantees of robustness to norm-bounded attacks, but they either do not scale
to large datasets or are limited in the types of models they can support. This
paper presents the first certified defense that both scales to large networks
and datasets (such as Google's Inception network for ImageNet) and applies
broadly to arbitrary model types. Our defense, called PixelDP, is based on a
novel connection between robustness against adversarial examples and
differential privacy, a cryptographically-inspired formalism, that provides a
rigorous, generic, and flexible foundation for defense.
| stat |
Random boosting and random^2 forests -- A random tree depth injection
approach | The induction of additional randomness in parallel and sequential ensemble
methods has proven to be worthwhile in many aspects. In this manuscript, we
propose and examine a novel random tree depth injection approach suitable for
sequential and parallel tree-based approaches including Boosting and Random
Forests. The resulting methods are called \emph{Random Boost} and
\emph{Random$^2$ Forest}. Both approaches serve as valuable extensions to the
existing literature on the gradient boosting framework and random forests. A
Monte Carlo simulation, in which tree-shaped data sets with different numbers
of final partitions are built, suggests that there are several scenarios where
\emph{Random Boost} and \emph{Random$^2$ Forest} can improve the prediction
performance of conventional hierarchical boosting and random forest approaches.
The new algorithms appear to be especially successful in cases where there are
merely a few high-order interactions in the generated data. In addition, our
simulations suggest that our random tree depth injection approach can improve
computation time by up to 40%, while at the same time the performance losses in
terms of prediction accuracy turn out to be minor or even negligible in most
cases.
| stat |
Collection of Historical Weather Data: Issues with Missing Values | Weather data collected from automated weather stations have become a crucial
component for making decisions in agriculture and in forestry. Over time,
weather stations may become out-of-order or stopped for maintenance, and
therefore, during those periods, the data values will be missing.
Unfortunately, this will cause huge problems when analysing the data. The main
aim of this study is to create high-quality historical weather datasets by
dealing efficiently with missing values. In this paper, we present a set of
missing data imputation methods and study their effectiveness. These methods
were used based on different types of missing values. The experimental results
show that two the proposed methods are very promising and can be used at larger
scale.
| stat |
Heterogeneous Dense Subhypergraph Detection | We study the problem of testing the existence of a heterogeneous dense
subhypergraph. The null hypothesis corresponds to a heterogeneous
Erd\"{o}s-R\'{e}nyi uniform random hypergraph and the alternative hypothesis
corresponds to a heterogeneous uniform random hypergraph that contains a dense
subhypergraph. We establish detection boundaries when the edge probabilities
are known and construct an asymptotically powerful test for distinguishing the
hypotheses. We also construct an adaptive test which does not involve edge
probabilities, and hence, is more practically useful.
| stat |
TaskNorm: Rethinking Batch Normalization for Meta-Learning | Modern meta-learning approaches for image classification rely on increasingly
deep networks to achieve state-of-the-art performance, making batch
normalization an essential component of meta-learning pipelines. However, the
hierarchical nature of the meta-learning setting presents several challenges
that can render conventional batch normalization ineffective, giving rise to
the need to rethink normalization in this setting. We evaluate a range of
approaches to batch normalization for meta-learning scenarios, and develop a
novel approach that we call TaskNorm. Experiments on fourteen datasets
demonstrate that the choice of batch normalization has a dramatic effect on
both classification accuracy and training time for both gradient based and
gradient-free meta-learning approaches. Importantly, TaskNorm is found to
consistently improve performance. Finally, we provide a set of best practices
for normalization that will allow fair comparison of meta-learning algorithms.
| stat |
Asymmetric Heavy Tails and Implicit Bias in Gaussian Noise Injections | Gaussian noise injections (GNIs) are a family of simple and widely-used
regularisation methods for training neural networks, where one injects additive
or multiplicative Gaussian noise to the network activations at every iteration
of the optimisation algorithm, which is typically chosen as stochastic gradient
descent (SGD). In this paper we focus on the so-called `implicit effect' of
GNIs, which is the effect of the injected noise on the dynamics of SGD. We show
that this effect induces an asymmetric heavy-tailed noise on SGD gradient
updates. In order to model this modified dynamics, we first develop a
Langevin-like stochastic differential equation that is driven by a general
family of asymmetric heavy-tailed noise. Using this model we then formally
prove that GNIs induce an `implicit bias', which varies depending on the
heaviness of the tails and the level of asymmetry. Our empirical results
confirm that different types of neural networks trained with GNIs are
well-modelled by the proposed dynamics and that the implicit effect of these
injections induces a bias that degrades the performance of networks.
| stat |
Predicting competitions by pairing conditional logistic regression and
subjective Bayes: An Academy Awards case study | Predicting the outcome of elections, sporting events, entertainment awards,
and other competitions has long captured the human imagination. Such prediction
is growing in sophistication in these areas, especially in the rapidly growing
field of data-driven journalism intended for a general audience as the
availability of historical information rapidly balloons. Providing statistical
methodology to probabilistically predict competition outcomes faces two main
challenges. First, a suitably general modeling approach is necessary to assign
probabilities to competitors. Second, the modeling framework must be able to
accommodate expert opinion, which is usually available but difficult to fully
encapsulate in typical data sets. We overcome these challenges with a combined
conditional logistic regression/subjective Bayes approach. To illustrate the
method, we re-analyze data from a recent Time.com piece in which the authors
attempted to predict the 2019 Best Picture Academy Award winner using standard
logistic regression. Towards engaging and educating a broad readership, we
discuss strategies to deploy the proposed method via an online application.
| stat |
Influence of weather natural variability on the thermal characterisation
of a building envelope | The thermal characterisation of a building envelope is usually best performed
from on site measurements with controlled heating power set points.
Occupant-friendly measurement conditions provide on the contrary less
informative data. Notwithstanding occupancy, the boundary conditions alone
contribute to a greater extent to the energy balance. Non intrusive conditions
question therefore the repeatability and relevance of such experiment.This
paper proposes an original numerical methodology to assess the repeatability
and accuracy of the estimation of an envelope's overall thermalresistance under
variable weather conditions. A comprehensive building energy model serves as
reference model to produce multiple synthetic datasets. Each is run with a
different weather dataset from a single location and serves for the calibration
of an appropriate model, which provides a thermal resistance estimate. The
estimate's accuracy is then assessed in the light of the particular weather
conditions that served for data generation. The originality also lies in the
use of stochastically generated weather datasets to perform an uncertaintyand
global sensitivity analysis of all estimates with respect to 6 weather
variables.The methodology is applied on simulated data from a one-storey house
case study serving as reference model. The thermal resistance estimations are
inferred from calibrated stochastic RC models. It is found that 11 days are
necessary to achieve robust estimations. The large air change rate in the case
study explains why the outdoor temperature and the wind speed are found highly
influential.
| stat |
How to choose between different Bayesian posterior indices for
hypothesis testing in practice | Hypothesis testing is an essential statistical method in psychology and the
cognitive sciences. The problems of traditional null hypothesis significance
testing (NHST) have been discussed widely, and among the proposed solutions to
the replication problems caused by the inappropriate use of significance tests
and $p$-values is a shift towards Bayesian data analysis. However, Bayesian
hypothesis testing is concerned with various posterior indices for significance
and the size of an effect. This complicates Bayesian hypothesis testing in
practice, as the availability of multiple Bayesian alternatives to the
traditional $p$-value causes confusion which one to select and why. In this
paper, we compare various Bayesian posterior indices which have been proposed
in the literature and discuss their benefits and limitations. Our comparison
shows that conceptually not all proposed Bayesian alternatives to NHST and
$p$-values are beneficial, and the usefulness of some indices strongly depends
on the study design and research goal. However, our comparison also reveals
that there exist at least two candidates among the available Bayesian posterior
indices which have appealing theoretical properties and are, to our best
knowledge, widely underused among psychologists.
| stat |
Identification of Underlying Dynamic System from Noisy Data with Splines | In this paper, we propose a two-stage method called Spline Assisted Partial
Differential Equation involved Model Identification (SAPDEMI) to efficiently
identify the underlying partial differential equation (PDE) models from the
noisy data. In the first stage -- functional estimation stage -- we employ the
cubic spline to estimate the unobservable derivatives, which serve as
candidates included the underlying PDE models. The contribution of this stage
is that, it is computational efficient because it only requires the
computational complexity of the linear polynomial of the sample size, which
achieves the lowest possible order of complexity. In the second stage -- model
identification stage -- we apply Least Absolute Shrinkage and Selection
Operator (Lasso) to identify the underlying PDE models. The contribution of
this stage is that, we focus on the model selections, while the existing
literature mostly focuses on parameter estimations. Moreover, we develop
statistical properties of our method for correct identification, where the main
tool we use is the primal-dual witness (PDW) method. Finally, we validate our
theory through various numerical examples.
| stat |
Hole or grain? A Section Pursuit Index for Finding Hidden Structure in
Multiple Dimensions | Multivariate data is often visualized using linear projections, produced by
techniques such as principal component analysis, linear discriminant analysis,
and projection pursuit. A problem with projections is that they obscure low and
high density regions near the center of the distribution. Sections, or slices,
can help to reveal them. This paper develops a section pursuit method, building
on the extensive work in projection pursuit, to search for interesting slices
of the data. Linear projections are used to define sections of the parameter
space, and to calculate interestingness by comparing the distribution of
observations, inside and outside a section. By optimizing this index, it is
possible to reveal features such as holes (low density) or grains (high
density). The optimization is incorporated into a guided tour so that the
search for structure can be dynamic. The approach can be useful for problems
when data distributions depart from uniform or normal, as in visually exploring
nonlinear manifolds, and functions in multivariate space. Two applications of
section pursuit are shown: exploring decision boundaries from classification
models, and exploring subspaces induced by complex inequality conditions from
multiple parameter model. The new methods are available in R, in the tourr
package.
| 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.