title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Sparsity-Agnostic Lasso Bandit | We consider a stochastic contextual bandit problem where the dimension $d$ of
the feature vectors is potentially large, however, only a sparse subset of
features of cardinality $s_0 \ll d$ affect the reward function. Essentially all
existing algorithms for sparse bandits require a priori knowledge of the value
of the sparsity index $s_0$. This knowledge is almost never available in
practice, and misspecification of this parameter can lead to severe
deterioration in the performance of existing methods. The main contribution of
this paper is to propose an algorithm that does not require prior knowledge of
the sparsity index $s_0$ and establish tight regret bounds on its performance
under mild conditions. We also comprehensively evaluate our proposed algorithm
numerically and show that it consistently outperforms existing methods, even
when the correct sparsity index is revealed to them but is kept hidden from our
algorithm.
| stat |
Multi-Armed Bandits with Correlated Arms | We consider a multi-armed bandit framework where the rewards obtained by
pulling different arms are correlated. We develop a unified approach to
leverage these reward correlations and present fundamental generalizations of
classic bandit algorithms to the correlated setting. We present a unified proof
technique to analyze the proposed algorithms. Rigorous analysis of C-UCB and
C-TS (the correlated bandit versions of Upper-confidence-bound and Thompson
sampling) reveals that the algorithms end up pulling certain sub-optimal arms,
termed as non-competitive, only O(1) times, as opposed to the O(log T) pulls
required by classic bandit algorithms such as UCB, TS etc. We present
regret-lower bound and show that when arms are correlated through a latent
random source, our algorithms obtain order-optimal regret. We validate the
proposed algorithms via experiments on the MovieLens and Goodreads datasets,
and show significant improvement over classical bandit algorithms.
| stat |
On the training dynamics of deep networks with $L_2$ regularization | We study the role of $L_2$ regularization in deep learning, and uncover
simple relations between the performance of the model, the $L_2$ coefficient,
the learning rate, and the number of training steps. These empirical relations
hold when the network is overparameterized. They can be used to predict the
optimal regularization parameter of a given model. In addition, based on these
observations we propose a dynamical schedule for the regularization parameter
that improves performance and speeds up training. We test these proposals in
modern image classification settings. Finally, we show that these empirical
relations can be understood theoretically in the context of infinitely wide
networks. We derive the gradient flow dynamics of such networks, and compare
the role of $L_2$ regularization in this context with that of linear models.
| stat |
Better Model Selection with a new Definition of Feature Importance | Feature importance aims at measuring how crucial each input feature is for
model prediction. It is widely used in feature engineering, model selection and
explainable artificial intelligence (XAI). In this paper, we propose a new
tree-model explanation approach for model selection. Our novel concept
leverages the Coefficient of Variation of a feature weight (measured in terms
of the contribution of the feature to the prediction) to capture the dispersion
of importance over samples. Extensive experimental results show that our novel
feature explanation performs better than general cross validation method in
model selection both in terms of time efficiency and accuracy performance.
| stat |
Certifiably Optimal Sparse Sufficient Dimension Reduction | Sufficient dimension reduction (SDR) is a popular tool in regression
analysis, which replaces the original predictors with a minimal set of their
linear combinations. However, the estimated linear combinations generally
contain all original predictors, which brings difficulties in interpreting the
results, especially when the number of predictors is large. In this paper, we
propose a customized branch and bound algorithm, optimal sparse generalized
eigenvalue problem (Optimal SGEP), which combines a SGEP formulation of many
SDR methods and efficient and accurate bounds allowing the algorithm to
converge quickly. Optimal SGEP exactly solves the underlying non-convex
optimization problem and thus produces certifiably optimal solutions. We
demonstrate the effectiveness of the proposed algorithm through simulation
studies.
| stat |
Meta-Learning Acquisition Functions for Transfer Learning in Bayesian
Optimization | Transferring knowledge across tasks to improve data-efficiency is one of the
open key challenges in the field of global black-box optimization. Readily
available algorithms are typically designed to be universal optimizers and,
therefore, often suboptimal for specific tasks. We propose a novel transfer
learning method to obtain customized optimizers within the well-established
framework of Bayesian optimization, allowing our algorithm to utilize the
proven generalization capabilities of Gaussian processes. Using reinforcement
learning to meta-train an acquisition function (AF) on a set of related tasks,
the proposed method learns to extract implicit structural information and to
exploit it for improved data-efficiency. We present experiments on a
simulation-to-real transfer task as well as on several synthetic functions and
on two hyperparameter search problems. The results show that our algorithm (1)
automatically identifies structural properties of objective functions from
available source tasks or simulations, (2) performs favourably in settings with
both scarse and abundant source data, and (3) falls back to the performance
level of general AFs if no particular structure is present.
| stat |
Simple estimators for network sampling | A new estimation method is presented for network sampling designs, including
Respondent Driven Sampling (RDS) and Snowball (SB) sampling. These types of
link-tracing designs are essential for studies of hidden populations, such as
people at risk for HIV. The simple idea behind the new method is to run a
fast-sampling process on the sample network data to estimate the inclusion
probabilities of the actual survey, and incorporate those in unequal
probability estimators of population means and proportions. Improved versions
of the usual RDS and SB designs are also proposed, termed RDS+ and SB+, to
obtain information on more of the within-sample links. In simulations using the
network from the Colorado Springs study on the heterosexual spread of HIV, the
new estimators produce in most cases lower bias and lower mean square than
current methods. For the variables having the largest mean square errors with
current estimators, the improvement with the new estimator is dramatic. The
estimates are improved even more with the enhanced design versions. For
estimating the population mean degree, the efficiency gains using he new method
are 29 for RDS, 54 for RDS+, 26 for SB and 80 for SB+. This means for example,
with the ordinary RDS design, the mean square error with the new estimator,
same data, is 1/29 that of currently used estimators. The new method is
computationally intricate but is fast and scales up well. The new estimation
method can be used to re-analyze existing network survey data. For new network
sampling studies, it is recommended to use the improved designs as well as the
new estimators.
| stat |
Log-concave Ridge Estimation | We develop a density ridge search algorithm based on a novel density ridge
definition. This definition is based on a conditional variance matrix and the
mode in the lower dimensional subspace. It is compared to the subspace
constraint mean shift algorithm, based on the gradient and Hessian of the
underlying probability density function. We show the advantages of the new
algorithm in a simulation study and estimate galaxy filaments from a data set
of the Baryon Oscillation Spectroscopic Survey.
| stat |
Minimum Relative Entropy Inference for Normal and Monte Carlo
Distributions | We represent affine sub-manifolds of exponential family distributions as
minimum relative entropy sub-manifolds. With such representation we derive
analytical formulas for the inference from partial information on expectations
and covariances of multivariate normal distributions; and we improve the
numerical implementation via Monte Carlo simulations for the inference from
partial information of generalized expectation type.
| stat |
Penalized Estimation and Forecasting of Multiple Subject Intensive
Longitudinal Data | Intensive Longitudinal Data (ILD) is an increasingly common data type in the
social and behavioral sciences. Despite the many benefits these data provide,
little work has been dedicated to realizing the potential such data hold for
forecasting dynamic processes at the individual level. To address this gap in
the literature we present the multi-VAR framework, a novel methodological
approach for penalized estimation and forecasting of ILD collected from
multiple individuals. Importantly, our approach estimates models for all
individuals simultaneously and is capable of adaptively adjusting to the amount
of heterogeneity exhibited across individual dynamic processes. To accomplish
this we propose proximal gradient descent algorithm for solving the multi-VAR
problem and prove the consistency of the recovered transition matrices. We
evaluate the forecasting performance of our method in comparison with a number
of benchmark forecasting methods and provide an illustrative example involving
the day-to-day emotional experiences of 16 individuals over an 11-week period.
| stat |
Using Simulation to Analyze Interrupted Time Series Designs | We are sometimes forced to use the Interrupted Time Series (ITS) design as an
identification strategy for potential policy change, such as when we only have
a single treated unit and no comparable controls. For example, with recent
county- and state-wide criminal justice reform efforts, where judicial bodies
have changed bail setting practices for everyone in their jurisdiction in order
to reduce rates of pre-trial detention while maintaining court order and public
safety, we have no natural comparison group other than the past. In these
contexts, it is imperative to model pre-policy trends with a light touch,
allowing for structures such as autoregressive departures from any pre-existing
trend, in order to accurately and realistically assess the statistical
uncertainty of our projections (beyond the stringent assumptions necessary for
the subsequent causal inferences). To tackle this problem we provide a
methodological approach rooted in commonly understood and used modeling
approaches that better captures uncertainty. We quantify uncertainty with
simulation, generating a distribution of plausible counterfactual trajectories
to compare to the observed; this approach naturally allows for incorporating
seasonality and other time varying covariates, and provides confidence
intervals along with point estimates for the potential impacts of policy
change. We find simulation provides a natural framework to capture and show
uncertainty in the ITS designs. It also allows for easy extensions such as
nonparametric smoothing in order to handle multiple post-policy time points or
more structural models to account for seasonality.
| stat |
Doubly Robust Direct Learning for Estimating Conditional Average
Treatment Effect | Inferring the heterogeneous treatment effect is a fundamental problem in the
sciences and commercial applications. In this paper, we focus on estimating
Conditional Average Treatment Effect (CATE), that is, the difference in the
conditional mean outcome between treatments given covariates. Traditionally,
Q-Learning based approaches rely on the estimation of conditional mean outcome
given treatment and covariates. However, they are subject to misspecification
of the main effect model. Recently, simple and flexible one-step methods to
directly learn (D-Learning) the CATE without model specifications have been
proposed. However, these methods are not robust against misspecification of the
propensity score model. We propose a new framework for CATE estimation, robust
direct learning (RD-Learning), leading to doubly robust estimators of the
treatment effect. The consistency for our CATE estimator is guaranteed if
either the main effect model or the propensity score model is correctly
specified. The framework can be used in both the binary and the multi-arm
settings and is general enough to allow different function spaces and
incorporate different generic learning algorithms. As a by-product, we develop
a competitive statistical inference tool for the treatment effect, assuming the
propensity score is known. We provide theoretical insights to the proposed
method using risk bounds under both linear and non-linear settings. The
effectiveness of our proposed method is demonstrated by simulation studies and
a real data example about an AIDS Clinical Trials study.
| stat |
Three-dimensional weights of evidence modeling of a deep-seated porphyry
Cu deposit | Given the challenges in data acquisition and modeling at the stage of
detailed exploration, it is difficult to develop a prospectivity model,
particularly for disseminated ore deposits. Recently, the weights of evidence
(WofE) method has demonstrated a high efficiency for modeling such deposits. In
this study, we propose a framework for creating a three-dimensional (3D)
weights of evidence-based prospectivity model of the Nochoun porphyry Cu
deposit in the Urmia-Dokhtar magmatic arc of Iran. The input data include
qualitative geological and quantitative geochemical information obtained from
boreholes and field observations. We combine ordinary and fuzzy weights of
evidence for integrating qualitative and quantitative exploration criteria in a
3D space constrained by a metallogenic model of the study area for identifying
a deep-seated ore body. Ordinary weights of evidence are determined for
geological data, including lithology, alteration, rock type, and structure.
Moreover, we determine the fuzzy weight of evidence for each class of
continuous geochemical models created based on Fe, Mo, and Zn concentration
values derived from boreholes. We integrate the input evidential models using
WofE and create two prospectivity models (i.e., posterior and studentized
posterior probability). We also determine anomalous voxels in the probability
models using concentration-volume fractal models and validate them using
prediction-volume plots. The modeling results indicate that the studentized
posterior probability model is more efficient in identifying voxels
representing copper mineralized rock volumes. We provide open source software
for the proposed framework which can be used for exploring deep-seated ore
bodies in other regions.
| stat |
Using Subset Log-Likelihoods to Trim Outliers in Gaussian Mixture Models | Mixtures of Gaussian distributions are a popular choice in model-based
clustering. Outliers can affect parameters estimation and, as such, must be
accounted for. Predicting the proportion of outliers correctly is paramount as
it minimizes misclassification error. It is proved that, for a finite Gaussian
mixture model, the log-likelihoods of the subset models are distributed
according to a mixture of beta distributions. An algorithm is then proposed
that predicts the proportion of outliers by measuring the adherence of a set of
subset log-likelihoods to a beta mixture reference distribution. This algorithm
removes the least likely points, which are deemed outliers, until model
assumptions are met.
| stat |
BdryGP: a new Gaussian process model for incorporating boundary
information | Gaussian processes (GPs) are widely used as surrogate models for emulating
computer code, which simulate complex physical phenomena. In many problems,
additional boundary information (i.e., the behavior of the phenomena along
input boundaries) is known beforehand, either from governing physics or
scientific knowledge. While there has been recent work on incorporating
boundary information within GPs, such models do not provide theoretical
insights on improved convergence rates. To this end, we propose a new GP model,
called BdryGP, for incorporating boundary information. We show that BdryGP not
only has improved convergence rates over existing GP models (which do not
incorporate boundaries), but is also more resistant to the
"curse-of-dimensionality" in nonparametric regression. Our proofs make use of a
novel connection between GP interpolation and finite-element modeling.
| stat |
Markov chain Monte Carlo algorithms with sequential proposals | We explore a general framework in Markov chain Monte Carlo (MCMC) sampling
where sequential proposals are tried as a candidate for the next state of the
Markov chain. This sequential-proposal framework can be applied to various
existing MCMC methods, including Metropolis-Hastings algorithms using random
proposals and methods that use deterministic proposals such as Hamiltonian
Monte Carlo (HMC) or the bouncy particle sampler. Sequential-proposal MCMC
methods construct the same Markov chains as those constructed by the delayed
rejection method under certain circumstances. In the context of HMC, the
sequential-proposal approach has been proposed as extra chance generalized
hybrid Monte Carlo (XCGHMC). We develop two novel methods in which the
trajectories leading to proposals in HMC are automatically tuned to avoid
doubling back, as in the No-U-Turn sampler (NUTS). The numerical efficiency of
these new methods compare favorably to the NUTS. We additionally show that the
sequential-proposal bouncy particle sampler enables the constructed Markov
chain to pass through regions of low target density and thus facilitates better
mixing of the chain when the target density is multimodal.
| stat |
Dealing With Ratio Metrics in A/B Testing at the Presence of Intra-User
Correlation and Segments | We study ratio metrics in A/B testing at the presence of correlation among
observations coming from the same user and provides practical guidance
especially when two metrics contradict each other. We propose new estimating
methods to quantitatively measure the intra-user correlation (within segments).
With the accurately estimated correlation, a uniformly minimum-variance
unbiased estimator of the population mean, called correlation-adjusted mean, is
proposed to account for such correlation structure. It is proved theoretically
and numerically better than the other two unbiased estimators, naive mean and
normalized mean (averaging within users first and then across users). The
correlation-adjusted mean method is unbiased and has reduced variance so it
gains additional power. Several simulation studies are designed to show the
estimation accuracy of the correlation structure, effectiveness in reducing
variance, and capability of obtaining more power. An application to the eBay
data is conducted to conclude this paper.
| stat |
A Supervised Hybrid Statistical Catch-up System Built on Gabece Gambian
Data | In this paper we want to find a statistical rule that assigns a passing or
failing grade to students who undertook at least three exams out of four in a
national exam, instead of completely dismissing them students. While it is
cruel to declare them as failing, especially if the reason for their absence it
not intentional, they should have demonstrated enough merit in the three exams
taken to deserve a chance to be declared passing. We use a special
classification method and nearest neighbors methods based on the average grade
and on the most modal grade to build a statistical rule in a supervised
learning process. The study is built on the national GABECE educational data
which is a considerable data covering seven years and all the six regions of
the Gambia.
| stat |
DAG With Omitted Objects Displayed (DAGWOOD): A framework for revealing
causal assumptions in DAGs | Directed acyclic graphs (DAGs) are frequently used in epidemiology as a guide
to encode causal inference assumptions. We propose the DAGWOOD framework to
bring many of those encoded assumptions to the forefront.
DAGWOOD combines a root DAG (the DAG in the proposed analysis) and a set of
branch DAGs (alternative hidden assumptions to the root DAG). All branch DAGs
share a common ruleset, and must 1) change the root DAG, 2) be a valid DAG, and
either 3a) change the minimally sufficient adjustment set or 3b) change the
number of frontdoor paths. Branch DAGs comprise a list of assumptions which
must be justified as negligible. We define two types of branch DAGs: exclusion
branch DAGs add a single- or bidirectional pathway between two nodes in the
root DAG (e.g. direct pathways and colliders), while misdirection branch DAGs
represent alternative pathways that could be drawn between objects (e.g.,
creating a collider by reversing the direction of causation for a controlled
confounder).
The DAGWOOD framework 1) organizes causal model assumptions, 2) reinforces
best DAG practices, 3) provides a framework for evaluation of causal models,
and 4) can be used for generating causal models.
| stat |
A Bayesian Dynamic Graphical Model for Recurrent Events in Public Health | To analyze the impacts of certain types of public health interventions we
need to estimate the treatment effects and outcomes as these apply to
heterogeneous open populations. Dynamically modifying populations containing
risk groups that can react very differently to changes in covariates are
inferentially challenging. Here we propose a novel Bayesian graphical model
called the Reduced Dynamic Chain Event Graph (RDCEG) customized to such
populations. These models generalize the tree-based Chain Event Graphs to a
particular class of graphically supported semi-Markov processes. They provide
an interface between natural language explanations about what might be
happening to individuals and a formal statistical analysis. Here we show how
the RDCEG is able to express the different possible progressions of each
vulnerable individual as well as hypotheses about probabilistic symmetries
within these progressions across different individuals within that population.
We demonstrate how well-developed Bayesian Network technologies can be
transferred almost seamlessly to this class. Our work is motivated by the
challenge of modeling non-pharmacological interventions for recurrent event
processes. We illustrate our methodology in two settings: an intervention to
reduce falls among the elderly and a trial to examine effects of deferred
treatment among individuals presenting with early epilepsy.
| stat |
Identifying Statistical Bias in Dataset Replication | Dataset replication is a useful tool for assessing whether improvements in
test accuracy on a specific benchmark correspond to improvements in models'
ability to generalize reliably. In this work, we present unintuitive yet
significant ways in which standard approaches to dataset replication introduce
statistical bias, skewing the resulting observations. We study ImageNet-v2, a
replication of the ImageNet dataset on which models exhibit a significant
(11-14%) drop in accuracy, even after controlling for a standard
human-in-the-loop measure of data quality. We show that after correcting for
the identified statistical bias, only an estimated $3.6\% \pm 1.5\%$ of the
original $11.7\% \pm 1.0\%$ accuracy drop remains unaccounted for. We conclude
with concrete recommendations for recognizing and avoiding bias in dataset
replication. Code for our study is publicly available at
http://github.com/MadryLab/dataset-replication-analysis .
| stat |
L\'{e}vy Adaptive B-spline Regression via Overcomplete Systems | The estimation of functions with varying degrees of smoothness is a
challenging problem in the nonparametric function estimation. In this paper, we
propose the LABS (L\'{e}vy Adaptive B-Spline regression) model, an extension of
the LARK models, for the estimation of functions with varying degrees of
smoothness. LABS model is a LARK with B-spline bases as generating kernels. The
B-spline basis consists of piecewise k degree polynomials with k-1 continuous
derivatives and can express systematically functions with varying degrees of
smoothness. By changing the orders of the B-spline basis, LABS can
systematically adapt the smoothness of functions, i.e., jump discontinuities,
sharp peaks, etc. Results of simulation studies and real data examples support
that this model catches not only smooth areas but also jumps and sharp peaks of
functions. The proposed model also has the best performance in almost all
examples. Finally, we provide theoretical results that the mean function for
the LABS model belongs to the certain Besov spaces based on the orders of the
B-spline basis and that the prior of the model has the full support on the
Besov spaces.
| stat |
Double-Counting Problem of the Bonus-Malus System | The bonus-malus system (BMS) is a widely used premium adjustment mechanism
based on policyholder's claim history. Most auto insurance BMSs assume that
policyholders in the same bonus-malus (BM) level share the same a posteriori
risk adjustment. This system reflects the policyholder's claim history in a
relatively simple manner. However, the typical system follows a single BM scale
and is known to suffer from the double-counting problem: policyholders in the
high-risk classes in terms of a priori characteristics are penalized too
severely (Taylor, 1997; Pitrebois et al., 2003). Thus, Pitrebois et al. (2003)
proposed a new system with multiple BM scales based on the a priori
characteristics. While this multiple-scale BMS removes the double-counting
problem, it loses the prime benefit of simplicity. Alternatively, we argue that
the double-counting problem can be viewed as an inefficiency of the
optimization process. Furthermore, we show that the double-counting problem can
be resolved by fully optimizing the BMS setting, but retaining the traditional
BMS format.
| stat |
On the probability of a causal inference is robust for internal validity | The internal validity of observational study is often subject to debate. In
this study, we define the counterfactuals as the unobserved sample and intend
to quantify its relationship with the null hypothesis statistical testing
(NHST). We propose the probability of a causal inference is robust for internal
validity, i.e., the PIV, as a robustness index of causal inference. Formally,
the PIV is the probability of rejecting the null hypothesis again based on both
the observed sample and the counterfactuals, provided the same null hypothesis
has already been rejected based on the observed sample. Under either
frequentist or Bayesian framework, one can bound the PIV of an inference based
on his bounded belief about the counterfactuals, which is often needed when the
unconfoundedness assumption is dubious. The PIV is equivalent to statistical
power when the NHST is thought to be based on both the observed sample and the
counterfactuals. We summarize the process of evaluating internal validity with
the PIV into an eight-step procedure and illustrate it with an empirical
example (i.e., Hong and Raudenbush (2005)).
| stat |
Implementing Monte Carlo Tests with P-value Buckets | Software packages usually report the results of statistical tests using
p-values. Users often interpret these by comparing them to standard thresholds,
e.g. 0.1%, 1% and 5%, which is sometimes reinforced by a star rating (***, **,
*). We consider an arbitrary statistical test whose p-value p is not available
explicitly, but can be approximated by Monte Carlo samples, e.g. by bootstrap
or permutation tests. The standard implementation of such tests usually draws a
fixed number of samples to approximate p. However, the probability that the
exact and the approximated p-value lie on different sides of a threshold (the
resampling risk) can be high, particularly for p-values close to a threshold.
We present a method to overcome this. We consider a finite set of
user-specified intervals which cover [0,1] and which can be overlapping. We
call these p-value buckets. We present algorithms that, with arbitrarily high
probability, return a p-value bucket containing p. We prove that for both a
bounded resampling risk and a finite runtime, overlapping buckets need to be
employed, and that our methods both bound the resampling risk and guarantee a
finite runtime for such overlapping buckets. To interpret decisions with
overlapping buckets, we propose an extension of the star rating system. We
demonstrate that our methods are suitable for use in standard software,
including for low p-value thresholds occurring in multiple testing settings,
and that they can be computationally more efficient than standard
implementations.
| stat |
Statistical Assessment of Safety Levels of Railway Operators | Recently the European Union Agency for Railways (ERA) has received a mandate
for 'the development of common safety methods for assessing the safety level
and the safety performance of railway operators at national and Union level'.
Currently, several methods are under development. It is of interest how a
possible candidate would behave and what would be the advantages and
disadvantages of a particular method. In this paper, we study a version of the
procedure. On the one hand side we analyze it based on the theory of
mathematical statistics. As a result, we present a statistically efficient
method the rate-ratio test based on a quantity that has smaller variance than
the quantity handled by the ERA. Then, we support the theoretical results with
the help of a simple simulation study in order to estimate failure
probabilities of the first and second kinds. We construct such alternative
distributions which the decision procedure cannot distinguish. We will show
that the use of procedures that are optimal in the sense of mathematical
statistics combined with the use of a characteristics that has small spread,
here the number of accidents, is advantageous.
| stat |
MCMC Algorithms for Posteriors on Matrix Spaces | We study Markov chain Monte Carlo (MCMC) algorithms for target distributions
defined on matrix spaces. Such an important sampling problem has yet to be
analytically explored. We carry out a major step in covering this gap by
developing the proper theoretical framework that allows for the identification
of ergodicity properties of typical MCMC algorithms, relevant in such a
context. Beyond the standard Random-Walk Metropolis (RWM) and preconditioned
Crank--Nicolson (pCN), a contribution of this paper in the development of a
novel algorithm, termed the `Mixed' pCN (MpCN). RWM and pCN are shown not to be
geometrically ergodic for an important class of matrix distributions with heavy
tails. In contrast, MpCN has very good empirical performance within this class.
Geometric ergodicity for MpCN is not fully proven in this work, as some
remaining drift conditions are quite challenging to obtain owing to the
complexity of the state space. We do, however, make a lot of progress towards a
proof, and show in detail the last steps left for future work. We illustrate
the computational performance of the various algorithms through simulation
studies, first for the trivial case of an Inverse-Wishart target, and then for
a challenging model arising in financial statistics.
| stat |
Estimation of Dirichlet distribution parameters with bias-reducing
adjusted score functions | The Dirichlet distribution, also known as multivariate beta, is the most used
to analyse frequencies or proportions data. Maximum likelihood is widespread
for estimation of Dirichlet's parameters. However, for small sample sizes, the
maximum likelihood estimator may shows a significant bias. In this paper,
Dirchlet's parameters estimation is obtained through modified score functions
aiming at mean and median bias reduction of the maximum likelihood estimator,
respectively. A simulation study and an application compare the adjusted score
approaches with maximum likelihood.
| stat |
Robust Deep Gaussian Processes | This report provides an in-depth overview over the implications and novelty
Generalized Variational Inference (GVI) (Knoblauch et al., 2019) brings to Deep
Gaussian Processes (DGPs) (Damianou & Lawrence, 2013). Specifically, robustness
to model misspecification as well as principled alternatives for uncertainty
quantification are motivated with an information-geometric view. These
modifications have clear interpretations and can be implemented in less than
100 lines of Python code. Most importantly, the corresponding empirical results
show that DGPs can greatly benefit from the presented enhancements.
| stat |
Efficient Bayesian estimation for GARCH-type models via Sequential Monte
Carlo | The advantages of sequential Monte Carlo (SMC) are exploited to develop
parameter estimation and model selection methods for GARCH (Generalized
AutoRegressive Conditional Heteroskedasticity) style models. It provides an
alternative method for quantifying estimation uncertainty relative to classical
inference. Even with long time series, it is demonstrated that the posterior
distribution of model parameters are non-normal, highlighting the need for a
Bayesian approach and an efficient posterior sampling method. Efficient
approaches for both constructing the sequence of distributions in SMC, and
leave-one-out cross-validation, for long time series data are also proposed.
Finally, an unbiased estimator of the likelihood is developed for the Bad
Environment-Good Environment model, a complex GARCH-type model, which permits
exact Bayesian inference not previously available in the literature.
| stat |
Sense and Sensitivity Analysis: Simple Post-Hoc Analysis of Bias Due to
Unobserved Confounding | It is a truth universally acknowledged that an observed association without
known mechanism must be in want of a causal estimate. However, causal
estimation from observational data often relies on the (untestable) assumption
of `no unobserved confounding'. Violations of this assumption can induce bias
in effect estimates. In principle, such bias could invalidate or reverse the
conclusions of a study. However, in some cases, we might hope that the
influence of unobserved confounders is weak relative to a `large' estimated
effect, so the qualitative conclusions are robust to bias from unobserved
confounding. The purpose of this paper is to develop \emph{Austen plots}, a
sensitivity analysis tool to aid such judgments by making it easier to reason
about potential bias induced by unobserved confounding. We formalize
confounding strength in terms of how strongly the confounder influences
treatment assignment and outcome. For a target level of bias, an Austen plot
shows the minimum values of treatment and outcome influence required to induce
that level of bias. Domain experts can then make subjective judgments about
whether such strong confounders are plausible. To aid this judgment, the Austen
plot additionally displays the estimated influence strength of (groups of) the
observed covariates. Austen plots generalize the classic sensitivity analysis
approach of Imbens [Imb03]. Critically, Austen plots allow any approach for
modeling the observed data and producing the initial estimate. We illustrate
the tool by assessing biases for several real causal inference problems, using
a variety of machine learning approaches for the initial data analysis. Code is
available at https://github.com/anishazaveri/austen_plots
| stat |
Slow and Stale Gradients Can Win the Race | Distributed Stochastic Gradient Descent (SGD) when run in a synchronous
manner, suffers from delays in runtime as it waits for the slowest workers
(stragglers). Asynchronous methods can alleviate stragglers, but cause gradient
staleness that can adversely affect the convergence error. In this work, we
present a novel theoretical characterization of the speedup offered by
asynchronous methods by analyzing the trade-off between the error in the
trained model and the actual training runtime(wallclock time). The main novelty
in our work is that our runtime analysis considers random straggling delays,
which helps us design and compare distributed SGD algorithms that strike a
balance between straggling and staleness. We also provide a new error
convergence analysis of asynchronous SGD variants without bounded or
exponential delay assumptions. Finally, based on our theoretical
characterization of the error-runtime trade-off, we propose a method of
gradually varying synchronicity in distributed SGD and demonstrate its
performance on CIFAR10 dataset.
| stat |
Non-Canonical Hamiltonian Monte Carlo | Hamiltonian Monte Carlo is typically based on the assumption of an underlying
canonical symplectic structure. Numerical integrators designed for the
canonical structure are incompatible with motion generated by non-canonical
dynamics. These non-canonical dynamics, motivated by examples in physics and
symplectic geometry, correspond to techniques such as preconditioning which are
routinely used to improve algorithmic performance. Indeed, recently, a special
case of non-canonical structure, magnetic Hamiltonian Monte Carlo, was
demonstrated to provide advantageous sampling properties. We present a
framework for Hamiltonian Monte Carlo using non-canonical symplectic
structures. Our experimental results demonstrate sampling advantages associated
to Hamiltonian Monte Carlo with non-canonical structure. To summarize our
contributions: (i) we develop non-canonical HMC from foundations in symplectic
geomtry; (ii) we construct an HMC procedure using implicit integration that
satisfies the detailed balance; (iii) we propose to accelerate the sampling
using an {\em approximate} explicit methodology; (iv) we study two novel,
randomly-generated non-canonical structures: magnetic momentum and the coupled
magnet structure, with implicit and explicit integration.
| stat |
Soft and subspace robust multivariate rank tests based on entropy
regularized optimal transport | In this paper, we extend the recently proposed multivariate rank energy
distance, based on the theory of optimal transport, for statistical testing of
distributional similarity, to soft rank energy distance. Being differentiable,
this in turn allows us to extend the rank energy to a subspace robust rank
energy distance, dubbed Projected soft-Rank Energy distance, which can be
computed via optimization over the Stiefel manifold. We show via experiments
that using projected soft rank energy one can trade-off the detection power vs
the false alarm via projections onto an appropriately selected low dimensional
subspace. We also show the utility of the proposed tests on unsupervised change
point detection in multivariate time series data. All codes are publicly
available at the link provided in the experiment section.
| stat |
Robust Experimental Designs for Model Calibration | A computer model can be used for predicting an output only after specifying
the values of some unknown physical constants known as calibration parameters.
The unknown calibration parameters can be estimated from real data by
conducting physical experiments. This paper presents an approach to optimally
design such a physical experiment. The problem of optimally designing physical
experiment, using a computer model, is similar to the problem of finding
optimal design for fitting nonlinear models. However, the problem is more
challenging than the existing work on nonlinear optimal design because of the
possibility of model discrepancy, that is, the computer model may not be an
accurate representation of the true underlying model. Therefore, we propose an
optimal design approach that is robust to potential model discrepancies. We
show that our designs are better than the commonly used physical experimental
designs that do not make use of the information contained in the computer model
and other nonlinear optimal designs that ignore potential model discrepancies.
We illustrate our approach using a toy example and a real example from
industry.
| stat |
A nonparametric framework for inferring orders of categorical data from
category-real ordered pairs | Given a dataset of careers and incomes, how large a difference of income
between any pair of careers would be? Given a dataset of travel time records,
how long do we need to spend more when choosing a public transportation mode
$A$ instead of $B$ to travel? In this paper, we propose a framework that is
able to infer orders of categories as well as magnitudes of difference of real
numbers between each pair of categories using Estimation statistics framework.
Not only reporting whether an order of categories exists, but our framework
also reports the magnitude of difference of each consecutive pairs of
categories in the order. In large dataset, our framework is scalable well
compared with the existing framework. The proposed framework has been applied
to two real-world case studies: 1) ordering careers by incomes based on
information of 350,000 households living in Khon Kaen province, Thailand, and
2) ordering sectors by closing prices based on 1060 companies' closing prices
of NASDAQ stock markets between years 2000 and 2016. The results of careers
ordering show income inequality among different careers. The stock market
results illustrate dynamics of sector domination that can change over time. Our
approach is able to be applied in any research area that has category-real
ordered pairs. Our proposed "Dominant-Distribution Network" provides a novel
approach to gain new insight of analyzing category orders. The software of this
framework is available for researchers or practitioners within R package:
EDOIF.
| stat |
Detecting the skewness of data from the sample size and the five-number
summary | For clinical studies with continuous outcomes, when the data are potentially
skewed, researchers may choose to report the whole or part of the five-number
summary (the sample median, the first and third quartiles, and the minimum and
maximum values), rather than the sample mean and standard deviation. For the
studies with skewed data, if we include them in the classical meta-analysis for
normal data, it may yield misleading or even wrong conclusions. In this paper,
we develop a flow chart and three new tests for detecting the skewness of data
from the sample size and the five-number summary. Simulation studies
demonstrate that our new tests are able to control the type I error rates, and
meanwhile provide good statistical power. A real data example is also analyzed
to demonstrate the usefulness of the skewness tests in meta-analysis and
evidence-based practice.
| stat |
Posterior computation with the Gibbs zig-zag sampler | Markov chain Monte Carlo (MCMC) sampling algorithms have dominated the
literature on posterior computation. However, MCMC faces substantial hurdles in
performing efficient posterior sampling for challenging Bayesian models,
particularly in high-dimensional and large data settings. Motivated in part by
such hurdles, an intriguing new class of piecewise deterministic Markov
processes (PDMPs) has recently been proposed as an alternative to MCMC. One of
the most popular types of PDMPs is known as the zig-zag (ZZ) sampler. Such
algorithms require a computational upper bound in a Poisson thinning step, with
performance improving for tighter bounds. In order to facilitate scaling to
larger classes of problems, we propose a general class of Gibbs zig-zag (GZZ)
samplers. GZZ allows parameters to be updated in blocks with ZZ applied to
certain parameters and traditional MCMC style updates to others. This provides
a flexible framework to combine PDMPs with the rich literature on MCMC
algorithms. We prove appealing theoretical properties of GZZ and demonstrate it
on posterior sampling for logistic models with shrinkage priors for
high-dimensional regression and random effects.
| stat |
Beyond A/B Testing: Sequential Randomization for Developing
Interventions in Scaled Digital Learning Environments | Randomized experiments ensure robust causal inference that are critical to
effective learning analytics research and practice. However, traditional
randomized experiments, like A/B tests, are limiting in large scale digital
learning environments. While traditional experiments can accurately compare two
treatment options, they are less able to inform how to adapt interventions to
continually meet learners' diverse needs. In this work, we introduce a trial
design for developing adaptive interventions in scaled digital learning
environments -- the sequential randomized trial (SRT). With the goal of
improving learner experience and developing interventions that benefit all
learners at all times, SRTs inform how to sequence, time, and personalize
interventions. In this paper, we provide an overview of SRTs, and we illustrate
the advantages they hold compared to traditional experiments. We describe a
novel SRT run in a large scale data science MOOC. The trial results
contextualize how learner engagement can be addressed through inclusive
culturally targeted reminder emails. We also provide practical advice for
researchers who aim to run their own SRTs to develop adaptive interventions in
scaled digital learning environments.
| stat |
Regression-based imputation of explanatory discrete missing data | Imputation of missing values is a strategy for handling non-responses in
surveys or data loss in measurement processes, which may be more effective than
ignoring them. When the variable represents a count, the literature dealing
with this issue is scarce. Likewise, if problems of over- or under-dispersion
are observed, generalisations of the Poisson distribution are recommended for
carrying out imputation. In order to assess the performance of various
regression models in the imputation of a discrete variable compared to
classical counting models, this work presents a comprehensive simulation study
considering a variety of scenarios and real data. To do so we compared the
results of estimations using only complete data, and using imputations based on
the Poisson, negative binomial, Hermite, and COMPoisson distributions, and the
ZIP and ZINB models for excesses of zeros. The results of this work reveal that
the COMPoisson distribution provides in general better results in any
dispersion scenario, especially when the amount of missing information is
large. When the variable presenting missing values is a count, the most widely
used method is to assume that a classical Poisson model is the best alternative
to impute the missing counts; however, in real-life research this assumption is
not always correct, and it is common to find count variables exhibiting
overdispersion or underdispersion, for which the Poisson model is no longer the
best to use in imputation. In several of the scenarios considered the
performance of the methods analysed differs, something which indicates that it
is important to analyse dispersion and the possible presence of excess zeros
before deciding on the imputation method to use. The COMPoisson model performs
well as it is flexible regarding the handling of counts with characteristics of
over- and under-dispersion, as well as with equidispersion.
| stat |
A Bayesian Model of Dose-Response for Cancer Drug Studies | Exploratory cancer drug studies test multiple tumor cell lines against
multiple candidate drugs. The goal in each paired (cell line, drug) experiment
is to map out the dose-response curve of the cell line as the dose level of the
drug increases. We propose Bayesian Tensor Filtering (BTF), a hierarchical
Bayesian model for dose-response modeling in multi-sample, multi-treatment
cancer drug studies. BTF uses low-dimensional embeddings to share statistical
strength between similar drugs and similar cell lines. Structured shrinkage
priors in BTF encourage smoothness in the dose-response curves while remaining
adaptive to sharp jumps when the data call for it. We focus on a pair of cancer
drug studies exhibiting a particular pathology in their experimental design,
leading us to a non-conjugate monotone mixture-of-Gammas likelihood. To perform
posterior inference, we develop a variant of the elliptical slice sampling
algorithm for sampling from linearly-constrained multivariate normal priors
with non-conjugate likelihoods. In benchmarks, BTF outperforms state-of-the-art
methods for covariance regression and dynamic Poisson matrix factorization. On
the two cancer drug studies, BTF outperforms the current standard approach in
biology and reveals potential new biomarkers of drug sensitivity in cancer.
Code is available at https://github.com/tansey/functionalmf.
| stat |
Epileptic seizure prediction using Pearson's product-moment correlation
coefficient of a linear classifier from generalized Gaussian modeling | To predict an epileptic event means the ability to determine in advance the
time of the seizure with the highest possible accuracy. A correct prediction
benchmark for epilepsy events in clinical applications is a typical problem in
biomedical signal processing that helps to an appropriate diagnosis and
treatment of this disease. In this work, we use Pearson's product-moment
correlation coefficient from generalized Gaussian distribution parameters
coupled with a linear-based classifier to predict between seizure and
non-seizure events in epileptic EEG signals. The performance in 36 epileptic
events from 9 patients showing good performance with 100% of effectiveness for
sensitivity and specificity greater than 83% for seizures events in all brain
rhythms. Pearson's test suggests that all brain rhythms are highly correlated
in non-seizure events but no during the seizure events. This suggests that our
model can be scaled with the Pearson's product-moment correlation coefficient
for the detection of epileptic seizures.
| stat |
A Unified Framework for Inference in Network Models with Degree
Heterogeneity and Homophily | The degree heterogeneity and homophily are two typical features in network
data. In this paper, we formulate a general model for undirected networks with
these two features and present the moment estimation for inferring the degree
and homophily parameters. The binary or nonbinary network edges are
simultaneously considered. We establish a unified theoretical framework under
which the consistency of the moment estimator holds as the size of networks
goes to infinity. We also derive the asymptotic representation of the moment
estimator that can be used to characterize its limiting distribution. The
asymptotic representation of the moment estimator of the homophily parameter
contains a bias term. Two applications are provided to illustrate the
theoretical result. Numerical studies and a real data analysis demonstrate our
theoretical findings.
| stat |
On the Robustness to Misspecification of $\alpha$-Posteriors and Their
Variational Approximations | $\alpha$-posteriors and their variational approximations distort standard
posterior inference by downweighting the likelihood and introducing variational
approximation errors. We show that such distortions, if tuned appropriately,
reduce the Kullback-Leibler (KL) divergence from the true, but perhaps
infeasible, posterior distribution when there is potential parametric model
misspecification. To make this point, we derive a Bernstein-von Mises theorem
showing convergence in total variation distance of $\alpha$-posteriors and
their variational approximations to limiting Gaussian distributions. We use
these distributions to evaluate the KL divergence between true and reported
posteriors. We show this divergence is minimized by choosing $\alpha$ strictly
smaller than one, assuming there is a vanishingly small probability of model
misspecification. The optimized value becomes smaller as the the
misspecification becomes more severe. The optimized KL divergence increases
logarithmically in the degree of misspecification and not linearly as with the
usual posterior.
| stat |
Diagnosis Prevalence vs. Efficacy in Machine-learning Based Diagnostic
Decision Support | Many recent studies use machine learning to predict a small number of
ICD-9-CM codes. In practice, on the other hand, physicians have to consider a
broader range of diagnoses. This study aims to put these previously incongruent
evaluation settings on a more equal footing by predicting ICD-9-CM codes based
on electronic health record properties and demonstrating the relationship
between diagnosis prevalence and system performance. We extracted patient
features from the MIMIC-III dataset for each admission. We trained and
evaluated 43 different machine learning classifiers. Among this pool, the most
successful classifier was a Multi-Layer Perceptron. In accordance with general
machine learning expectation, we observed all classifiers' F1 scores to drop as
disease prevalence decreased. Scores fell from 0.28 for the 50 most prevalent
ICD-9-CM codes to 0.03 for the 1000 most prevalent ICD-9-CM codes. Statistical
analyses showed a moderate positive correlation between disease prevalence and
efficacy (0.5866).
| stat |
The fraud loss for selecting the model complexity in fraud detection | In fraud detection applications, the investigator is typically limited to
controlling a restricted number k of cases. The most efficient manner of
allocating the resources is then to try selecting the k cases with the highest
probability of being fraudulent. The prediction model used for this purpose
must normally be regularized to avoid overfitting and consequently bad
prediction performance. A new loss function, denoted the fraud loss, is
proposed for selecting the model complexity via a tuning parameter. A
simulation study is performed to find the optimal settings for validation.
Further, the performance of the proposed procedure is compared to the most
relevant competing procedure, based on the area under the receiver operating
characteristic curve (AUC), in a set of simulations, as well as on a VAT fraud
dataset. In most cases, choosing the complexity of the model according to the
fraud loss, gave a better than, or comparable performance to the AUC in terms
of the fraud loss.
| stat |
A Fast Particle-Based Approach for Calibrating a 3-D Model of the
Antarctic Ice Sheet | We consider the scientifically challenging and policy-relevant task of
understanding the past and projecting the future dynamics of the Antarctic ice
sheet. The Antarctic ice sheet has shown a highly nonlinear threshold response
to past climate forcings. Triggering such a threshold response through
anthropogenic greenhouse gas emissions would drive drastic and potentially fast
sea level rise with important implications for coastal flood risks. Previous
studies have combined information from ice sheet models and observations to
calibrate model parameters. These studies have broken important new ground but
have either adopted simple ice sheet models or have limited the number of
parameters to allow for the use of more complex models. These limitations are
largely due to the computational challenges posed by calibration as models
become more computationally intensive or when the number of parameters
increases. Here we propose a method to alleviate this problem: a fast
sequential Monte Carlo method that takes advantage of the massive
parallelization afforded by modern high performance computing systems. We use
simulated examples to demonstrate how our sample-based approach provides
accurate approximations to the posterior distributions of the calibrated
parameters. The drastic reduction in computational times enables us to provide
new insights into important scientific questions, for example, the impact of
Pliocene era data and prior parameter information on sea level projections.
These studies would be computationally prohibitive with other computational
approaches for calibration such as Markov chain Monte Carlo or emulation-based
methods. We also find considerable differences in the distributions of sea
level projections when we account for a larger number of uncertain parameters.
| stat |
Localizing Changes in High-Dimensional Regression Models | This paper addresses the problem of localizing change points in
high-dimensional linear regression models with piecewise constant regression
coefficients. We develop a dynamic programming approach to estimate the
locations of the change points whose performance improves upon the current
state-of-the-art, even as the dimensionality, the sparsity of the regression
coefficients, the temporal spacing between two consecutive change points, and
the magnitude of the difference of two consecutive regression coefficient
vectors are allowed to vary with the sample size. Furthermore, we devise a
computationally-efficient refinement procedure that provably reduces the
localization error of preliminary estimates of the change points. We
demonstrate minimax lower bounds on the localization error that nearly match
the upper bound on the localization error of our methodology and show that the
signal-to-noise condition we impose is essentially the weakest possible based
on information-theoretic arguments. Extensive numerical results support our
theoretical findings, and experiments on real air quality data reveal change
points supported by historical information not used by the algorithm.
| stat |
Real Time Anomaly Detection And Categorisation | The ability to quickly and accurately detect anomalous structure within data
sequences is an inference challenge of growing importance. This work extends
recently proposed post-hoc (offline) anomaly detection methodology to the
sequential setting. The resultant procedure is capable of real-time analysis
and categorisation between baseline and two forms of anomalous structure: point
and collective anomalies. Various theoretical properties of the procedure are
derived. These, together with an extensive simulation study, highlight that the
average run length to false alarm and the average detection delay of the
proposed online algorithm are very close to that of the offline version.
Experiments on simulated and real data are provided to demonstrate the benefits
of the proposed method.
| stat |
Random autoregressive models: A structured overview | Models characterized by autoregressive structure and random coefficients are
powerful tools for the analysis of high-frequency, high-dimensional and
volatile time series. The available literature on such models is broad, but
also sectorial, overlapping, and confusing. Most models focus on one property
of the data, while much can be gained by combining the strength of various
models and their sources of heterogeneity. We present a structured overview of
the literature on autoregressive models with random coefficients. We describe
hierarchy and analogies among models, and for each we systematically list
properties, estimation methods, tests, software packages and typical
applications.
| stat |
Using machine learning to identify nontraditional spatial dependence in
occupancy data | Spatial models for occupancy data are used to estimate and map the true
presence of a species, which may depend on biotic and abiotic factors as well
as spatial autocorrelation. Traditionally researchers have accounted for
spatial autocorrelation in occupancy data by using a correlated normally
distributed site-level random effect, which might be incapable of identifying
nontraditional spatial dependence such as discontinuities and abrupt
transitions. Machine learning approaches have the potential to identify and
model nontraditional spatial dependence, but these approaches do not account
for observer errors such as false absences. By combining the flexibility of
Bayesian hierarchal modeling and machine learning approaches, we present a
general framework to model occupancy data that accounts for both traditional
and nontraditional spatial dependence as well as false absences. We demonstrate
our framework using six synthetic occupancy data sets and two real data sets.
Our results demonstrate how to identify and model both traditional and
nontraditional spatial dependence in occupancy data which enables a broader
class of spatial occupancy models that can be used to improve predictive
accuracy and model adequacy.
| stat |
Adaptively stacking ensembles for influenza forecasting with incomplete
data | Seasonal influenza infects between 10 and 50 million people in the United
States every year, overburdening hospitals during weeks of peak incidence.
Named by the CDC as an important tool to fight the damaging effects of these
epidemics, accurate forecasts of influenza and influenza-like illness (ILI)
forewarn public health officials about when, and where, seasonal influenza
outbreaks will hit hardest. Multi-model ensemble forecasts---weighted
combinations of component models---have shown positive results in forecasting.
Ensemble forecasts of influenza outbreaks have been static, training on all
past ILI data at the beginning of a season, generating a set of optimal weights
for each model in the ensemble, and keeping the weights constant. We propose an
adaptive ensemble forecast that (i) changes model weights week-by-week
throughout the influenza season, (ii) only needs the current influenza season's
data to make predictions, and (iii) by introducing a prior distribution,
shrinks weights toward the reference equal weighting approach and adjusts for
observed ILI percentages that are subject to future revisions. We investigate
the prior's ability to impact adaptive ensemble performance and, after finding
an optimal prior via a cross-validation approach, compare our adaptive
ensemble's performance to equal-weighted and static ensembles. Applied to
forecasts of short-term ILI incidence at the regional and national level in the
US, our adaptive model outperforms a naive equal-weighted ensemble, and has
similar or better performance to the static ensemble, which requires multiple
years of training data. Adaptive ensembles are able to quickly train and
forecast during epidemics, and provide a practical tool to public health
officials looking for forecasts that can conform to unique features of a
specific season.
| stat |
Finite-sample Efficient Conformal Prediction | Conformal prediction is a generic methodology for finite-sample valid
distribution-free prediction. This technique has garnered a lot of attention in
the literature partly because it can be applied with any machine learning
algorithm that provides point predictions to yield valid prediction regions. Of
course, the efficiency (width/volume) of the resulting prediction region
depends on the performance of the machine learning algorithm. In this paper, we
consider the problem of obtaining the smallest conformal prediction region
given a family of machine learning algorithms. We provide two general-purpose
selection algorithms and consider coverage as well as width properties of the
final prediction region. The first selection method yields the smallest width
prediction region among the family of conformal prediction regions for all
sample sizes, but only has an approximate coverage guarantee. The second
selection method has a finite sample coverage guarantee but only attains close
to the smallest width. The approximate optimal width property of the second
method is quantified via an oracle inequality. Asymptotic oracle inequalities
are also considered when the family of algorithms is given by ridge regression
with different penalty parameters.
| stat |
Likelihood Assignment for Out-of-Distribution Inputs in Deep Generative
Models is Sensitive to Prior Distribution Choice | Recent work has shown that deep generative models assign higher likelihood to
out-of-distribution inputs than to training data. We show that a factor
underlying this phenomenon is a mismatch between the nature of the prior
distribution and that of the data distribution, a problem found in widely used
deep generative models such as VAEs and Glow. While a typical choice for a
prior distribution is a standard Gaussian distribution, properties of
distributions of real data sets may not be consistent with a unimodal prior
distribution. This paper focuses on the relationship between the choice of a
prior distribution and the likelihoods assigned to out-of-distribution inputs.
We propose the use of a mixture distribution as a prior to make likelihoods
assigned by deep generative models sensitive to out-of-distribution inputs.
Furthermore, we explain the theoretical advantages of adopting a mixture
distribution as the prior, and we present experimental results to support our
claims. Finally, we demonstrate that a mixture prior lowers the
out-of-distribution likelihood with respect to two pairs of real image data
sets: Fashion-MNIST vs. MNIST and CIFAR10 vs. SVHN.
| stat |
Community detection in sparse latent space models | We show that a simple community detection algorithm originated from
stochastic blockmodel literature achieves consistency, and even optimality, for
a broad and flexible class of sparse latent space models. The class of models
includes latent eigenmodels (arXiv:0711.1146). The community detection
algorithm is based on spectral clustering followed by local refinement via
normalized edge counting.
| stat |
Computationally Efficient Deep Bayesian Unit-Level Modeling of Survey
Data under Informative Sampling for Small Area Estimation | The topic of deep learning has seen a surge of interest in recent years both
within and outside of the field of Statistics. Deep models leverage both
nonlinearity and interaction effects to provide superior predictions in many
cases when compared to linear or generalized linear models. However, one of the
main challenges with deep modeling approaches is quantification of uncertainty.
The use of random weight models, such as the popularized "Extreme Learning
Machine," offer a potential solution in this regard. In addition to uncertainty
quantification, these models are extremely computationally efficient as they do
not require optimization through stochastic gradient descent, which is what is
typically done for deep learning. We show how the use of random weights in a
deep model can fit into a likelihood based framework to allow for uncertainty
quantification of the model parameters and any desired estimates. Furthermore,
we show how this approach can be used to account for informative sampling of
survey data through the use of a pseudo-likelihood. We illustrate the
effectiveness of this methodology through simulation and with a real survey
data application involving American National Election Studies data.
| stat |
Total Stability of SVMs and Localized SVMs | Regularized kernel-based methods such as support vector machines (SVMs)
typically depend on the underlying probability measure $\mathrm{P}$
(respectively an empirical measure $\mathrm{D}_n$ in applications) as well as
on the regularization parameter $\lambda$ and the kernel $k$. Whereas classical
statistical robustness only considers the effect of small perturbations in
$\mathrm{P}$, the present paper investigates the influence of simultaneous
slight variations in the whole triple $(\mathrm{P},\lambda,k)$, respectively
$(\mathrm{D}_n,\lambda_n,k)$, on the resulting predictor. Existing results from
the literature are considerably generalized and improved. In order to also make
them applicable to big data, where regular SVMs suffer from their super-linear
computational requirements, we show how our results can be transferred to the
context of localized learning. Here, the effect of slight variations in the
applied regionalization, which might for example stem from changes in
$\mathrm{P}$ respectively $\mathrm{D}_n$, is considered as well.
| stat |
Optimal Bias Correction of the Log-periodogram Estimator of the
Fractional Parameter: A Jackknife Approach | We use the jackknife to bias correct the log-periodogram regression(LPR)
estimator of the fractional parameter in a stationary fractionally integrated
model. The weights for the jackknife estimator are chosen in such a way that
bias reduction is achieved without the usual increase in asymptotic variance,
with the estimator viewed as `optimal' in this sense. The theoretical results
are valid under both the non-overlapping and moving-block sub-sampling schemes
that can be used in the jackknife technique, and do not require the assumption
of Gaussianity for the data generating process. A Monte Carlo study explores
the finite sample performance of different versions of the jackknife estimator,
under a variety of scenarios. The simulation experiments reveal that when the
weights are constructed using the parameter values of the true data generating
process, a version of the optimal jackknife estimator almost always
out-performs alternative semi-parametric bias-corrected estimators. A feasible
version of the jackknife estimator, in which the weights are constructed using
estimates of the unknown parameters, whilst not dominant overall, is still the
least biased estimator in some cases. Even when misspecified short run dynamics
are assumed in the construction of the weights, the feasible jackknife still
shows significant reduction in bias under certain designs. As is not
surprising, parametric maximum likelihood estimation out-performs all
semi-parametric methods when the true values of the short memory parameters are
known, but is dominated by the semi-parametric methods (in terms of bias) when
the short memory parameters need to be estimated, and in particular when the
model is misspecified.
| stat |
Currents and K-functions for Fiber Point Processes | Analysis of images of sets of fibers such as myelin sheaths or skeletal
muscles must account for both the spatial distribution of fibers and
differences in fiber shape. This necessitates a combination of point process
and shape analysis methodology. In this paper, we develop a K-function for
shape-valued point processes by embedding shapes as currents, thus equipping
the point process domain with metric structure inherited from a reproducing
kernel Hilbert space. We extend Ripley's K-function which measures deviations
from spatial homogeneity of point processes to fiber data. The paper provides a
theoretical account of the statistical foundation of the K-function and its
extension to fiber data, and we test the developed K-function on simulated as
well as real data sets. This includes a fiber data set consisting of myelin
sheaths, visualizing the spatial and fiber shape behavior of myelin
configurations at different debts.
| stat |
Unconstrained representation of orthogonal matrices with application to
common principle components | Many statistical problems involve the estimation of a $\left(d\times
d\right)$ orthogonal matrix $\textbf{Q}$. Such an estimation is often
challenging due to the orthonormality constraints on $\textbf{Q}$. To cope with
this problem, we propose a very simple decomposition for orthogonal matrices
which we abbreviate as PLR decomposition. It produces a one-to-one
correspondence between $\textbf{Q}$ and a $\left(d\times d\right)$ unit lower
triangular matrix $\textbf{L}$ whose $d\left(d-1\right)/2$ entries below the
diagonal are unconstrained real values. Once the decomposition is applied,
regardless of the objective function under consideration, we can use any
classical unconstrained optimization method to find the minimum (or maximum) of
the objective function with respect to $\textbf{L}$. For illustrative purposes,
we apply the PLR decomposition in common principle components analysis (CPCA)
for the maximum likelihood estimation of the common orthogonal matrix when a
multivariate leptokurtic-normal distribution is assumed in each group. Compared
to the commonly used normal distribution, the leptokurtic-normal has an
additional parameter governing the excess kurtosis; this makes the estimation
of $\textbf{Q}$ in CPCA more robust against mild outliers. The usefulness of
the PLR decomposition in leptokurtic-normal CPCA is illustrated by two
biometric data analyses.
| stat |
Parallel tempering as a mechanism for facilitating inference in
hierarchical hidden Markov models | The study of animal behavioural states inferred through hidden Markov models
and similar state switching models has seen a significant increase in
popularity in recent years. The ability to account for varying levels of
behavioural scale has become possible through hierarchical hidden Markov
models, but additional levels lead to higher complexity and increased
correlation between model components. Maximum likelihood approaches to
inference using the EM algorithm and direct optimisation of likelihoods are
more frequently used, with Bayesian approaches being less favoured due to
computational demands. Given these demands, it is vital that efficient
estimation algorithms are developed when Bayesian methods are preferred. We
study the use of various approaches to improve convergence times and mixing in
Markov chain Monte Carlo methods applied to hierarchical hidden Markov models,
including parallel tempering as an inference facilitation mechanism. The method
shows promise for analysing complex stochastic models with high levels of
correlation between components, but our results show that it requires careful
tuning in order to maximise that potential.
| stat |
Bayesian exponential random graph models for populations of networks | The collection of data on populations of networks is becoming increasingly
common, where each data point can be seen as a realisation of a network-valued
random variable. A canonical example is that of brain networks: a typical
neuroimaging study collects one or more brain scans across multiple
individuals, each of which can be modelled as a network with nodes
corresponding to distinct brain regions and edges corresponding to structural
or functional connections between these regions. Most statistical network
models, however, were originally proposed to describe a single underlying
relational structure, although recent years have seen a drive to extend these
models to populations of networks. Here, we propose one such extension: a
multilevel framework for populations of networks based on exponential random
graph models. By pooling information across the individual networks, this
framework provides a principled approach to characterise the relational
structure for an entire population. To perform inference, we devise a novel
exchange-within-Gibbs MCMC algorithm that generates samples from the
doubly-intractable posterior. To illustrate our framework, we use it to assess
group-level variations in networks derived from fMRI scans, enabling the
inference of age-related differences in the topological structure of the
brain's functional connectivity.
| stat |
Causal Regularization | I argue that regularizing terms in standard regression methods not only help
against overfitting finite data, but sometimes also yield better causal models
in the infinite sample regime. I first consider a multi-dimensional variable
linearly influencing a target variable with some multi-dimensional unobserved
common cause, where the confounding effect can be decreased by keeping the
penalizing term in Ridge and Lasso regression even in the population limit.
Choosing the size of the penalizing term, is however challenging, because cross
validation is pointless. Here it is done by first estimating the strength of
confounding via a method proposed earlier, which yielded some reasonable
results for simulated and real data.
Further, I prove a `causal generalization bound' which states (subject to a
particular model of confounding) that the error made by interpreting any
non-linear regression as causal model can be bounded from above whenever
functions are taken from a not too rich class. In other words, the bound
guarantees "generalization" from observational to interventional distributions,
which is usually not subject of statistical learning theory (and is only
possible due to the underlying symmetries of the confounder model).
| stat |
Partial Correlations in Compositional Data Analysis | Partial correlations quantify linear association between two variables
adjusting for the influence of the remaining variables. They form the backbone
for graphical models and are readily obtained from the inverse of the
covariance matrix. For compositional data, the covariance structure is
specified from log ratios of variables, so unless we try to "open" the data via
a normalization, this implies changes in the definition and interpretation of
partial correlations. In the present work, we elucidate how results derived by
Aitchison (1986) lead to a natural definition of partial correlation that has a
number of advantages over current measures of association. For this, we show
that the residuals of log-ratios between a variable with a reference, when
adjusting for all remaining variables including the reference, are
reference-independent. Since the reference itself can be controlled for,
correlations between residuals are defined for the variables directly without
the necessity to recur to ratios except when specifying which variables are
partialled out. Thus, perhaps surprisingly, partial correlations do not have
the problems commonly found with measures of pairwise association on
compositional data. They are well-defined between two variables, are properly
scaled, and allow for negative association. By design, they are
subcompositionally incoherent, but they share this property with conventional
partial correlations (where results change when adjusting for the influence of
fewer variables). We discuss the equivalence with normalization-based
approaches whenever the normalizing variables are controlled for. We also
discuss the partial variances and correlations we obtain from a previously
studied data set of Roman glass cups.
| stat |
Non Proportional Odds Models are Widely Dispensable -- Sparser Modeling
based on Parametric and Additive Location-Shift Approaches | The potential of location-shift models to find adequate models between the
proportional odds model and the non-proportional odds model is investigated. It
is demonstrated that these models are very useful in ordinal modeling. While
proportional odds models are often too simple, non proportional odds models are
typically unnecessary complicated and seem widely dispensable. The class of
location-shift models is also extended to allow for smooth effects. The
additive location-shift model contains two functions for each explanatory
variable, one for the location and one for dispersion. It is much sparser than
hard-to-handle additive models with category-specific covariate functions but
more flexible than common vector generalized additive models.
| stat |
Predicting the Long-Term Outcomes of Biologics in Psoriasis Patients
Using Machine Learning | Background. Real-world data show that approximately 50% of psoriasis patients
treated with a biologic agent will discontinue the drug because of loss of
efficacy. History of previous therapy with another biologic, female sex and
obesity were identified as predictors of drug discontinuations, but their
individual predictive value is low. Objectives. To determine whether machine
learning algorithms can produce models that can accurately predict outcomes of
biologic therapy in psoriasis on individual patient level. Results. All tested
machine learning algorithms could accurately predict the risk of drug
discontinuation and its cause (e.g. lack of efficacy vs adverse event). The
learned generalized linear model achieved diagnostic accuracy of 82%, requiring
under 2 seconds per patient using the psoriasis patients dataset. Input
optimization analysis established a profile of a patient who has best chances
of long-term treatment success: biologic-naive patient under 49 years,
early-onset plaque psoriasis without psoriatic arthritis, weight < 100 kg, and
moderate-to-severe psoriasis activity (DLQI $\geq$ 16; PASI $\geq$ 10).
Moreover, a different generalized linear model is used to predict the length of
treatment for each patient with mean absolute error (MAE) of 4.5 months.
However Pearson Correlation Coefficient indicates 0.935 linear dependencies
between the actual treatment lengths and predicted ones. Conclusions. Machine
learning algorithms predict the risk of drug discontinuation and treatment
duration with accuracy exceeding 80%, based on a small set of predictive
variables. This approach can be used as a decision-making tool, communicating
expected outcomes to the patient, and development of evidence-based guidelines.
| stat |
Optimal Sampling for Generalized Linear Models under Measurement
Constraints | Under "measurement constraints," responses are expensive to measure and
initially unavailable on most of records in the dataset, but the covariates are
available for the entire dataset. Our goal is to sample a relatively small
portion of the dataset where the expensive responses will be measured and the
resultant sampling estimator is statistically efficient. Measurement
constraints require the sampling probabilities can only depend on a very small
set of the responses. A sampling procedure that uses responses at most only on
a small pilot sample will be called "response-free." We propose a response-free
sampling procedure \mbox{(OSUMC)} for generalized linear models (GLMs). Using
the A-optimality criterion, i.e., the trace of the asymptotic variance, the
resultant estimator is statistically efficient within a class of sampling
estimators. We establish the unconditional asymptotic distribution of a general
class of response-free sampling estimators. This result is novel compared with
the existing conditional results obtained by conditioning on both covariates
and responses. Under our unconditional framework, the subsamples are no longer
independent and new martingale techniques are developed for our asymptotic
theory. We further derive the A-optimal response-free sampling distribution.
Since this distribution depends on population level quantities, we propose the
Optimal Sampling Under Measurement Constraints (OSUMC) algorithm to approximate
the theoretical optimal sampling. Finally, we conduct an intensive empirical
study to demonstrate the advantages of OSUMC algorithm over existing methods in
both statistical and computational perspectives.
| stat |
Optimal False Discovery Rate Control for Large Scale Multiple Testing
with Auxiliary Information | Large-scale multiple testing is a fundamental problem in high dimensional
statistical inference. It is increasingly common that various types of
auxiliary information, reflecting the structural relationship among the
hypotheses, are available. Exploiting such auxiliary information can boost
statistical power. To this end, we propose a framework based on a two-group
mixture model with varying probabilities of being null for different hypotheses
a priori, where a shape-constrained relationship is imposed between the
auxiliary information and the prior probabilities of being null. An optimal
rejection rule is designed to maximize the expected number of true positives
when average false discovery rate is controlled. Focusing on the ordered
structure, we develop a robust EM algorithm to estimate the prior probabilities
of being null and the distribution of $p$-values under the alternative
hypothesis simultaneously. We show that the proposed method has better power
than state-of-the-art competitors while controlling the false discovery rate,
both empirically and theoretically. Extensive simulations demonstrate the
advantage of the proposed method. Datasets from genome-wide association studies
are used to illustrate the new methodology.
| stat |
Prediction with Spatio-temporal Point Processes with Self Organizing
Decision Trees | We study the spatio-temporal prediction problem, which has attracted the
attention of many researchers due to its critical real-life applications. In
particular, we introduce a novel approach to this problem. Our approach is
based on the Hawkes process, which is a non-stationary and self-exciting point
process. We extend the formulations of a standard point process model that can
represent time-series data to represent a spatio-temporal data. We model the
data as nonstationary in time and space. Furthermore, we partition the spatial
region we are working on into subregions via an adaptive decision tree and
model the source statistics in each subregion with individual but mutually
interacting point processes. We also provide a gradient based joint
optimization algorithm for the point process and decision tree parameters.
Thus, we introduce a model that can jointly infer the source statistics and an
adaptive partitioning of the spatial region. Finally, we provide experimental
results on real-life data, which provides significant improvement due to space
adaptation and joint optimization compared to standard well-known methods in
the literature.
| stat |
Active Bayesian Assessment for Black-Box Classifiers | Recent advances in machine learning have led to increased deployment of
black-box classifiers across a wide variety of applications. In many such
situations there is a critical need to both reliably assess the performance of
these pre-trained models and to perform this assessment in a label-efficient
manner (given that labels may be scarce and costly to collect). In this paper,
we introduce an active Bayesian approach for assessment of classifier
performance to satisfy the desiderata of both reliability and label-efficiency.
We begin by developing inference strategies to quantify uncertainty for common
assessment metrics such as accuracy, misclassification cost, and calibration
error. We then propose a general framework for active Bayesian assessment using
inferred uncertainty to guide efficient selection of instances for labeling,
enabling better performance assessment with fewer labels. We demonstrate
significant gains from our proposed active Bayesian approach via a series of
systematic empirical experiments assessing the performance of modern neural
classifiers (e.g., ResNet and BERT) on several standard image and text
classification datasets.
| stat |
Statistical Efficiency of Thompson Sampling for Combinatorial
Semi-Bandits | We investigate stochastic combinatorial multi-armed bandit with semi-bandit
feedback (CMAB). In CMAB, the question of the existence of an efficient policy
with an optimal asymptotic regret (up to a factor poly-logarithmic with the
action size) is still open for many families of distributions, including
mutually independent outcomes, and more generally the multivariate sub-Gaussian
family. We propose to answer the above question for these two families by
analyzing variants of the Combinatorial Thompson Sampling policy (CTS). For
mutually independent outcomes in $[0,1]$, we propose a tight analysis of CTS
using Beta priors. We then look at the more general setting of multivariate
sub-Gaussian outcomes and propose a tight analysis of CTS using Gaussian
priors. This last result gives us an alternative to the Efficient Sampling for
Combinatorial Bandit policy (ESCB), which, although optimal, is not
computationally efficient.
| stat |
An Inverse-free Truncated Rayleigh-Ritz Method for Sparse Generalized
Eigenvalue Problem | This paper considers the sparse generalized eigenvalue problem (SGEP), which
aims to find the leading eigenvector with at most $k$ nonzero entries. SGEP
naturally arises in many applications in machine learning, statistics, and
scientific computing, for example, the sparse principal component analysis
(SPCA), the sparse discriminant analysis (SDA), and the sparse canonical
correlation analysis (SCCA). In this paper, we focus on the development of a
three-stage algorithm named {\em inverse-free truncated Rayleigh-Ritz method}
({\em IFTRR}) to efficiently solve SGEP. In each iteration of IFTRR, only a
small number of matrix-vector products is required. This makes IFTRR
well-suited for large scale problems. Particularly, a new truncation strategy
is proposed, which is able to find the support set of the leading eigenvector
effectively. Theoretical results are developed to explain why IFTRR works well.
Numerical simulations demonstrate the merits of IFTRR.
| stat |
Improving Maximum Likelihood Training for Text Generation with Density
Ratio Estimation | Auto-regressive sequence generative models trained by Maximum Likelihood
Estimation suffer the exposure bias problem in practical finite sample
scenarios. The crux is that the number of training samples for Maximum
Likelihood Estimation is usually limited and the input data distributions are
different at training and inference stages. Many method shave been proposed to
solve the above problem (Yu et al., 2017; Lu et al., 2018), which relies on
sampling from the non-stationary model distribution and suffers from high
variance or biased estimations. In this paper, we propose{\psi}-MLE, a new
training scheme for auto-regressive sequence generative models, which is
effective and stable when operating at large sample space encountered in text
generation. We derive our algorithm from a new perspective of self-augmentation
and introduce bias correction with density ratio estimation. Extensive
experimental results on synthetic data and real-world text generation tasks
demonstrate that our method stably outperforms Maximum Likelihood Estimation
and other state-of-the-art sequence generative models in terms of both quality
and diversity.
| stat |
Influence of single observations on the choice of the penalty parameter
in ridge regression | Penalized regression methods, such as ridge regression, heavily rely on the
choice of a tuning, or penalty, parameter, which is often computed via
cross-validation. Discrepancies in the value of the penalty parameter may lead
to substantial differences in regression coefficient estimates and predictions.
In this paper, we investigate the effect of single observations on the optimal
choice of the tuning parameter, showing how the presence of influential points
can dramatically change it. We distinguish between points as "expanders" and
"shrinkers", based on their effect on the model complexity. Our approach
supplies a visual exploratory tool to identify influential points, naturally
implementable for high-dimensional data where traditional approaches usually
fail. Applications to real data examples, both low- and high-dimensional, and a
simulation study are presented.
| stat |
Conditionally structured variational Gaussian approximation with
importance weights | We develop flexible methods of deriving variational inference for models with
complex latent variable structure. By splitting the variables in these models
into "global" parameters and "local" latent variables, we define a class of
variational approximations that exploit this partitioning and go beyond
Gaussian variational approximation. This approximation is motivated by the fact
that in many hierarchical models, there are global variance parameters which
determine the scale of local latent variables in their posterior conditional on
the global parameters. We also consider parsimonious parametrizations by using
conditional independence structure, and improved estimation of the log marginal
likelihood and variational density using importance weights. These methods are
shown to improve significantly on Gaussian variational approximation methods
for a similar computational cost. Application of the methodology is illustrated
using generalized linear mixed models and state space models.
| stat |
Clustering piecewise stationary processes | The problem of time-series clustering is considered in the case where each
data-point is a sample generated by a piecewise stationary ergodic process.
Stationary processes are perhaps the most general class of processes considered
in non-parametric statistics and allow for arbitrary long-range dependence
between variables. Piecewise stationary processes studied here for the first
time in the context of clustering, relax the last remaining assumption in this
model: stationarity. A natural formulation is proposed for this problem and a
notion of consistency is introduced which requires the samples to be placed in
the same cluster if and only if the piecewise stationary distributions that
generate them have the same set of stationary distributions. Simple,
computationally efficient algorithms are proposed and are shown to be
consistent without any additional assumptions beyond piecewise stationarity.
| stat |
A machine learning methodology for real-time forecasting of the
2019-2020 COVID-19 outbreak using Internet searches, news alerts, and
estimates from mechanistic models | We present a timely and novel methodology that combines disease estimates
from mechanistic models with digital traces, via interpretable machine-learning
methodologies, to reliably forecast COVID-19 activity in Chinese provinces in
real-time. Specifically, our method is able to produce stable and accurate
forecasts 2 days ahead of current time, and uses as inputs (a) official health
reports from Chinese Center Disease for Control and Prevention (China CDC), (b)
COVID-19-related internet search activity from Baidu, (c) news media activity
reported by Media Cloud, and (d) daily forecasts of COVID-19 activity from
GLEAM, an agent-based mechanistic model. Our machine-learning methodology uses
a clustering technique that enables the exploitation of geo-spatial
synchronicities of COVID-19 activity across Chinese provinces, and a data
augmentation technique to deal with the small number of historical disease
activity observations, characteristic of emerging outbreaks. Our model's
predictive power outperforms a collection of baseline models in 27 out of the
32 Chinese provinces, and could be easily extended to other geographies
currently affected by the COVID-19 outbreak to help decision makers.
| stat |
Bayesian Nonparametric Cost-Effectiveness Analyses: Causal Estimation
and Adaptive Subgroup Discovery | Cost-effectiveness analyses (CEAs) are at the center of health economic
decision making. While these analyses help policy analysts and economists
determine coverage, inform policy, and guide resource allocation, they are
statistically challenging for several reasons. Cost and effectiveness are
correlated and follow complex joint distributions which are difficult to
capture parametrically. Effectiveness (often measured as increased survival
time) and accumulated cost tends to be right-censored in many applications.
Moreover, CEAs are often conducted using observational data with non-random
treatment assignment. Policy-relevant causal estimation therefore requires
robust confounding control. Finally, current CEA methods do not address
cost-effectiveness heterogeneity in a principled way - often presenting
population-averaged estimates even though significant effect heterogeneity may
exist. Motivated by these challenges, we develop a nonparametric Bayesian model
for joint cost-survival distributions in the presence of censoring. Our
approach utilizes a joint Enriched Dirichlet Process prior on the covariate
effects of cost and survival time, while using a Gamma Process prior on the
baseline survival time hazard. Causal CEA estimands, with policy-relevant
interpretations, are identified and estimated via a Bayesian nonparametric
g-computation procedure. Finally, we outline how the induced clustering of the
Enriched Dirichlet Process can be used to adaptively detect presence of
subgroups with different cost-effectiveness profiles. We outline an MCMC
procedure for full posterior inference and evaluate frequentist properties via
simulations. We use our model to assess the cost-efficacy of chemotherapy
versus radiation adjuvant therapy for treating endometrial cancer in the
SEER-Medicare database.
| stat |
Testing equality of spectral density operators for functional linear
processes | The problem of testing equality of the entire second order structure of two
independent functional linear processes is considered. A fully functional
$L^2$-type test is developed which evaluates, over all frequencies, the
Hilbert-Schmidt distance between the estimated spectral density operators of
the two processes. The asymptotic behavior of the test statistic is
investigated and its limiting distribution under the null hypothesis is
derived. Furthermore, a novel frequency domain bootstrap method is developed
which approximates more accurately the distribution of the test statistic under
the null than the large sample Gaussian approximation obtained. Asymptotic
validity of the bootstrap procedure is established and consistency of the
bootstrap-based test under the alternative is proved. Numerical simulations
show that, even for small samples, the bootstrap-based test has very good size
and power behavior. An application to meteorological functional time series is
also presented.
| stat |
Deep Learning in Characteristics-Sorted Factor Models | To study the characteristics-sorted factor model in asset pricing, we develop
a bottom-up approach with state-of-the-art deep learning optimization. With an
economic objective to minimize pricing errors, we train a non-reduced-form
neural network using firm characteristics [inputs], and generate factors
[intermediate features], to fit security returns [outputs]. Sorting securities
on firm characteristics provides a nonlinear activation to create long-short
portfolio weights, as a hidden layer, from lag characteristics to realized
returns. Our model offers an alternative perspective for dimension reduction on
firm characteristics [inputs], rather than factors [intermediate features], and
allows for both nonlinearity and interactions on inputs. Our empirical findings
are twofold. We find robust statistical and economic evidence in out-of-sample
portfolios and individual stock returns. To interpret our deep factors, we show
highly significant results in factor investing via the squared Sharpe ratio
test, as well as improvement in dissecting anomalies.
| stat |
A better measure of relative prediction accuracy for model selection and
model estimation | Surveys show that the mean absolute percentage error (MAPE) is the most
widely used measure of forecast accuracy in businesses and organizations. It is
however, biased: When used to select among competing prediction methods it
systematically selects those whose predictions are too low. This is not widely
discussed and so is not generally known among practitioners. We explain why
this happens. We investigate an alternative relative accuracy measure which
avoids this bias: the log of the accuracy ratio: log (prediction / actual).
Relative accuracy is particularly relevant if the scatter in the data grows as
the value of the variable grows (heteroscedasticity). We demonstrate using
simulations that for heteroscedastic data (modelled by a multiplicative error
factor) the proposed metric is far superior to MAPE for model selection.
Another use for accuracy measures is in fitting parameters to prediction
models. Minimum MAPE models do not predict a simple statistic and so
theoretical analysis is limited. We prove that when the proposed metric is used
instead, the resulting least squares regression model predicts the geometric
mean. This important property allows its theoretical properties to be
understood.
| stat |
Beating SGD Saturation with Tail-Averaging and Minibatching | While stochastic gradient descent (SGD) is one of the major workhorses in
machine learning, the learning properties of many practically used variants are
poorly understood. In this paper, we consider least squares learning in a
nonparametric setting and contribute to filling this gap by focusing on the
effect and interplay of multiple passes, mini-batching and averaging, and in
particular tail averaging. Our results show how these different variants of SGD
can be combined to achieve optimal learning errors, hence providing practical
insights. In particular, we show for the first time in the literature that tail
averaging allows faster convergence rates than uniform averaging in the
nonparametric setting. Finally, we show that a combination of tail-averaging
and minibatching allows more aggressive step-size choices than using any one of
said components.
| stat |
Assessing contribution of treatment phases through tipping point
analyses using rank preserving structural failure time models | In clinical trials, an experimental treatment is sometimes added on to a
standard of care or control therapy in multiple treatment phases (e.g.,
concomitant and maintenance phases) to improve patient outcomes. When the new
regimen provides meaningful benefit over the control therapy in such cases, it
proves difficult to separately assess the contribution of each phase to the
overall effect observed. This article provides an approach for assessing the
importance of a specific treatment phase in such a situation through tipping
point analyses of a time-to-event endpoint using
rank-preserving-structural-failure-time (RPSFT) modeling. A tipping-point
analysis is commonly used in situations where it is suspected that a
statistically significant difference between treatment arms could be a result
of missing or unobserved data instead of a real treatment effect.
Rank-preserving-structural-failure-time modeling is an approach for causal
inference that is typically used to adjust for treatment switching in clinical
trials with time to event endpoints. The methodology proposed in this article
is an amalgamation of these two ideas to investigate the contribution of a
treatment phase of interest to the effect of a regimen comprising multiple
treatment phases. We provide two different variants of the method corresponding
to two different effects of interest. We provide two different tipping point
thresholds depending on inferential goals. The proposed approaches are
motivated and illustrated with data from a recently concluded, real-life phase
3 cancer clinical trial. We then conclude with several considerations and
recommendations.
| stat |
Differentially Private Confidence Intervals | Confidence intervals for the population mean of normally distributed data are
some of the most standard statistical outputs one might want from a database.
In this work we give practical differentially private algorithms for this task.
We provide five algorithms and then compare them to each other and to prior
work. We give concrete, experimental analysis of their accuracy and find that
our algorithms provide much more accurate confidence intervals than prior work.
For example, in one setting (with {\epsilon} = 0.1 and n = 2782) our algorithm
yields an interval that is only 1/15th the size of the standard set by prior
work.
| stat |
Escaping the curse of dimensionality in Bayesian model based clustering | In many applications, there is interest in clustering very high-dimensional
data. A common strategy is first stage dimensionality reduction followed by a
standard clustering algorithm, such as k-means. This approach does not target
dimension reduction to the clustering objective, and fails to quantify
uncertainty. Model-based Bayesian approaches provide an appealing alternative,
but often have poor performance in high-dimensions, producing too many or too
few clusters. This article provides an explanation for this behavior through
studying the clustering posterior in a non-standard setting with fixed sample
size and increasing dimensionality. We show that the finite sample posterior
tends to either assign every observation to a different cluster or all
observations to the same cluster as dimension grows, depending on the kernels
and prior specification but not on the true data-generating model. To find
models avoiding this pitfall, we define a Bayesian oracle for clustering, with
the oracle clustering posterior based on the true values of low-dimensional
latent variables. We define a class of LAtent Mixtures for Bayesian (Lamb)
clustering that have equivalent behavior to this oracle as dimension grows.
Lamb is shown to have good performance in simulation studies and an application
to inferring cell types based on scRNAseq.
| stat |
Learning to detect chest radiographs containing lung nodules using
visual attention networks | Machine learning approaches hold great potential for the automated detection
of lung nodules in chest radiographs, but training the algorithms requires vary
large amounts of manually annotated images, which are difficult to obtain. Weak
labels indicating whether a radiograph is likely to contain pulmonary nodules
are typically easier to obtain at scale by parsing historical free-text
radiological reports associated to the radiographs. Using a repositotory of
over 700,000 chest radiographs, in this study we demonstrate that promising
nodule detection performance can be achieved using weak labels through
convolutional neural networks for radiograph classification. We propose two
network architectures for the classification of images likely to contain
pulmonary nodules using both weak labels and manually-delineated bounding
boxes, when these are available. Annotated nodules are used at training time to
deliver a visual attention mechanism informing the model about its localisation
performance. The first architecture extracts saliency maps from high-level
convolutional layers and compares the estimated position of a nodule against
the ground truth, when this is available. A corresponding localisation error is
then back-propagated along with the softmax classification error. The second
approach consists of a recurrent attention model that learns to observe a short
sequence of smaller image portions through reinforcement learning. When a
nodule annotation is available at training time, the reward function is
modified accordingly so that exploring portions of the radiographs away from a
nodule incurs a larger penalty. Our empirical results demonstrate the potential
advantages of these architectures in comparison to competing methodologies.
| stat |
Training Generative Adversarial Networks by Solving Ordinary
Differential Equations | The instability of Generative Adversarial Network (GAN) training has
frequently been attributed to gradient descent. Consequently, recent methods
have aimed to tailor the models and training procedures to stabilise the
discrete updates. In contrast, we study the continuous-time dynamics induced by
GAN training. Both theory and toy experiments suggest that these dynamics are
in fact surprisingly stable. From this perspective, we hypothesise that
instabilities in training GANs arise from the integration error in discretising
the continuous dynamics. We experimentally verify that well-known ODE solvers
(such as Runge-Kutta) can stabilise training - when combined with a regulariser
that controls the integration error. Our approach represents a radical
departure from previous methods which typically use adaptive optimisation and
stabilisation techniques that constrain the functional space (e.g. Spectral
Normalisation). Evaluation on CIFAR-10 and ImageNet shows that our method
outperforms several strong baselines, demonstrating its efficacy.
| stat |
Source-specific contributions of particulate matter to asthma-related
pediatric emergency department utilization | Few studies have linked specific sources of ambient particulate matter
smaller than 2.5 $\mu$m (PM2.5) and asthma. In this study, we estimated the
contributions of specific sources to PM2.5 and examined their association with
daily asthma hospital utilization in Cincinnati, Ohio, USA. We used Poisson
regression models to estimate the daily number of asthma ED visits the day of
and one, and two days following separate increases in PM2.5 and its source
components, adjusting for temporal trends, holidays, temperature, and humidity.
In addition, we used a model-based clustering method to group days with similar
source-specific contributions into six distinct clusters. Specifically,
elevated PM2.5 concentrations occurring on days characterized by low
contributions of coal combustion showed a significantly reduced risk of
hospital utilization for asthma (rate ratio: 0.86, 95% CI: [0.77, 0.95])
compared to other clusters. Reducing the contribution of coal combustion to
PM2.5 levels could be an effective intervention for reducing asthma-related
hospital utilization.
| stat |
Estimation of Extreme Survival Probabilities with Cox Model | We propose an extension of the regular Cox's proportional hazards model which
allows the estimation of the probabilities of rare events. It is known that
when the data are heavily censored at the upper end of the survival
distribution, the estimation of the tail of the survival distribution is not
reliable. To estimate the distribution beyond the last observed data, we
suppose that the survival data are in the domain of attraction of the Fr\'echet
distribution conditionally to covariates. Under this condition, by the
Fisher-Tippett-Gnedenko theorem, the tail of the baseline distribution can be
adjusted by a Pareto distribution with parameter $\theta$ beyond a threshold
$\tau$. The survival distributions conditioned to the covariates are easily
computed from the baseline. We also propose an aggregated estimate of the
survival probabilities. A procedure allowing an automatic choice of the
threshold and an application on two data sets are given.
| stat |
Variational Bayesian Sequence-to-Sequence Networks for Memory-Efficient
Sign Language Translation | Memory-efficient continuous Sign Language Translation is a significant
challenge for the development of assisted technologies with real-time
applicability for the deaf. In this work, we introduce a paradigm of designing
recurrent deep networks whereby the output of the recurrent layer is derived
from appropriate arguments from nonparametric statistics. A novel variational
Bayesian sequence-to-sequence network architecture is proposed that consists of
a) a full Gaussian posterior distribution for data-driven memory compression
and b) a nonparametric Indian Buffet Process prior for regularization applied
on the Gated Recurrent Unit non-gate weights. We dub our approach
Stick-Breaking Recurrent network and show that it can achieve a substantial
weight compression without diminishing modeling performance.
| stat |
A Potential Outcomes Calculus for Identifying Conditional Path-Specific
Effects | The do-calculus is a well-known deductive system for deriving connections
between interventional and observed distributions, and has been proven complete
for a number of important identifiability problems in causal inference.
Nevertheless, as it is currently defined, the do-calculus is inapplicable to
causal problems that involve complex nested counterfactuals which cannot be
expressed in terms of the "do" operator. Such problems include analyses of
path-specific effects and dynamic treatment regimes. In this paper we present
the potential outcome calculus (po-calculus), a natural generalization of
do-calculus for arbitrary potential outcomes. We thereby provide a bridge
between identification approaches which have their origins in artificial
intelligence and statistics, respectively. We use po-calculus to give a
complete identification algorithm for conditional path-specific effects with
applications to problems in mediation analysis and algorithmic fairness.
| stat |
Robust Bayesian variable selection for gene-environment interactions | Gene-environment (G$\times$E) interactions have important implications to
elucidate the etiology of complex diseases beyond the main genetic and
environmental effects. Outliers and data contamination in disease phenotypes of
G$\times$E studies have been commonly encountered, leading to the development
of a broad spectrum of robust regularization methods. Nevertheless, within the
Bayesian framework, the issue has not been taken care of in existing studies.
We develop a fully Bayesian robust variable selection method for G$\times$E
interaction studies. The proposed Bayesian method can effectively accommodate
heavy-tailed errors and outliers in the response variable while conducting
variable selection by accounting for structural sparsity. In particular, for
the robust sparse group selection, the spike-and-slab priors have been imposed
on both individual and group levels to identify important main and interaction
effects robustly. An efficient Gibbs sampler has been developed to facilitate
fast computation. Extensive simulation studies and analysis of both the
diabetes data with SNP measurements from the Nurses' Health Study and TCGA
melanoma data with gene expression measurements demonstrate the superior
performance of the proposed method over multiple competing alternatives.
| stat |
Differentiable Learning Under Triage | Multiple lines of evidence suggest that predictive models may benefit from
algorithmic triage. Under algorithmic triage, a predictive model does not
predict all instances but instead defers some of them to human experts.
However, the interplay between the prediction accuracy of the model and the
human experts under algorithmic triage is not well understood. In this work, we
start by formally characterizing under which circumstances a predictive model
may benefit from algorithmic triage. In doing so, we also demonstrate that
models trained for full automation may be suboptimal under triage. Then, given
any model and desired level of triage, we show that the optimal triage policy
is a deterministic threshold rule in which triage decisions are derived
deterministically by thresholding the difference between the model and human
errors on a per-instance level. Building upon these results, we introduce a
practical gradient-based algorithm that is guaranteed to find a sequence of
triage policies and predictive models of increasing performance. Experiments on
a wide variety of supervised learning tasks using synthetic and real data from
two important applications -- content moderation and scientific discovery --
illustrate our theoretical results and show that the models and triage policies
provided by our gradient-based algorithm outperform those provided by several
competitive baselines.
| stat |
SMIM: a unified framework of Survival sensitivity analysis using
Multiple Imputation and Martingale | Censored survival data are common in clinical trial studies. We propose a
unified framework for sensitivity analysis to censoring at random in survival
data using multiple imputation and martingale, called SMIM. The proposed
framework adopts the \delta-adjusted and control-based models, indexed by the
sensitivity parameter, entailing censoring at random and a wide collection of
censoring not at random assumptions. Also, it targets for a broad class of
treatment effect estimands defined as functionals of treatment-specific
survival functions, taking into account of missing data due to censoring.
Multiple imputation facilitates the use of simple full-sample estimation;
however, the standard Rubin's combining rule may overestimate the variance for
inference in the sensitivity analysis framework. We decompose the multiple
imputation estimator into a martingale series based on the sequential
construction of the estimator and propose the wild bootstrap inference by
resampling the martingale series. The new bootstrap inference has a theoretical
guarantee for consistency and is computationally efficient compared to the
non-parametric bootstrap counterpart. We evaluate the finite-sample performance
of the proposed SMIM through simulation and applications on HIV and
cardiovascular clinical trials.
| stat |
Modified Dorfman procedure for pool tests with dilution -- COVID-19 case
study | The outbreak of the global COVID-19 pandemic results in unprecedented demand
for fast and efficient testing of large numbers of patients for the presence of
SARS-CoV-2 coronavirus. Beside technical improvements of the cost and speed of
individual tests, pool testing may be used to improve efficiency and throughput
of a population test. Dorfman pool testing procedure is one of the best known
and studied methods of this kind. This procedure is, however, based on
unrealistic assumptions that the pool test has perfect sensitivity and the only
objective is to minimize the number of tests, and is not well adapted to the
case of imperfect pool tests. We propose and analyze a simple modification of
this procedure in which test of a pool with negative result is independently
repeated up to several times. The proposed procedure is evaluated in a
computational study using recent data about dilution effect for SARS-CoV-2 PCR
tests, showing that the proposed approach significantly reduces the number of
false negatives with a relatively small increase of the number of tests,
especially for small prevalence rates. For example, for prevalence rate 0.001
the number of tests could be reduced to 22.1% of individual tests, increasing
the expected number of false negatives by no more than 1%, and to 16.8% of
individual tests increasing the expected number of false negatives by no more
than 10%. At the same time, a similar reduction of the expected number of tests
in the standard Dorfman procedure would yield 675% and 821% increase of the
expected number of false negatives, respectively. This makes the proposed
procedure an interesting choice for screening tests in the case of diseases
like COVID-19.
| stat |
Modeling Heterogeneous Peer Assortment Effects using Finite Mixture
Exponential Random Graph Models | This article develops a class of models called Sender/Receiver Finite Mixture
Exponential Random Graph Models (SRFM-ERGMs) that enables inference on
networks. This class of models extends the existing Exponential Random Graph
Modeling framework to allow analysts to model unobserved heterogeneity in the
effects of nodal covariates and network features. An empirical example
regarding substance use among adolescents is presented. Simulations across a
variety of conditions are used to evaluate the performance of this technique.
We conclude that that unobserved heterogeneity in effects of nodal covariates
can be a major cause of mis-fit in network models, and the SRFM-ERGM approach
can alleviate this misfit. Implications for the analysis of social networks in
psychological science are discussed.
| stat |
Visually Communicating and Teaching Intuition for Influence Functions | Estimators based on influence functions (IFs) have been shown to be effective
in many settings, especially when combined with machine learning techniques. By
focusing on estimating a specific target of interest (e.g., the average effect
of a treatment), rather than on estimating the full underlying data generating
distribution, IF-based estimators are often able to achieve asymptotically
optimal mean-squared error. Still, many researchers find IF-based estimators to
be opaque or overly technical, which makes their use less prevalent and their
benefits less available. To help foster understanding and trust in IF-based
estimators, we present tangible, visual illustrations of when and how IF-based
estimators can outperform standard ``plug-in'' estimators. The figures we show
are based on connections between IFs, gradients, linear approximations, and
Newton-Raphson.
| stat |
Variational Autoencoders and Nonlinear ICA: A Unifying Framework | The framework of variational autoencoders allows us to efficiently learn deep
latent-variable models, such that the model's marginal distribution over
observed variables fits the data. Often, we're interested in going a step
further, and want to approximate the true joint distribution over observed and
latent variables, including the true prior and posterior distributions over
latent variables. This is known to be generally impossible due to
unidentifiability of the model. We address this issue by showing that for a
broad family of deep latent-variable models, identification of the true joint
distribution over observed and latent variables is actually possible up to very
simple transformations, thus achieving a principled and powerful form of
disentanglement. Our result requires a factorized prior distribution over the
latent variables that is conditioned on an additionally observed variable, such
as a class label or almost any other observation. We build on recent
developments in nonlinear ICA, which we extend to the case with noisy,
undercomplete or discrete observations, integrated in a maximum likelihood
framework. The result also trivially contains identifiable flow-based
generative models as a special case.
| stat |
PAPRIKA: Private Online False Discovery Rate Control | In hypothesis testing, a false discovery occurs when a hypothesis is
incorrectly rejected due to noise in the sample. When adaptively testing
multiple hypotheses, the probability of a false discovery increases as more
tests are performed. Thus the problem of False Discovery Rate (FDR) control is
to find a procedure for testing multiple hypotheses that accounts for this
effect in determining the set of hypotheses to reject. The goal is to minimize
the number (or fraction) of false discoveries, while maintaining a high true
positive rate (i.e., correct discoveries).
In this work, we study False Discovery Rate (FDR) control in multiple
hypothesis testing under the constraint of differential privacy for the sample.
Unlike previous work in this direction, we focus on the online setting, meaning
that a decision about each hypothesis must be made immediately after the test
is performed, rather than waiting for the output of all tests as in the offline
setting. We provide new private algorithms based on state-of-the-art results in
non-private online FDR control. Our algorithms have strong provable guarantees
for privacy and statistical performance as measured by FDR and power. We also
provide experimental results to demonstrate the efficacy of our algorithms in a
variety of data environments.
| stat |
Weighted Fisher Discriminant Analysis in the Input and Feature Spaces | Fisher Discriminant Analysis (FDA) is a subspace learning method which
minimizes and maximizes the intra- and inter-class scatters of data,
respectively. Although, in FDA, all the pairs of classes are treated the same
way, some classes are closer than the others. Weighted FDA assigns weights to
the pairs of classes to address this shortcoming of FDA. In this paper, we
propose a cosine-weighted FDA as well as an automatically weighted FDA in which
weights are found automatically. We also propose a weighted FDA in the feature
space to establish a weighted kernel FDA for both existing and newly proposed
weights. Our experiments on the ORL face recognition dataset show the
effectiveness of the proposed weighting schemes.
| stat |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.