title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
MoFlow: An Invertible Flow Model for Generating Molecular Graphs | Generating molecular graphs with desired chemical properties driven by deep
graph generative models provides a very promising way to accelerate drug
discovery process. Such graph generative models usually consist of two steps:
learning latent representations and generation of molecular graphs. However, to
generate novel and chemically-valid molecular graphs from latent
representations is very challenging because of the chemical constraints and
combinatorial complexity of molecular graphs. In this paper, we propose MoFlow,
a flow-based graph generative model to learn invertible mappings between
molecular graphs and their latent representations. To generate molecular
graphs, our MoFlow first generates bonds (edges) through a Glow based model,
then generates atoms (nodes) given bonds by a novel graph conditional flow, and
finally assembles them into a chemically valid molecular graph with a posthoc
validity correction. Our MoFlow has merits including exact and tractable
likelihood training, efficient one-pass embedding and generation, chemical
validity guarantees, 100\% reconstruction of training data, and good
generalization ability. We validate our model by four tasks: molecular graph
generation and reconstruction, visualization of the continuous latent space,
property optimization, and constrained property optimization. Our MoFlow
achieves state-of-the-art performance, which implies its potential efficiency
and effectiveness to explore large chemical space for drug discovery.
| stat |
Minimax Classification with 0-1 Loss and Performance Guarantees | Supervised classification techniques use training samples to find
classification rules with small expected 0-1 loss. Conventional methods achieve
efficient learning and out-of-sample generalization by minimizing surrogate
losses over specific families of rules. This paper presents minimax risk
classifiers (MRCs) that do not rely on a choice of surrogate loss and family of
rules. MRCs achieve efficient learning and out-of-sample generalization by
minimizing worst-case expected 0-1 loss w.r.t. uncertainty sets that are
defined by linear constraints and include the true underlying distribution. In
addition, MRCs' learning stage provides performance guarantees as lower and
upper tight bounds for expected 0-1 loss. We also present MRCs' finite-sample
generalization bounds in terms of training size and smallest minimax risk, and
show their competitive classification performance w.r.t. state-of-the-art
techniques using benchmark datasets.
| stat |
Probabilistic Structure Learning for EEG/MEG Source Imaging with
Hierarchical Graph Prior | Brain source imaging is an important method for noninvasively characterizing
brain activity using Electroencephalogram (EEG) or Magnetoencephalography (MEG)
recordings. Traditional EEG/MEG Source Imaging (ESI) methods usually assume
that either source activity at different time points is unrelated, or that
similar spatiotemporal patterns exist across an entire study period. The former
assumption makes ESI analyses sensitive to noise, while the latter renders ESI
analyses unable to account for time-varying patterns of activity. To
effectively deal with noise while maintaining flexibility and continuity among
brain activation patterns, we propose a novel probabilistic ESI model based on
a hierarchical graph prior. Under our method, a spanning tree constraint
ensures that activity patterns have spatiotemporal continuity. An efficient
algorithm based on alternating convex search is presented to solve the proposed
model and is provably convergent. Comprehensive numerical studies using
synthetic data on a real brain model are conducted under different levels of
signal-to-noise ratio (SNR) from both sensor and source spaces. We also examine
the EEG/MEG data in a real application, in which our ESI reconstructions are
neurologically plausible. All the results demonstrate significant improvements
of the proposed algorithm over the benchmark methods in terms of source
localization performance, especially at high noise levels.
| stat |
Efficient Importance Sampling for the Left Tail of Positive Gaussian
Quadratic Forms | Estimating the left tail of quadratic forms in Gaussian random vectors is of
major practical importance in many applications. In this letter, we propose an
efficient importance sampling estimator that is endowed with the bounded
relative error property. This property significantly reduces the number of
simulation runs required by the proposed estimator compared to naive Monte
Carlo (MC), especially when the probability of interest is very small. Selected
simulation results are presented to illustrate the efficiency of our estimator
compared to naive MC as well as some of the well-known approximations.
| stat |
Joint Interaction and Trajectory Prediction for Autonomous Driving using
Graph Neural Networks | In this work, we aim to predict the future motion of vehicles in a traffic
scene by explicitly modeling their pairwise interactions. Specifically, we
propose a graph neural network that jointly predicts the discrete interaction
modes and 5-second future trajectories for all agents in the scene. Our model
infers an interaction graph whose nodes are agents and whose edges capture the
long-term interaction intents among the agents. In order to train the model to
recognize known modes of interaction, we introduce an auto-labeling function to
generate ground truth interaction labels. Using a large-scale real-world
driving dataset, we demonstrate that jointly predicting the trajectories along
with the explicit interaction types leads to significantly lower trajectory
error than baseline methods. Finally, we show through simulation studies that
the learned interaction modes are semantically meaningful.
| stat |
Dirichlet-tree multinomial mixtures for clustering microbiome
compositions | Studying the human microbiome has gained substantial interest in recent
years, and a common task in the analysis of these data is to cluster microbiome
compositions into subtypes. This subdivision of samples into subgroups serves
as an intermediary step in achieving personalized diagnosis and treatment. In
applying existing clustering methods to modern microbiome studies including the
American Gut Project (AGP) data, we found that this seemingly standard task,
however, is very challenging in the microbiome composition context due to
several key features of such data. Standard distance-based clustering
algorithms generally do not produce reliable results as they do not take into
account the heterogeneity of the cross-sample variability among the bacterial
taxa, while existing model-based approaches do not allow sufficient flexibility
for the identification of complex within-cluster variation from cross-cluster
variation. Direct applications of such methods generally lead to overly
dispersed clusters in the AGP data and such phenomenon is common for other
microbiome data. To overcome these challenges, we introduce Dirichlet-tree
multinomial mixtures (DTMM) as a Bayesian generative model for clustering
amplicon sequencing data in microbiome studies. DTMM models the microbiome
population with a mixture of Dirichlet-tree kernels that utilizes the
phylogenetic tree to offer a more flexible covariance structure in
characterizing within-cluster variation, and it provides a means for
identifying a subset of signature taxa that distinguish the clusters. We
perform extensive simulation studies to evaluate the performance of DTMM and
compare it to state-of-the-art model-based and distance-based clustering
methods in the microbiome context. Finally, we report a case study on the fecal
data from the AGP to identify compositional clusters among individuals with
inflammatory bowel disease and diabetes.
| stat |
Theoretical bounds on estimation error for meta-learning | Machine learning models have traditionally been developed under the
assumption that the training and test distributions match exactly. However,
recent success in few-shot learning and related problems are encouraging signs
that these models can be adapted to more realistic settings where train and
test distributions differ. Unfortunately, there is severely limited theoretical
support for these algorithms and little is known about the difficulty of these
problems. In this work, we provide novel information-theoretic lower-bounds on
minimax rates of convergence for algorithms that are trained on data from
multiple sources and tested on novel data. Our bounds depend intuitively on the
information shared between sources of data, and characterize the difficulty of
learning in this setting for arbitrary algorithms. We demonstrate these bounds
on a hierarchical Bayesian model of meta-learning, computing both upper and
lower bounds on parameter estimation via maximum-a-posteriori inference.
| stat |
Multi-view Data Visualisation via Manifold Learning | Non-linear dimensionality reduction can be performed by \textit{manifold
learning} approaches, such as Stochastic Neighbour Embedding (SNE), Locally
Linear Embedding (LLE) and Isometric Feature Mapping (ISOMAP). These methods
aim to produce two or three latent embeddings, primarily to visualise the data
in intelligible representations. This manuscript proposes extensions of
Student's t-distributed SNE (t-SNE), LLE and ISOMAP, for dimensionality
reduction and visualisation of multi-view data. Multi-view data refers to
multiple types of data generated from the same samples. The proposed multi-view
approaches provide more comprehensible projections of the samples compared to
the ones obtained by visualising each data-view separately. Commonly
visualisation is used for identifying underlying patterns within the samples.
By incorporating the obtained low-dimensional embeddings from the multi-view
manifold approaches into the K-means clustering algorithm, it is shown that
clusters of the samples are accurately identified. Through the analysis of real
and synthetic data the proposed multi-SNE approach is found to have the best
performance. We further illustrate the applicability of the multi-SNE approach
for the analysis of multi-omics single-cell data, where the aim is to visualise
and identify cell heterogeneity and cell types in biological tissues relevant
to health and disease.
| stat |
The Identification and Estimation of Direct and Indirect Effects in A/B
Tests through Causal Mediation Analysis | E-commerce companies have a number of online products, such as organic
search, sponsored search, and recommendation modules, to fulfill customer
needs. Although each of these products provides a unique opportunity for users
to interact with a portion of the overall inventory, they are all similar
channels for users and compete for limited time and monetary budgets of users.
To optimize users' overall experiences on an E-commerce platform, instead of
understanding and improving different products separately, it is important to
gain insights into the evidence that a change in one product would induce users
to change their behaviors in others, which may be due to the fact that these
products are functionally similar. In this paper, we introduce causal mediation
analysis as a formal statistical tool to reveal the underlying causal
mechanisms. Existing literature provides little guidance on cases where
multiple unmeasured causally-dependent mediators exist, which are common in A/B
tests. We seek a novel approach to identify in those scenarios direct and
indirect effects of the treatment. In the end, we demonstrate the effectiveness
of the proposed method in data from Etsy's real A/B tests and shed lights on
complex relationships between different products.
| stat |
Modeling High-Dimensional Time Series: A Factor Model with Dynamically
Dependent Factors and Diverging Eigenvalues | This article proposes a new approach to modeling high-dimensional time series
by treating a $p$-dimensional time series as a nonsingular linear
transformation of certain common factors and idiosyncratic components. Unlike
the approximate factor models, we assume that the factors capture all the
non-trivial dynamics of the data, but the cross-sectional dependence may be
explained by both the factors and the idiosyncratic components. Under the
proposed model, (a) the factor process is dynamically dependent and the
idiosyncratic component is a white noise process, and (b) the largest
eigenvalues of the covariance matrix of the idiosyncratic components may
diverge to infinity as the dimension $p$ increases. We propose a white noise
testing procedure for high-dimensional time series to determine the number of
white noise components and, hence, the number of common factors, and introduce
a projected Principal Component Analysis (PCA) to eliminate the diverging
effect of the idiosyncratic noises. Asymptotic properties of the proposed
method are established for both fixed $p$ and diverging $p$ as the sample size
$n$ increases to infinity. We use both simulated data and real examples to
assess the performance of the proposed method. We also compare our method with
two commonly used methods in the literature concerning the forecastability of
the extracted factors and find that the proposed approach not only provides
interpretable results, but also performs well in out-of-sample forecasting.
Supplementary materials of the article are available online.
| stat |
Having a Ball: evaluating scoring streaks and game excitement using
in-match trend estimation | Many popular sports involve matches between two teams or players where each
team have the possibility of scoring points throughout the match. While the
overall match winner and result is interesting, it conveys little information
about the underlying scoring trends throughout the match. Modeling approaches
that accommodate a finer granularity of the score difference throughout the
match is needed to evaluate in-game strategies, discuss scoring streaks, teams
strengths, and other aspects of the game.
We propose a latent Gaussian process to model the score difference between
two teams and introduce the Trend Direction Index as an easily interpretable
probabilistic measure of the current trend in the match as well as a measure of
post-game trend evaluation. In addition we propose the Excitement Trend Index -
the expected number of monotonicity changes in the running score difference -
as a measure of overall game excitement.
Our proposed methodology is applied to all 1143 matches from the 2019-2020
National Basketball Association (NBA) season. We show how the trends can be
interpreted in individual games and how the excitement score can be used to
cluster teams according to how exciting they are to watch.
| stat |
Stochastic tree ensembles for regularized nonlinear regression | This paper develops a novel stochastic tree ensemble method for nonlinear
regression, referred to as Accelerated Bayesian Additive Regression Trees, or
XBART. By combining regularization and stochastic search strategies from
Bayesian modeling with computationally efficient techniques from recursive
partitioning algorithms, XBART attains state-of-the-art performance at
prediction and function estimation. Simulation studies demonstrate that XBART
provides accurate point-wise estimates of the mean function and does so faster
than popular alternatives, such as BART, XGBoost, and neural networks (using
Keras) on a variety of test functions. Additionally, it is demonstrated that
using XBART to initialize the standard BART MCMC algorithm considerably
improves credible interval coverage and reduces total run-time. Finally, two
basic theoretical results are established: the single tree version of the model
is asymptotically consistent and the Markov chain produced by the ensemble
version of the algorithm has a unique stationary distribution.
| stat |
A Nearest-Neighbor Based Nonparametric Test for Viral Remodeling in
Heterogeneous Single-Cell Proteomic Data | An important problem in contemporary immunology studies based on single-cell
protein expression data is to determine whether cellular expressions are
remodeled post infection by a pathogen. One natural approach for detecting such
changes is to use non-parametric two-sample statistical tests. However, in
single-cell studies, direct application of these tests is often inadequate
because single-cell level expression data from uninfected populations often
contains attributes of several latent sub-populations with highly heterogeneous
characteristics. As a result, viruses often infect these different
sub-populations at different rates in which case the traditional nonparametric
two-sample tests for checking similarity in distributions are no longer
conservative. We propose a new nonparametric method for Testing Remodeling
Under Heterogeneity (TRUH) that can accurately detect changes in the infected
samples compared to possibly heterogeneous uninfected samples. Our testing
framework is based on composite nulls and is designed to allow the null model
to encompass the possibility that the infected samples, though unaltered by the
virus, might be dominantly arising from under-represented sub-populations in
the baseline data. The TRUH statistic, which uses nearest neighbor projections
of the infected samples into the baseline uninfected population, is calibrated
using a novel bootstrap algorithm. We demonstrate the non-asymptotic
performance of the test via simulation experiments and derive the large sample
limit of the test statistic, which provides theoretical support towards
consistent asymptotic calibration of the test. We use the TRUH statistic for
studying remodeling in tonsillar T cells under different types of HIV infection
and find that unlike traditional tests, TRUH based statistical inference
conforms to the biologically validated immunological theories on HIV infection.
| stat |
Manifold structure in graph embeddings | Statistical analysis of a graph often starts with embedding, the process of
representing its nodes as points in space. How to choose the embedding
dimension is a nuanced decision in practice, but in theory a notion of true
dimension is often available. In spectral embedding, this dimension may be very
high. However, this paper shows that existing random graph models, including
graphon and other latent position models, predict the data should live near a
much lower-dimensional set. One may therefore circumvent the curse of
dimensionality by employing methods which exploit hidden manifold structure.
| stat |
Projecting "better than randomly": How to reduce the dimensionality of
very large datasets in a way that outperforms random projections | For very large datasets, random projections (RP) have become the tool of
choice for dimensionality reduction. This is due to the computational
complexity of principal component analysis. However, the recent development of
randomized principal component analysis (RPCA) has opened up the possibility of
obtaining approximate principal components on very large datasets. In this
paper, we compare the performance of RPCA and RP in dimensionality reduction
for supervised learning. In Experiment 1, study a malware classification task
on a dataset with over 10 million samples, almost 100,000 features, and over 25
billion non-zero values, with the goal of reducing the dimensionality to a
compressed representation of 5,000 features. In order to apply RPCA to this
dataset, we develop a new algorithm called large sample RPCA (LS-RPCA), which
extends the RPCA algorithm to work on datasets with arbitrarily many samples.
We find that classification performance is much higher when using LS-RPCA for
dimensionality reduction than when using random projections. In particular,
across a range of target dimensionalities, we find that using LS-RPCA reduces
classification error by between 37% and 54%. Experiment 2 generalizes the
phenomenon to multiple datasets, feature representations, and classifiers.
These findings have implications for a large number of research projects in
which random projections were used as a preprocessing step for dimensionality
reduction. As long as accuracy is at a premium and the target dimensionality is
sufficiently less than the numeric rank of the dataset, randomized PCA may be a
superior choice. Moreover, if the dataset has a large number of samples, then
LS-RPCA will provide a method for obtaining the approximate principal
components.
| stat |
Efficient computation and analysis of distributional Shapley values | Distributional data Shapley value (DShapley) has recently been proposed as a
principled framework to quantify the contribution of individual datum in
machine learning. DShapley develops the foundational game theory concept of
Shapley values into a statistical framework and can be applied to identify data
points that are useful (or harmful) to a learning algorithm. Estimating
DShapley is computationally expensive, however, and this can be a major
challenge to using it in practice. Moreover, there has been little mathematical
analyses of how this value depends on data characteristics. In this paper, we
derive the first analytic expressions for DShapley for the canonical problems
of linear regression, binary classification, and non-parametric density
estimation. These analytic forms provide new algorithms to estimate DShapley
that are several orders of magnitude faster than previous state-of-the-art
methods. Furthermore, our formulas are directly interpretable and provide
quantitative insights into how the value varies for different types of data. We
demonstrate the practical efficacy of our approach on multiple real and
synthetic datasets.
| stat |
Optimistic search strategy: Change point detection for large-scale data
via adaptive logarithmic queries | As a classical and ever reviving topic, change point detection is often
formulated as a search for the maximum of a gain function describing improved
fits when segmenting the data. Searching through all candidate split points on
the grid for finding the best one requires $O(T)$ evaluations of the gain
function for an interval with $T$ observations. If each evaluation is
computationally demanding (e.g. in high-dimensional models), this can become
infeasible. Instead, we propose optimistic search strategies with $O(\log T)$
evaluations exploiting specific structure of the gain function.
Towards solid understanding of our strategies, we investigate in detail the
classical univariate Gaussian change in mean setup. For some of our proposals
we prove asymptotic minimax optimality for single and multiple change point
scenarios. Our search strategies generalize far beyond the theoretically
analyzed univariate setup. We illustrate, as an example, massive computational
speedup in change point detection for high-dimensional Gaussian graphical
models. More generally, we demonstrate empirically that our optimistic search
methods lead to competitive estimation performance while heavily reducing
run-time.
| stat |
Spillover Effects in Cluster Randomized Trials with Noncompliance | Cluster randomized trials (CRTs) are popular in public health and in the
social sciences to evaluate a new treatment or policy where the new policy is
randomly allocated to clusters of units rather than individual units. CRTs
often feature both noncompliance, when individuals within a cluster are not
exposed to the intervention, and individuals within a cluster may influence
each other through treatment spillovers where those who comply with the new
policy may affect the outcomes of those who do not. Here, we study the
identification of causal effects in CRTs when both noncompliance and treatment
spillovers are present. We prove that the standard analysis of CRT data with
noncompliance using instrumental variables does not identify the usual complier
average causal effect when treatment spillovers are present. We extend this
result and show that no analysis of CRT data can unbiasedly estimate local
network causal effects. Finally, we develop bounds for these causal effects
under the assumption that the treatment is not harmful compared to the control.
We demonstrate these results with an empirical study of a deworming
intervention in Kenya.
| stat |
Learning ReLU Networks on Linearly Separable Data: Algorithm,
Optimality, and Generalization | Neural networks with REctified Linear Unit (ReLU) activation functions
(a.k.a. ReLU networks) have achieved great empirical success in various
domains. Nonetheless, existing results for learning ReLU networks either pose
assumptions on the underlying data distribution being e.g. Gaussian, or require
the network size and/or training size to be sufficiently large. In this
context, the problem of learning a two-layer ReLU network is approached in a
binary classification setting, where the data are linearly separable and a
hinge loss criterion is adopted. Leveraging the power of random noise
perturbation, this paper presents a novel stochastic gradient descent (SGD)
algorithm, which can \emph{provably} train any single-hidden-layer ReLU network
to attain global optimality, despite the presence of infinitely many bad local
minima, maxima, and saddle points in general. This result is the first of its
kind, requiring no assumptions on the data distribution, training/network size,
or initialization. Convergence of the resultant iterative algorithm to a global
minimum is analyzed by establishing both an upper bound and a lower bound on
the number of non-zero updates to be performed. Moreover, generalization
guarantees are developed for ReLU networks trained with the novel SGD
leveraging classic compression bounds. These guarantees highlight a key
difference (at least in the worst case) between reliably learning a ReLU
network as well as a leaky ReLU network in terms of sample complexity.
Numerical tests using both synthetic data and real images validate the
effectiveness of the algorithm and the practical merits of the theory.
| stat |
Community Vibrancy and its Relationship with Safety in Philadelphia | To what extent can the strength of a local urban community impact
neighborhood safety? We constructed measures of community vibrancy based on a
unique dataset of block party permit approvals from the City of Philadelphia.
We use both regression modeling and propensity score matching to control for
the economic, demographic and land use characteristics of the surrounding
neighborhood when examining the relationship betweeen crime and our measures of
community vibrancy. We conduct our analysis on aggregate levels of crime and
community vibrancy from 2006 to 2015 as well as the trends in community
vibrancy and crime over this time period. We find that neighborhoods with a
higher number of block parties have a significantly higher crime rate, while
those holding a greater proportion of spontaneous block party events have a
significantly lower crime rate. We also find that neighborhoods with an
increase in the ratio spontaneous block parties over time are significantly
more likely to have a decreasing trend in total crime incidence over that same
time period.
| stat |
Multi-Parameter Regression Survival Modelling: An Alternative to
Proportional Hazards | It is standard practice for covariates to enter a parametric model through a
single distributional parameter of interest, for example, the scale parameter
in many standard survival models. Indeed, the well-known proportional hazards
model is of this kind. In this paper we discuss a more general approach whereby
covariates enter the model through more than one distributional parameter
simultaneously (e.g., scale and shape parameters). We refer to this practice as
"multi-parameter regression" (MPR) modelling and explore its use in a survival
analysis context. We find that multi-parameter regression leads to more
flexible models which can offer greater insight into the underlying data
generating process. To illustrate the concept, we consider the two-parameter
Weibull model which leads to time-dependent hazard ratios, thus relaxing the
typical proportional hazards assumption and motivating a new test of
proportionality. A novel variable selection strategy is introduced for such
multi-parameter regression models. It accounts for the correlation arising
between the estimated regression coefficients in two or more linear predictors
-- a feature which has not been considered by other authors in similar
settings. The methods discussed have been implemented in the mpr package in R.
| stat |
Sparse Semi-supervised Heterogeneous Interbattery Bayesian Analysis | The Bayesian approach to feature extraction, known as factor analysis (FA),
has been widely studied in machine learning to obtain a latent representation
of the data. An adequate selection of the probabilities and priors of these
bayesian models allows the model to better adapt to the data nature (i.e.
heterogeneity, sparsity), obtaining a more representative latent space.
The objective of this article is to propose a general FA framework capable of
modelling any problem. To do so, we start from the Bayesian Inter-Battery
Factor Analysis (BIBFA) model, enhancing it with new functionalities to be able
to work with heterogeneous data, include feature selection, and handle missing
values as well as semi-supervised problems.
The performance of the proposed model, Sparse Semi-supervised Heterogeneous
Interbattery Bayesian Analysis (SSHIBA) has been tested on 4 different
scenarios to evaluate each one of its novelties, showing not only a great
versatility and an interpretability gain, but also outperforming most of the
state-of-the-art algorithms.
| stat |
On optimal two-stage testing of multiple mediators | Mediation analysis in high-dimensional settings often involves identifying
potential mediators among a large number of measured variables. For this
purpose, a two-step familywise error rate procedure called ScreenMin has been
recently proposed (Djordjilovi\'c et al. 2019). In ScreenMin, variables are
first screened and only those that pass the screening are tested. The proposed
threshold for selection has been shown to guarantee asymptotic familywise error
rate. In this work, we investigate the impact of the selection threshold on the
finite sample familywise error rate. We derive a power maximizing selection
threshold and show that it is well approximated by an adaptive threshold of
Wang et al. (2016). We illustrate the investigated procedures on a case-control
study examining the effect of fish intake on the risk of colorectal adenoma.
| stat |
Information Plane Analysis of Deep Neural Networks via Matrix-Based
Renyi's Entropy and Tensor Kernels | Analyzing deep neural networks (DNNs) via information plane (IP) theory has
gained tremendous attention recently as a tool to gain insight into, among
others, their generalization ability. However, it is by no means obvious how to
estimate mutual information (MI) between each hidden layer and the
input/desired output, to construct the IP. For instance, hidden layers with
many neurons require MI estimators with robustness towards the high
dimensionality associated with such layers. MI estimators should also be able
to naturally handle convolutional layers, while at the same time being
computationally tractable to scale to large networks. None of the existing IP
methods to date have been able to study truly deep Convolutional Neural
Networks (CNNs), such as the e.g.\ VGG-16. In this paper, we propose an IP
analysis using the new matrix--based R\'enyi's entropy coupled with tensor
kernels over convolutional layers, leveraging the power of kernel methods to
represent properties of the probability distribution independently of the
dimensionality of the data. The obtained results shed new light on the previous
literature concerning small-scale DNNs, however using a completely new
approach. Importantly, the new framework enables us to provide the first
comprehensive IP analysis of contemporary large-scale DNNs and CNNs,
investigating the different training phases and providing new insights into the
training dynamics of large-scale neural networks.
| stat |
Tempered Sigmoid Activations for Deep Learning with Differential Privacy | Because learning sometimes involves sensitive data, machine learning
algorithms have been extended to offer privacy for training data. In practice,
this has been mostly an afterthought, with privacy-preserving models obtained
by re-running training with a different optimizer, but using the model
architectures that already performed well in a non-privacy-preserving setting.
This approach leads to less than ideal privacy/utility tradeoffs, as we show
here. Instead, we propose that model architectures are chosen ab initio
explicitly for privacy-preserving training.
To provide guarantees under the gold standard of differential privacy, one
must bound as strictly as possible how individual training points can possibly
affect model updates. In this paper, we are the first to observe that the
choice of activation function is central to bounding the sensitivity of
privacy-preserving deep learning. We demonstrate analytically and
experimentally how a general family of bounded activation functions, the
tempered sigmoids, consistently outperform unbounded activation functions like
ReLU. Using this paradigm, we achieve new state-of-the-art accuracy on MNIST,
FashionMNIST, and CIFAR10 without any modification of the learning procedure
fundamentals or differential privacy analysis.
| stat |
Benefit of deep learning with non-convex noisy gradient descent:
Provable excess risk bound and superiority to kernel methods | Establishing a theoretical analysis that explains why deep learning can
outperform shallow learning such as kernel methods is one of the biggest issues
in the deep learning literature. Towards answering this question, we evaluate
excess risk of a deep learning estimator trained by a noisy gradient descent
with ridge regularization on a mildly overparameterized neural network, and
discuss its superiority to a class of linear estimators that includes neural
tangent kernel approach, random feature model, other kernel methods, $k$-NN
estimator and so on. We consider a teacher-student regression model, and
eventually show that any linear estimator can be outperformed by deep learning
in a sense of the minimax optimal rate especially for a high dimension setting.
The obtained excess bounds are so-called fast learning rate which is faster
than $O(1/\sqrt{n})$ that is obtained by usual Rademacher complexity analysis.
This discrepancy is induced by the non-convex geometry of the model and the
noisy gradient descent used for neural network training provably reaches a near
global optimal solution even though the loss landscape is highly non-convex.
Although the noisy gradient descent does not employ any explicit or implicit
sparsity inducing regularization, it shows a preferable generalization
performance that dominates linear estimators.
| stat |
Optimal dose calibration in radiotherapy | In this paper, the tools provided by the theory of Optimal Experimental
Design are applied to a nonlinear calibration model. This is motivated by the
need of estimating radiation doses using radiochromic films for radiotherapy
purposes. The calibration model is in this case nonlinear and the explanatory
variable cannot be worked out explicitly from the model. In this case an
experimental design has to be found on the dependent variable. For that, the
inverse function theorem will be used to obtain an information matrix to be
optimized. Optimal designs on the response variable are computed from two
different perspectives, first for fitting the model and estimating each of the
parameters and then for predicting the proper dose to be applied to the
patient. While the first is a common point of view in a general context of the
Optimal Experimental Design, the latter is actually the main objective of the
calibration problem for the practitioners and algorithms for computing these
optimal designs are also provided.
| stat |
On the Interplay of Regional Mobility, Social Connectedness, and the
Spread of COVID-19 in Germany | Since the primary mode of respiratory virus transmission is person-to-person
interaction, we are required to reconsider physical interaction patterns to
mitigate the number of people infected with COVID-19. Since non-pharmaceutical
interventions (NPI) had an evident impact on national mobility patterns, we
investigate the relative regional mobility behaviour to assess the effect of
human movement on the spread of COVID-19. In this paper, we, therefore, explore
the impact of human mobility and social connectivity derived from Facebook
activities on the weekly rate of new infections in Germany between March 3rd
and June 22nd, 2020. Our results confirm that reduced social activity lowers
the infection rate, accounting for regional and temporal patterns. The extent
of social distancing, quantified by the percentage of people staying put within
a federal administrative district, has an overall negative effect on the
incidence of infections. Additionally, our results show spatial infection
patterns based on geographic as well as social distances.
| stat |
A Bayesian decision-theoretic approach to incorporate preclinical
information into phase I oncology trials | Leveraging preclinical animal data for a phase I first-in-man trial is
appealing yet challenging. A prior based on animal data may place large
probability mass on values of the dose-toxicity model parameter(s), which
appear infeasible in light of data accrued from the ongoing phase I clinical
trial. In this paper, we seek to use animal data to improve decision making in
a model-based dose-escalation procedure for phase I oncology trials.
Specifically, animal data are incorporated via a robust mixture prior for the
parameters of the dose-toxicity relationship. This prior changes dynamically as
the trial progresses. After completion of treatment for each cohort, the weight
allocated to the informative component, obtained based on animal data alone, is
updated using a decision-theoretic approach to assess the commensurability of
the animal data with the human toxicity data observed thus far. In particular,
we measure commensurability as a function of the utility of optimal prior
predictions for the human responses (toxicity or no toxicity) on each
administered dose. The proposed methodology is illustrated through several
examples and an extensive simulation study. Results show that our proposal can
address difficulties in coping with prior-data conflict commencing in
sequential trials with a small sample size.
| stat |
Magnify Your Population: Statistical Downscaling to Augment the Spatial
Resolution of Socioeconomic Census Data | Fine resolution estimates of demographic and socioeconomic attributes are
crucial for planning and policy development. While several efforts have been
made to produce fine-scale gridded population estimates, socioeconomic features
are typically not available at scales finer than Census units, which may hide
local heterogeneity and disparity. In this paper we present a new statistical
downscaling approach to derive fine-scale estimates of key socioeconomic
attributes. The method leverages demographic and geographical extensive
covariates available at multiple scales and additional Census covariates only
available at coarse resolution, which are included in the model hierarchically
within a "forward learning" approach. For each selected socioeconomic variable,
a Random Forest model is trained on the source Census units and then used to
generate fine-scale gridded predictions, which are then adjusted to ensure the
best possible consistency with the coarser Census data. As a case study, we
apply this method to Census data in the United States, downscaling the selected
socioeconomic variables available at the block group level, to a grid of ~300
spatial resolution. The accuracy of the method is assessed at both spatial
scales, first computing a pseudo cross-validation coefficient of determination
for the predictions at the block group level and then, for extensive variables
only, also for the (unadjusted) predicted counts summed by block group. Based
on these scores and on the inspection of the downscaled maps, we conclude that
our method is able to provide accurate, smoother, and more detailed
socioeconomic estimates than the available Census data.
| stat |
Path Sample-Analytic Gradient Estimators for Stochastic Binary Networks | In neural networks with binary activations and or binary weights the training
by gradient descent is complicated as the model has piecewise constant
response. We consider stochastic binary networks, obtained by adding noises in
front of activations. The expected model response becomes a smooth function of
parameters, its gradient is well defined but it is challenging to estimate it
accurately. We propose a new method for this estimation problem combining
sampling and analytic approximation steps. The method has a significantly
reduced variance at the price of a small bias which gives a very practical
tradeoff in comparison with existing unbiased and biased estimators. We further
show that one extra linearization step leads to a deep straight-through
estimator previously known only as an ad-hoc heuristic. We experimentally show
higher accuracy in gradient estimation and demonstrate a more stable and better
performing training in deep convolutional models with both proposed methods.
| stat |
A Bayesian Model of Cash Bail Decisions | The use of cash bail as a mechanism for detaining defendants pre-trial is an
often-criticized system that many have argued violates the presumption of
"innocent until proven guilty." Many studies have sought to understand both the
long-term effects of cash bail's use and the disparate rate of cash bail
assignments along demographic lines (race, gender, etc). However, such work is
often susceptible to problems of infra-marginality -- that the data we observe
can only describe average outcomes, and not the outcomes associated with the
marginal decision.
In this work, we address this problem by creating a hierarchical Bayesian
model of cash bail assignments. Specifically, our approach models cash bail
decisions as a probabilistic process whereby judges balance the relative costs
of assigning cash bail with the cost of defendants potentially skipping court
dates, and where these skip probabilities are estimated based upon features of
the individual case. We then use Monte Carlo inference to sample the
distribution over these costs for different magistrates and across different
races. We fit this model to a data set we have collected of over 50,000 court
cases in the Allegheny and Philadelphia counties in Pennsylvania. Our analysis
of 50 separate judges shows that they are uniformly more likely to assign cash
bail to black defendants than to white defendants, even given identical
likelihood of skipping a court appearance. This analysis raises further
questions about the equity of the practice of cash bail, irrespective of its
underlying legal justification.
| stat |
Supervised Visualization for Data Exploration | Dimensionality reduction is often used as an initial step in data
exploration, either as preprocessing for classification or regression or for
visualization. Most dimensionality reduction techniques to date are
unsupervised; they do not take class labels into account (e.g., PCA, MDS,
t-SNE, Isomap). Such methods require large amounts of data and are often
sensitive to noise that may obfuscate important patterns in the data. Various
attempts at supervised dimensionality reduction methods that take into account
auxiliary annotations (e.g., class labels) have been successfully implemented
with goals of increased classification accuracy or improved data visualization.
Many of these supervised techniques incorporate labels in the loss function in
the form of similarity or dissimilarity matrices, thereby creating
over-emphasized separation between class clusters, which does not realistically
represent the local and global relationships in the data. In addition, these
approaches are often sensitive to parameter tuning, which may be difficult to
configure without an explicit quantitative notion of visual superiority. In
this paper, we describe a novel supervised visualization technique based on
random forest proximities and diffusion-based dimensionality reduction. We
show, both qualitatively and quantitatively, the advantages of our approach in
retaining local and global structures in data, while emphasizing important
variables in the low-dimensional embedding. Importantly, our approach is robust
to noise and parameter tuning, thus making it simple to use while producing
reliable visualizations for data exploration.
| stat |
Sparse Popularity Adjusted Stochastic Block Model | In the present paper we study a sparse stochastic network enabled with a
block structure. The popular Stochastic Block Model (SBM) and the Degree
Corrected Block Model (DCBM) address sparsity by placing an upper bound on the
maximum probability of connections between any pair of nodes. As a result,
sparsity describes only the behavior of network as a whole, without
distinguishing between the block-dependent sparsity patterns. To the best of
our knowledge, the recently introduced Popularity Adjusted Block Model (PABM)
is the only block model that allows to introduce a {\it structural sparsity}
where some probabilities of connections are identically equal to zero while the
rest of them remain above a certain threshold. The latter presents a more
nuanced view of the network.
| stat |
Efficient inference for stochastic differential equation mixed-effects
models using correlated particle pseudo-marginal algorithms | Stochastic differential equation mixed-effects models (SDEMEMs) are flexible
hierarchical models that are able to account for random variability inherent in
the underlying time-dynamics, as well as the variability between experimental
units and, optionally, account for measurement error. Fully Bayesian inference
for state-space SDEMEMs is performed, using data at discrete times that may be
incomplete and subject to measurement error. However, the inference problem is
complicated by the typical intractability of the observed data likelihood which
motivates the use of sampling-based approaches such as Markov chain Monte
Carlo. A Gibbs sampler is proposed to target the marginal posterior of all
parameter values of interest. The algorithm is made computationally efficient
through careful use of blocking strategies and correlated pseudo-marginal
Metropolis-Hastings steps within the Gibbs scheme. The resulting methodology is
flexible and is able to deal with a large class of SDEMEMs. The methodology is
demonstrated on three case studies, including tumor growth dynamics and
neuronal data. The gains in terms of increased computational efficiency are
model and data dependent, but unless bespoke sampling strategies requiring
analytical derivations are possible for a given model, we generally observe an
efficiency increase of one order of magnitude when using correlated particle
methods together with our blocked-Gibbs strategy.
| stat |
Modelling the clustering of extreme events for short-term risk
assessment | Having reliable estimates of the occurrence rates of extreme events is highly
important for insurance companies, government agencies and the general public.
The rarity of an extreme event is typically expressed through its return
period, i.e., the expected waiting time between events of the observed size if
the extreme events of the processes are independent and identically
distributed. A major limitation with this measure is when an unexpectedly high
number of events occur within the next few months immediately after a
\textit{T} year event, with \textit{T} large. Such events undermine the trust
in the quality of these risk estimates. The clustering of apparently
independent extreme events can occur as a result of local non-stationarity of
the process, which can be explained by covariates or random effects. We show
how accounting for these covariates and random effects provides more accurate
estimates of return levels and aids short-term risk assessment through the use
of a new risk measure, which provides evidence of risk which is complementary
to the return period.
| stat |
High-dimensional vector autoregressive time series modeling via tensor
decomposition | The classical vector autoregressive model is a fundamental tool for
multivariate time series analysis. However, it involves too many parameters
when the number of time series and lag order are even moderately large. This
paper proposes to rearrange the transition matrices of the model into a tensor
form such that the parameter space can be restricted along three directions
simultaneously via tensor decomposition. In contrast, the reduced-rank
regression method can restrict the parameter space in only one direction.
Besides achieving substantial dimension reduction, the proposed model is
interpretable from the factor modeling perspective. Moreover, to handle
high-dimensional time series, this paper considers imposing sparsity on factor
matrices to improve the model interpretability and estimation efficiency, which
leads to a sparsity-inducing estimator. For the low-dimensional case, we derive
asymptotic properties of the proposed least squares estimator and introduce an
alternating least squares algorithm. For the high-dimensional case, we
establish non-asymptotic properties of the sparsity-inducing estimator and
propose an ADMM algorithm for regularized estimation. Simulation experiments
and a real data example demonstrate the advantages of the proposed approach
over various existing methods.
| stat |
Neural Network Gaussian Process Considering Input Uncertainty for
Composite Structures Assembly | Developing machine learning enabled smart manufacturing is promising for
composite structures assembly process. To improve production quality and
efficiency of the assembly process, accurate predictive analysis on dimensional
deviations and residual stress of the composite structures is required. The
novel composite structures assembly involves two challenges: (i) the highly
nonlinear and anisotropic properties of composite materials; and (ii)
inevitable uncertainty in the assembly process. To overcome those problems, we
propose a neural network Gaussian process model considering input uncertainty
for composite structures assembly. Deep architecture of our model allows us to
approximate a complex process better, and consideration of input uncertainty
enables robust modeling with complete incorporation of the process uncertainty.
Based on simulation and case study, the NNGPIU can outperform other benchmark
methods when the response function is nonsmooth and nonlinear. Although we use
composite structure assembly as an example, the proposed methodology can be
applicable to other engineering systems with intrinsic uncertainties.
| stat |
Rejoinder for the discussion of the paper "A novel algorithmic approach
to Bayesian Logic Regression" | In this rejoinder we summarize the comments, questions and remarks on the
paper "A novel algorithmic approach to Bayesian Logic Regression" from the
discussants. We then respond to those comments, questions and remarks, provide
several extensions of the original model and give a tutorial on our R-package
EMJMCMC (http://aliaksah.github.io/EMJMCMC2016/)
| stat |
Kendall's Tau for Functional Data Analysis | We treat the problem of testing for association between a functional variable
belonging to Hilbert space and a scalar variable. Particularly, we propose a
distribution-free test statistic based on Kendall's Tau which is one of the
most popular methods to determine the association between two random variables.
The distribution of the test statistic under the null hypothesis of
independence is established using the theory of U-statistics taking values in a
Hilbert space. We also consider the case when the functional data is sparsely
observed, a situation that arises in many applications. Simulations show that
the proposed method outperforms the alternatives under several conditions
demonstrating the robustness of our approach. We provide data applications that
further consolidate the utility of our method.
| stat |
Indicators of retention in remote digital health studies: A cross-study
evaluation of 100,000 participants | Digital technologies such as smartphones are transforming the way scientists
conduct biomedical research using real-world data. Several remotely-conducted
studies have recruited thousands of participants over a span of a few months.
Unfortunately, these studies are hampered by substantial participant attrition,
calling into question the representativeness of the collected data including
generalizability of findings from these studies. We report the challenges in
retention and recruitment in eight remote digital health studies comprising
over 100,000 participants who participated for more than 850,000 days,
completing close to 3.5 million remote health evaluations. Survival modeling
surfaced several factors significantly associated(P < 1e-16) with increase in
median retention time i) Clinician referral(increase of 40 days), ii) Effect of
compensation (22 days), iii) Clinical conditions of interest to the study (7
days) and iv) Older adults(4 days). Additionally, four distinct patterns of
daily app usage behavior that were also associated(P < 1e-10) with participant
demographics were identified. Most studies were not able to recruit a
representative sample, either demographically or regionally. Combined together
these findings can help inform recruitment and retention strategies to enable
equitable participation of populations in future digital health research.
| stat |
Testing for Normality with Neural Networks | In this paper, we treat the problem of testing for normality as a binary
classification problem and construct a feedforward neural network that can
successfully detect normal distributions by inspecting small samples from them.
The numerical experiments conducted on small samples with no more than 100
elements indicated that the neural network which we trained was more accurate
and far more powerful than the most frequently used and most powerful standard
tests of normality: Shapiro-Wilk, Anderson-Darling, Lilliefors and
Jarque-Berra, as well as the kernel tests of goodness-of-fit. The neural
network had the AUROC score of almost 1, which corresponds to the perfect
binary classifier. Additionally, the network's accuracy was higher than 96% on
a set of larger samples with 250-1000 elements. Since the normality of data is
an assumption of numerous techniques for analysis and inference, the neural
network constructed in this study has a very high potential for use in everyday
practice of statistics, data analysis and machine learning in both science and
industry.
| stat |
Semi-orthogonal Non-negative Matrix Factorization with an Application in
Text Mining | Emergency Department (ED) crowding is a worldwide issue that affects the
efficiency of hospital management and the quality of patient care. This occurs
when the request for an admit ward-bed to receive a patient is delayed until an
admission decision is made by a doctor. To reduce the overcrowding and waiting
time of ED, we build a classifier to predict the disposition of patients using
manually-typed nurse notes collected during triage, thereby allowing hospital
staff to begin necessary preparation beforehand. However, these triage notes
involve high dimensional, noisy, and also sparse text data which makes model
fitting and interpretation difficult. To address this issue, we propose the
semi-orthogonal non-negative matrix factorization (SONMF) for both continuous
and binary design matrices to first bi-cluster the patients and words into a
reduced number of topics. The subjects can then be interpreted as a
non-subtractive linear combination of orthogonal basis topic vectors. These
generated topic vectors provide the hospital with a direct understanding of the
cause of admission. We show that by using a transformation of basis, the
classification accuracy can be further increased compared to the conventional
bag-of-words model and alternative matrix factorization approaches. Through
simulated data experiments, we also demonstrate that the proposed method
outperforms other non-negative matrix factorization (NMF) methods in terms of
factorization accuracy, rate of convergence, and degree of orthogonality.
| stat |
Estimating Average Treatment Effects Utilizing Fractional Imputation
when Confounders are Subject to Missingness | The problem of missingness in observational data is ubiquitous. When the
confounders are missing at random, multiple imputation is commonly used;
however, the method requires congeniality conditions for valid inferences,
which may not be satisfied when estimating average causal treatment effects.
Alternatively, fractional imputation, proposed by Kim 2011, has been
implemented to handling missing values in regression context. In this article,
we develop fractional imputation methods for estimating the average treatment
effects with confounders missing at random. We show that the fractional
imputation estimator of the average treatment effect is asymptotically normal,
which permits a consistent variance estimate. Via simulation study, we compare
fractional imputation's accuracy and precision with that of multiple
imputation.
| stat |
Bayesian Shape Invariant Model for Latent Growth Curve with
Time-Invariant Covariates | In the attention-deficit hyperactivity disorder (ADHD) study, children are
prescribed different stimulant medications. The height measurements are
recorded longitudinally along with the medication time. Differences among the
patients are captured by the parameters suggested the Superimposition by
Translation and Rotation (SITAR) model using three subject-specific parameters
to estimate their deviation from the mean growth curve. In this paper, we
generalize the SITAR model in a Bayesian way with time-invariant covariates.
The time-invariant model allows us to predict latent growth factors. Since
patients suffer from a common disease, they usually exhibit a similar pattern,
and it is natural to build a nonlinear model that is shaped invariant. The
model is semi-parametric, where the population time curve is modeled with a
natural cubic spline. The original shape invariant growth curve model,
motivated by epidemiological research on the evolution of pubertal heights over
time, fits the underlying shape function for height over age and estimates
subject-specific deviations from this curve in terms of size, tempo, and
velocity using maximum likelihood. The usefulness of the model is illustrated
in the attention deficit hyperactivity disorder (ADHD) study. Further, we
demonstrated the effect of stimulant medications on pubertal growth by gender.
| stat |
Space-time smoothing models for sub-national measles routine
immunization coverage estimation with complex survey data | Despite substantial advances in global measles vaccination, measles disease
burden remains high in many low- and middle-income countries. A key public
health strategy for controling measles in such high-burden settings is to
conduct supplementary immunization activities (SIAs) in the form of mass
vaccination campaigns, in addition to delivering scheduled vaccination through
routine immunization (RI) programs. To achieve balanced implementations of RI
and SIAs, robust measurement of sub-national RI-specific coverage is crucial.
In this paper, we develop a space-time smoothing model for estimating
RI-specific coverage of the first dose of measles-containing-vaccines (MCV1) at
sub-national level using complex survey data. The application that motivated
this work is estimation of the RI-specific MCV1 coverage in Nigeria's 36 states
and the Federal Capital Territory. Data come from four Demographic and Health
Surveys, three Multiple Indicator Cluster Surveys, and two National Nutrition
and Health Surveys conducted in Nigeria between 2003 and 2018. Our method
incorporates information from the SIA calendar published by the World Health
Organization and accounts for the impact of SIAs on the overall MCV1 coverage,
as measured by cross-sectional surveys. The model can be used to analyze data
from multiple surveys with different data collection schemes and construct
coverage estimates with uncertainty that reflects the various sampling designs.
Implementation of our method can be done efficiently using integrated nested
Laplace approximation (INLA).
| stat |
University rankings from the revealed preferences of the applicants | A methodology is presented to rank universities on the basis of the lists of
programmes the students applied for. We exploit a crucial feature of the
centralised assignment system to higher education in Hungary: a student is
admitted to the first programme where the score limit is achieved. This makes
it possible to derive a partial preference order of each applicant. Our
approach integrates the information from all students participating in the
system, is free of multicollinearity among the indicators, and contains few ad
hoc parameters. The procedure is implemented to rank faculties in the Hungarian
higher education between 2001 and 2016. We demonstrate that the ranking given
by the least squares method has favourable theoretical properties, is robust
with respect to the aggregation of preferences, and performs well in practice.
The suggested ranking is worth considering as a reasonable alternative to the
standard composite indices.
| stat |
A Study on the Association between Maternal Childhood Trauma Exposure
and Placental-fetal Stress Physiology during Pregnancy | It has been found that the effect of childhood trauma (CT) exposure may pass
on to the next generation. Scientists have hypothesized that the association
between CT exposure and placental-fetal stress physiology is the mechanism. A
study was conducted to examine the hypothesis. To examine the association
between CT exposure and placental corticotrophin-releasing hormone (pCRH),
linear mixed effect model and hierarchical Bayesian linear model were
constructed. In Bayesian inference, by providing conditionally conjugate
priors, Gibbs sampler was used to draw MCMC samples. Piecewise linear mixed
effect model was conducted in order to adjust to the dramatic change of pCRH at
around week 20 into pregnancy. Pearson residual, QQ, ACF and trace plots were
used to justify the model adequacy. Likelihood ratio test and DIC were utilized
to model selection. The association between CT exposure and pCRH during
pregnancy is obvious. The effect of CT exposure on pCRH varies dramatically
over gestational age. Women with one childhood trauma would experience 11.9%
higher in pCRH towards the end of pregnancy than those without childhood
trauma. The increase rate of pCRH after week 20 is almost four-fold larger than
that before week 20. Frequentist and Bayesian inference produce similar
results. The findings support the hypothesis that the effect of CT exposure on
pCRH over GA exists. The effect changes dramatically at around week 20 into
pregnancy.
| stat |
Deep learning for detecting bid rigging: Flagging cartel participants
based on convolutional neural networks | Adding to the literature on the data-driven detection of bid-rigging cartels,
we propose a novel approach based on deep learning (a subfield of artificial
intelligence) that flags cartel participants based on their pairwise bidding
interactions with other firms. More concisely, we combine a so-called
convolutional neural network for image recognition with graphs that in a
pairwise manner plot the normalized bid values of some reference firm against
the normalized bids of any other firms participating in the same tenders as the
reference firm. Based on Japanese and Swiss procurement data, we construct such
graphs for both collusive and competitive episodes (i.e when a bid-rigging
cartel is or is not active) and use a subset of graphs to train the neural
network such that it learns distinguishing collusive from competitive bidding
patterns. We use the remaining graphs to test the neural network's
out-of-sample performance in correctly classifying collusive and competitive
bidding interactions. We obtain a very decent average accuracy of around 90% or
slightly higher when either applying the method within Japanese, Swiss, or
mixed data (in which Swiss and Japanese graphs are pooled). When using data
from one country for training to test the trained model's performance in the
other country (i.e. transnationally), predictive performance decreases (likely
due to institutional differences in procurement procedures across countries),
but often remains satisfactorily high. All in all, the generally quite high
accuracy of the convolutional neural network despite being trained in a rather
small sample of a few 100 graphs points to a large potential of deep learning
approaches for flagging and fighting bid-rigging cartels.
| stat |
Stein Variational Gradient Descent: many-particle and long-time
asymptotics | Stein variational gradient descent (SVGD) refers to a class of methods for
Bayesian inference based on interacting particle systems. In this paper, we
consider the originally proposed deterministic dynamics as well as a stochastic
variant, each of which represent one of the two main paradigms in Bayesian
computational statistics: variational inference and Markov chain Monte Carlo.
As it turns out, these are tightly linked through a correspondence between
gradient flow structures and large-deviation principles rooted in statistical
physics. To expose this relationship, we develop the cotangent space
construction for the Stein geometry, prove its basic properties, and determine
the large-deviation functional governing the many-particle limit for the
empirical measure. Moreover, we identify the Stein-Fisher information (or
kernelised Stein discrepancy) as its leading order contribution in the
long-time and many-particle regime in the sense of $\Gamma$-convergence,
shedding some light on the finite-particle properties of SVGD. Finally, we
establish a comparison principle between the Stein-Fisher information and
RKHS-norms that might be of independent interest.
| stat |
Finding our Way in the Dark: Approximate MCMC for Approximate Bayesian
Methods | With larger data at their disposal, scientists are emboldened to tackle
complex questions that require sophisticated statistical models. It is not
unusual for the latter to have likelihood functions that elude analytical
formulations. Even under such adversity, when one can simulate from the
sampling distribution, Bayesian analysis can be conducted using approximate
methods such as Approximate Bayesian Computation (ABC) or Bayesian Synthetic
Likelihood (BSL). A significant drawback of these methods is that the number of
required simulations can be prohibitively large, thus severely limiting their
scope. In this paper we design perturbed MCMC samplers that can be used within
the ABC and BSL paradigms to significantly accelerate computation while
maintaining control on computational efficiency. The proposed strategy relies
on recycling samples from the chain's past. The algorithmic design is supported
by a theoretical analysis while practical performance is examined via a series
of simulation examples and data analyses.
| stat |
Causal Inference on Networks under Continuous Treatment Interference | This paper presents a methodology to draw causal inference in a
non-experimental setting subject to network interference. Specifically, we
develop a generalized propensity score-based estimator that allows us to
estimate both direct and spillover effects of a continuous treatment, which
spreads through weighted and directed edges of a network. To showcase this
methodology, we investigate whether and how spillover effects shape the optimal
level of policy interventions in agricultural markets. Our results show that,
in this context, neglecting interference may lead to a downward bias when
assessing policy effectiveness.
| stat |
Bayesian Probabilistic Numerical Integration with Tree-Based Models | Bayesian quadrature (BQ) is a method for solving numerical integration
problems in a Bayesian manner, which allows users to quantify their uncertainty
about the solution. The standard approach to BQ is based on a Gaussian process
(GP) approximation of the integrand. As a result, BQ is inherently limited to
cases where GP approximations can be done in an efficient manner, thus often
prohibiting very high-dimensional or non-smooth target functions. This paper
proposes to tackle this issue with a new Bayesian numerical integration
algorithm based on Bayesian Additive Regression Trees (BART) priors, which we
call BART-Int. BART priors are easy to tune and well-suited for discontinuous
functions. We demonstrate that they also lend themselves naturally to a
sequential design setting and that explicit convergence rates can be obtained
in a variety of settings. The advantages and disadvantages of this new
methodology are highlighted on a set of benchmark tests including the Genz
functions, and on a Bayesian survey design problem.
| stat |
Multinomial Models with Linear Inequality Constraints: Overview and
Improvements of Computational Methods for Bayesian Inference | Many psychological theories can be operationalized as linear inequality
constraints on the parameters of multinomial distributions (e.g., discrete
choice analysis). These constraints can be described in two equivalent ways:
Either as the solution set to a system of linear inequalities or as the convex
hull of a set of extremal points (vertices). For both representations, we
describe a general Gibbs sampler for drawing posterior samples in order to
carry out Bayesian analyses. We also summarize alternative sampling methods for
estimating Bayes factors for these model representations using the encompassing
Bayes factor method. We introduce the R package multinomineq, which provides an
easily-accessible interface to a computationally efficient implementation of
these techniques.
| stat |
Kernel-based Prediction of Non-Markovian Time Series | A nonparametric method to predict non-Markovian time series of partially
observed dynamics is developed. The prediction problem we consider is a
supervised learning task of finding a regression function that takes a delay
embedded observable to the observable at a future time. When delay embedding
theory is applicable, the proposed regression function is a consistent
estimator of the flow map induced by the delay embedding. Furthermore, the
corresponding Mori-Zwanzig equation governing the evolution of the observable
simplifies to only a Markovian term, represented by the regression function. We
realize this supervised learning task with a class of kernel-based linear
estimators, the kernel analog forecast (KAF), which are consistent in the limit
of large data. In a scenario with a high-dimensional covariate space, we employ
a Markovian kernel smoothing method which is computationally cheaper than the
Nystr\"om projection method for realizing KAF. In addition to the guaranteed
theoretical convergence, we numerically demonstrate the effectiveness of this
approach on higher-dimensional problems where the relevant kernel features are
difficult to capture with the Nystr\"om method. Given noisy training data, we
propose a nonparametric smoother as a de-noising method. Numerically, we show
that the proposed smoother is more accurate than EnKF and 4Dvar in de-noising
signals corrupted by independent (but not necessarily identically distributed)
noise, even if the smoother is constructed using a data set corrupted by white
noise. We show skillful prediction using the KAF constructed from the denoised
data.
| stat |
A Symmetric Loss Perspective of Reliable Machine Learning | When minimizing the empirical risk in binary classification, it is a common
practice to replace the zero-one loss with a surrogate loss to make the
learning objective feasible to optimize. Examples of well-known surrogate
losses for binary classification include the logistic loss, hinge loss, and
sigmoid loss. It is known that the choice of a surrogate loss can highly
influence the performance of the trained classifier and therefore it should be
carefully chosen. Recently, surrogate losses that satisfy a certain symmetric
condition (aka., symmetric losses) have demonstrated their usefulness in
learning from corrupted labels. In this article, we provide an overview of
symmetric losses and their applications. First, we review how a symmetric loss
can yield robust classification from corrupted labels in balanced error rate
(BER) minimization and area under the receiver operating characteristic curve
(AUC) maximization. Then, we demonstrate how the robust AUC maximization method
can benefit natural language processing in the problem where we want to learn
only from relevant keywords and unlabeled documents. Finally, we conclude this
article by discussing future directions, including potential applications of
symmetric losses for reliable machine learning and the design of non-symmetric
losses that can benefit from the symmetric condition.
| stat |
Regression of exchangeable relational arrays | Relational arrays represent measures of association between pairs of actors,
often in varied contexts or over time. Such data appear as trade flows between
countries, financial transactions between individuals, contact frequencies
between school children in classrooms, and dynamic protein-protein
interactions. Elements of a relational array are often modeled as a linear
function of observable covariates, where the regression coefficients are the
subjects of inference. The structure of the relational array engenders
dependence among relations that involve the same actor. Uncertainty estimates
for regression coefficient estimators -- and ideally the coefficient estimators
themselves -- must account for this relational dependence. Existing estimators
of standard errors that recognize such relational dependence rely on estimating
extremely complex, heterogeneous structure across actors. This paper proposes a
new class of parsimonious coefficient and standard error estimators for
regressions of relational arrays. We leverage an exchangeability assumption to
derive standard error estimators that pool information across actors and are
substantially more accurate than existing estimators in a variety of settings.
This exchangeability assumption is pervasive in network and array models in the
statistics literature, but not previously considered when adjusting for
dependence in a regression setting with relational data. We demonstrate
improvements in inference theoretically, via a simulation study, and by
analysis of a data set involving international trade.
| stat |
A general Bayesian bootstrap for censored data based on the beta-Stacy
process | We introduce a novel procedure to perform Bayesian non-parametric inference
with right-censored data, the \emph{beta-Stacy bootstrap}. This approximates
the posterior law of summaries of the survival distribution (e.g. the mean
survival time), which is often difficult in the non-parametric case. More
precisely, our procedure approximates the joint posterior law of functionals of
the beta-Stacy process, a non-parametric process prior widely used in survival
analysis. It also represents the missing link that unifies other common
Bayesian bootstraps for complete or censored data based on non-parametric
priors. It is defined by an exact sampling algorithm that does not require
tuning of Markov Chain Monte Carlo steps. We illustrate the beta-Stacy
bootstrap by analyzing survival data from a real clinical trial.
| stat |
A Spatial Bayesian Semiparametric Mixture Model for Positive Definite
Matrices with Applications to Diffusion Tensor Imaging | Diffusion tensor imaging (DTI) is a popular magnetic resonance imaging
technique used to characterize microstructural changes in the brain. DTI
studies quantify the diffusion of water molecules in a voxel using an estimated
3x3 symmetric positive definite diffusion tensor matrix. Statistical analysis
of DTI data is challenging because the data are positive definite matrices.
Matrix-variate information is often summarized by a univariate quantity, such
as the fractional anisotropy (FA), leading to a loss of information.
Furthermore, DTI analyses often ignore the spatial association of neighboring
voxels, which can lead to imprecise estimates. Although the spatial modeling
literature is abundant, modeling spatially dependent positive definite matrices
is challenging. To mitigate these issues, we propose a matrix-variate Bayesian
semiparametric mixture model, where the positive definite matrices are
distributed as a mixture of inverse Wishart distributions with the spatial
dependence captured by a Markov model for the mixture component labels.
Conjugacy and the double Metropolis-Hastings algorithm result in fast and
elegant Bayesian computing. Our simulation study shows that the proposed method
is more powerful than non-spatial methods. We also apply the proposed method to
investigate the effect of cocaine use on brain structure. The contribution of
our work is to provide a novel statistical inference tool for DTI analysis by
extending spatial statistics to matrix-variate data.
| stat |
Correct power for cluster-randomized difference-in-difference trials
with loss to follow-up | Cluster randomized trials with measurements at baseline can improve power
over post-test only designs by using difference in difference designs. However,
subjects may be lost to follow-up between the baseline and follow-up periods.
While equations for sample size and variance have been developed assuming no
loss to follow-up ("cohort") and completely different subjects at baseline and
follow-up ("cross-sectional") difference in difference designs, equations have
yet to be developed when some subjects are observed in both periods ("mixture"
designs). We present a general equation for calculating the variance in
difference in difference designs and derive special cases assuming loss to
follow-up with replacement of lost subjects and assuming loss to follow-up with
no replacement but retaining the baseline measurements of all subjects.
Relative efficiency plots, plots of variance against subject autocorrelation,
and plots of variance by follow-up rate and subject autocorrelation are used to
compare cohort, cross-sectional, and mixture approaches. Results indicate that
when loss to follow-up to uncommon, mixture designs are almost as efficient as
cohort designs with a given initial sample size. When loss to follow-up is
common, mixture designs with full replacement maintain efficiency relative to
cohort designs. Finally, our results provide guidance on whether to replace
lost subjects during trial design and analysis.
| stat |
Optimal subsampling for quantile regression in big data | We investigate optimal subsampling for quantile regression. We derive the
asymptotic distribution of a general subsampling estimator and then derive two
versions of optimal subsampling probabilities. One version minimizes the trace
of the asymptotic variance-covariance matrix for a linearly transformed
parameter estimator and the other minimizes that of the original parameter
estimator. The former does not depend on the densities of the responses given
covariates and is easy to implement. Algorithms based on optimal subsampling
probabilities are proposed and asymptotic distributions and asymptotic
optimality of the resulting estimators are established. Furthermore, we propose
an iterative subsampling procedure based on the optimal subsampling
probabilities in the linearly transformed parameter estimation which has great
scalability to utilize available computational resources. In addition, this
procedure yields standard errors for parameter estimators without estimating
the densities of the responses given the covariates. We provide numerical
examples based on both simulated and real data to illustrate the proposed
method.
| stat |
A principled distance-based prior for the shape of the Weibull model | The use of flat or weakly informative priors is popular due to the objective
a priori belief in the absence of strong prior information. In the case of the
Weibull model the improper uniform, equal parameter gamma and joint Jeffrey's
priors for the shape parameter are popular choices. The effects and behaviors
of these priors have yet to be established from a modeling viewpoint,
especially their ability to reduce to the simpler exponential model. In this
work we propose a new principled prior for the shape parameter of the Weibull
model, originating from a prior on the distance function, and advocate this new
prior as a principled choice in the absence of strong prior information. This
new prior can then be used in models with a Weibull modeling component, like
competing risks, joint and spatial models, to mention a few. This prior is
available in the R-INLA for use, and is applied in a joint
longitudinal-survival model framework using the INLA method.
| stat |
Functional L-Optimality Subsampling for Massive Data | Massive data bring the big challenges of memory and computation to
researchers, which can be tackled to some extent by taking subsamples from the
full data as a surrogate. For functional data, it is common to collect
measurements intensively over their domains, which require more memory and
computation time when the sample size is large. The situation would be much
worse when the statistical inference is made through bootstrap samples. To the
best of our knowledge, there is no work to study the subsampling for the
functional linear regression or its generation systematically. In this article,
based on the functional L-optimality criterion we propose an optimal
subsampling method for the functional linear model. When the response is a
discrete or categorical variable, we further extend this subsampling method to
the functional generalized linear model. We establish the asymptotic properties
of the resultant estimators by the subsampling methods. The finite sample
performance of our proposed subsampling methods is investigated by extensive
simulation studies. We also apply our proposed subsampling methods to analyze
the global climate data and the kidney transplant data. The results from the
analysis of these data show that the optimal subsampling methods motivated by
the functional L-optimality criterion are much better than the uniform
subsampling method and can well approximate the results based on full data.
| stat |
Factor Analysis, Probabilistic Principal Component Analysis, Variational
Inference, and Variational Autoencoder: Tutorial and Survey | This is a tutorial and survey paper on factor analysis, probabilistic
Principal Component Analysis (PCA), variational inference, and Variational
Autoencoder (VAE). These methods, which are tightly related, are dimensionality
reduction and generative models. They asssume that every data point is
generated from or caused by a low-dimensional latent factor. By learning the
parameters of distribution of latent space, the corresponding low-dimensional
factors are found for the sake of dimensionality reduction. For their
stochastic and generative behaviour, these models can also be used for
generation of new data points in the data space. In this paper, we first start
with variational inference where we derive the Evidence Lower Bound (ELBO) and
Expectation Maximization (EM) for learning the parameters. Then, we introduce
factor analysis, derive its joint and marginal distributions, and work out its
EM steps. Probabilistic PCA is then explained, as a special case of factor
analysis, and its closed-form solutions are derived. Finally, VAE is explained
where the encoder, decoder and sampling from the latent space are introduced.
Training VAE using both EM and backpropagation are explained.
| stat |
Asymmetric tail dependence modeling, with application to cryptocurrency
market data | Since the inception of Bitcoin in 2008, cryptocurrencies have played an
increasing role in the world of e-commerce, but the recent turbulence in the
cryptocurrency market in 2018 has raised some concerns about their stability
and associated risks. For investors, it is crucial to uncover the dependence
relationships between cryptocurrencies for a more resilient portfolio
diversification. Moreover, the stochastic behavior in both tails is important,
as long positions are sensitive to a decrease in prices (lower tail), while
short positions are sensitive to an increase in prices (upper tail). In order
to assess both risk types, we develop in this paper a flexible copula model
which is able to distinctively capture asymptotic dependence or independence in
its lower and upper tails simultaneously. Our proposed model is parsimonious
and smoothly bridges (in each tail) both extremal dependence classes in the
interior of the parameter space. Inference is performed using a full or
censored likelihood approach, and we investigate by simulation the estimators'
efficiency under three different censoring schemes which reduce the impact of
non-extreme observations. We also develop a local likelihood approach to
capture the temporal dynamics of extremal dependence among two leading
cryptocurrencies. We here apply our model to historical closing prices of five
leading cryotocurrencies, which share most of the cryptocurrency market
capitalizations. The results show that our proposed copula model outperforms
alternative copula models and that the lower tail dependence level between most
pairs of leading cryptocurrencies -- and in particular Bitcoin and Ethereum --
has become stronger over time, smoothly transitioning from an asymptotic
independence regime to an asymptotic dependence regime in recent years, whilst
the upper tail has been relatively more stable overall at a weaker dependence
level.
| stat |
A Reproducing Kernel Hilbert Space log-rank test for the two-sample
problem | Weighted log-rank tests are arguably the most widely used tests by
practitioners for the two-sample problem in the context of right-censored data.
Many approaches have been considered to make weighted log-rank tests more
robust against a broader family of alternatives, among them, considering linear
combinations of weighted log-rank tests, and taking the maximum among a finite
collection of them. In this paper, we propose as test statistic the supremum of
a collection of (potentially infinite) weight-indexed log-rank tests where the
index space is the unit ball in a reproducing kernel Hilbert space (RKHS). By
using some desirable properties of RKHSs we provide an exact and simple
evaluation of the test statistic and establish connections with previous tests
in the literature. Additionally, we show that for a special family of RKHSs,
the proposed test is omnibus. We finalise by performing an empirical evaluation
of the proposed methodology and show an application to a real data scenario.
Our theoretical results are proved using techniques for double integrals with
respect to martingales that may be of independent interest.
| stat |
ALMA: Alternating Minimization Algorithm for Clustering Mixture
Multilayer Network | The paper considers a Mixture Multilayer Stochastic Block Model (MMLSBM),
where layers can be partitioned into groups of similar networks, and networks
in each group are equipped with a distinct Stochastic Block Model. The goal is
to partition the multilayer network into clusters of similar layers, and to
identify communities in those layers. Jing et al. (2020) introduced the MMLSBM
and developed a clustering methodology, TWIST, based on regularized tensor
decomposition.
The present paper proposes a different technique, an alternating minimization
algorithm (ALMA), that aims at simultaneous recovery of the layer partition,
together with estimation of the matrices of connection probabilities of the
distinct layers. Compared to TWIST, ALMA achieves higher accuracy both
theoretically and numerically.
| stat |
Bayesian Stacked Parametric Survival with Frailty Components and
Interval Censored Failure Times | To better understand effects of exposure to food allergens, food challenge
studies are designed to slowly increase the dose of an allergen delivered to
allergic individuals until an objective reaction occurs. These dose-to-failure
studies are used to determine acceptable intake levels and are analyzed using
parametric failure time models. Though these models can provide estimates of
the survival curve, their parametric form may misrepresent the survival
function for doses of interest, and different models that describe the data
similarly may produce different dose-to-failure estimates. Motivated by
predictive inference, we developed a Bayesian approach to combine survival
estimates based upon posterior predictive stacking, where the weights are
formed to maximize posterior predictive accuracy. The approach allows for the
inclusion of flexible models, and, in our case, allows us to include random
effects to account for frailty components entering the model through
study-to-study heterogeneity. The methodology is investigated in simulation,
and is used to estimate allergic population eliciting doses for multiple food
allergens.
| stat |
Cascade of Phase Transitions for Multi-Scale Clustering | We present a novel framework exploiting the cascade of phase transitions
occurring during a simulated annealing of the Expectation-Maximisation
algorithm to cluster datasets with multi-scale structures. Using the weighted
local covariance, we can extract, a posteriori and without any prior knowledge,
information on the number of clusters at different scales together with their
size. We also study the linear stability of the iterative scheme to derive the
threshold at which the first transition occurs and show how to approximate the
next ones. Finally, we combine simulated annealing together with recent
developments of regularised Gaussian mixture models to learn a principal graph
from spatially structured datasets that can also exhibit many scales.
| stat |
An Efficient Method for Computing Expected Value of Sample Information
for Survival Data from an Ongoing Trial | The European Medicines Agency has in recent years allowed licensing of new
pharmaceuticals at an earlier stage in the clinical trial process. When trial
evidence is obtained at an early stage, the events of interest, such as disease
progression or death, may have only been observed in a small proportion of
patients. Health care authorities therefore must decide on the adoption of new
technologies based on less mature evidence than previously, resulting in
greater uncertainty about clinical- and cost-effectiveness. When a trial is
ongoing at the point of decision making, there may be value in continuing the
trial in order to collect additional data before making an adoption decision.
This can be quantified by the Expected Value of Sample Information (EVSI).
However, no guidance exists on how to compute the EVSI for survival data from
an ongoing trial, nor on how to account for uncertainty about the choice of
survival model in the EVSI calculations. In this article, we describe
algorithms for computing the EVSI of extending a trial's follow-up, both where
a single known survival model is assumed, and where we are uncertain about the
true survival model. We compare a nested Markov Chain Monte Carlo procedure
with a non-parametric regression-based method in two synthetic case studies,
and find close agreement between the two methods. The regression-based method
is fast and straightforward to implement, and scales easily to include any
number of candidate survival models in the model uncertainty case. EVSI for
ongoing trials can help decision makers determine whether early patient access
to a new technology can be justified on the basis of the current evidence or
whether more mature evidence is needed.
| stat |
Finding Statistically Significant Interactions between Continuous
Features | The search for higher-order feature interactions that are statistically
significantly associated with a class variable is of high relevance in fields
such as Genetics or Healthcare, but the combinatorial explosion of the
candidate space makes this problem extremely challenging in terms of
computational efficiency and proper correction for multiple testing. While
recent progress has been made regarding this challenge for binary features, we
here present the first solution for continuous features. We propose an
algorithm which overcomes the combinatorial explosion of the search space of
higher-order interactions by deriving a lower bound on the p-value for each
interaction, which enables us to massively prune interactions that can never
reach significance and to thereby gain more statistical power. In our
experiments, our approach efficiently detects all significant interactions in a
variety of synthetic and real-world datasets.
| stat |
Learning bounded subsets of $L_p$ | We study learning problems in which the underlying class is a bounded subset
of $L_p$ and the target $Y$ belongs to $L_p$. Previously, minimax sample
complexity estimates were known under such boundedness assumptions only when
$p=\infty$. We present a sharp sample complexity estimate that holds for any $p
> 4$. It is based on a learning procedure that is suited for heavy-tailed
problems.
| stat |
PCA Reduced Gaussian Mixture Models with Applications in Superresolution | Despite the rapid development of computational hardware, the treatment of
large and high dimensional data sets is still a challenging problem. This paper
provides a twofold contribution to the topic. First, we propose a Gaussian
Mixture Model in conjunction with a reduction of the dimensionality of the data
in each component of the model by principal component analysis, called PCA-GMM.
To learn the (low dimensional) parameters of the mixture model we propose an EM
algorithm whose M-step requires the solution of constrained optimization
problems. Fortunately, these constrained problems do not depend on the usually
large number of samples and can be solved efficiently by an (inertial) proximal
alternating linearized minimization algorithm. Second, we apply our PCA-GMM for
the superresolution of 2D and 3D material images based on the approach of
Sandeep and Jacob. Numerical results confirm the moderate influence of the
dimensionality reduction on the overall superresolution result.
| stat |
A scoping review of causal methods enabling predictions under
hypothetical interventions | Background and Aims: The methods with which prediction models are usually
developed mean that neither the parameters nor the predictions should be
interpreted causally. However, when prediction models are used to support
decision making, there is often a need for predicting outcomes under
hypothetical interventions. We aimed to identify published methods for
developing and validating prediction models that enable risk estimation of
outcomes under hypothetical interventions, utilizing causal inference: their
main methodological approaches, underlying assumptions, targeted estimands, and
potential pitfalls and challenges with using the method, and unresolved
methodological challenges.
Methods: We systematically reviewed literature published by December 2019,
considering papers in the health domain that used causal considerations to
enable prediction models to be used for predictions under hypothetical
interventions.
Results: We identified 4919 papers through database searches and a further
115 papers through manual searches, of which 13 were selected for inclusion,
from both the statistical and the machine learning literature. Most of the
identified methods for causal inference from observational data were based on
marginal structural models and g-estimation.
Conclusions: There exist two broad methodological approaches for allowing
prediction under hypothetical intervention into clinical prediction models: 1)
enriching prediction models derived from observational studies with estimated
causal effects from clinical trials and meta-analyses; and 2) estimating
prediction models and causal effects directly from observational data. These
methods require extending to dynamic treatment regimes, and consideration of
multiple interventions to operationalise a clinical decision support system.
Techniques for validating 'causal prediction models' are still in their
infancy.
| stat |
Surprise sampling: improving and extending the local case-control
sampling | Fithian and Hastie (2014) proposed a new sampling scheme called local
case-control (LCC) sampling that achieves stability and efficiency by utilizing
a clever adjustment pertained to the logistic model. It is particularly useful
for classification with large and imbalanced data. This paper proposes a more
general sampling scheme based on a working principle that data points deserve
higher sampling probability if they contain more information or appear
"surprising" in the sense of, for example, a large error of pilot prediction or
a large absolute score. Compared with the relevant existing sampling schemes,
as reported in Fithian and Hastie (2014) and Ai, et al. (2018), the proposed
one has several advantages. It adaptively gives out the optimal forms to a
variety of objectives, including the LCC and Ai et al. (2018)'s sampling as
special cases. Under same model specifications, the proposed estimator also
performs no worse than those in the literature. The estimation procedure is
valid even if the model is misspecified and/or the pilot estimator is
inconsistent or dependent on full data. We present theoretical justifications
of the claimed advantages and optimality of the estimation and the sampling
design. Different from Ai, et al. (2018), our large sample theory are
population-wise rather than data-wise. Moreover, the proposed approach can be
applied to unsupervised learning studies, since it essentially only requires a
specific loss function and no response-covariate structure of data is needed.
Numerical studies are carried out and the evidence in support of the theory is
shown.
| stat |
A Large Dimensional Analysis of Least Squares Support Vector Machines | In this article, a large dimensional performance analysis of kernel least
squares support vector machines (LS-SVMs) is provided under the assumption of a
two-class Gaussian mixture model for the input data. Building upon recent
advances in random matrix theory, we show, when the dimension of data $p$ and
their number $n$ are both large, that the LS-SVM decision function can be well
approximated by a normally distributed random variable, the mean and variance
of which depend explicitly on a local behavior of the kernel function. This
theoretical result is then applied to the MNIST and Fashion-MNIST datasets
which, despite their non-Gaussianity, exhibit a convincingly close behavior.
Most importantly, our analysis provides a deeper understanding of the mechanism
into play in SVM-type methods and in particular of the impact on the choice of
the kernel function as well as some of their theoretical limits in separating
high dimensional Gaussian vectors.
| stat |
Bayesian Robustness: A Nonasymptotic Viewpoint | We study the problem of robustly estimating the posterior distribution for
the setting where observed data can be contaminated with potentially
adversarial outliers. We propose Rob-ULA, a robust variant of the Unadjusted
Langevin Algorithm (ULA), and provide a finite-sample analysis of its sampling
distribution. In particular, we show that after $T=
\tilde{\mathcal{O}}(d/\varepsilon_{\textsf{acc}})$ iterations, we can sample
from $p_T$ such that $\text{dist}(p_T, p^*) \leq \varepsilon_{\textsf{acc}} +
\tilde{\mathcal{O}}(\epsilon)$, where $\epsilon$ is the fraction of
corruptions. We corroborate our theoretical analysis with experiments on both
synthetic and real-world data sets for mean estimation, regression and binary
classification.
| stat |
Rate-optimal refinement strategies for local approximation MCMC | Many Bayesian inference problems involve target distributions whose density
functions are computationally expensive to evaluate. Replacing the target
density with a local approximation based on a small number of carefully chosen
density evaluations can significantly reduce the computational expense of
Markov chain Monte Carlo (MCMC) sampling. Moreover, continual refinement of the
local approximation can guarantee asymptotically exact sampling. We devise a
new strategy for balancing the decay rate of the bias due to the approximation
with that of the MCMC variance. We prove that the error of the resulting local
approximation MCMC (LA-MCMC) algorithm decays at roughly the expected
$1/\sqrt{T}$ rate, and we demonstrate this rate numerically. We also introduce
an algorithmic parameter that guarantees convergence given very weak tail
bounds, significantly strengthening previous convergence results. Finally, we
apply LA-MCMC to a computationally intensive Bayesian inverse problem arising
in groundwater hydrology.
| stat |
Likelihood-free parameter estimation for dynamic queueing networks: case
study of passenger flow in an international airport terminal | Dynamic queueing networks (DQN) model queueing systems where demand varies
strongly with time, such as airport terminals. With rapidly rising global air
passenger traffic placing increasing pressure on airport terminals, efficient
allocation of resources is more important than ever. Parameter inference and
quantification of uncertainty are key challenges for developing decision
support tools. The DQN likelihood function is, in general, intractable and
current approaches to simulation make likelihood-free parameter inference
methods, such as approximate Bayesian computation (ABC), infeasible since
simulating from these models is computationally expensive. By leveraging a
recent advance in computationally efficient queueing simulation, we develop the
first parameter inference approach for DQNs. We demonstrate our approach with
data of passenger flows in a real airport terminal, and we show that our model
accurately recreates the behaviour of the system and is useful for decision
support. Special care must be taken in developing the distance for ABC since
any useful output must vary with time. We use maximum mean discrepancy, a
metric on probability measures, as the distance function for ABC. Prediction
intervals of performance measures for decision support tools are easily
constructed using draws from posterior samples, which we demonstrate with a
scenario of a delayed flight.
| stat |
Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse
Coding | We construct an unsupervised learning model that achieves nonlinear
disentanglement of underlying factors of variation in naturalistic videos.
Previous work suggests that representations can be disentangled if all but a
few factors in the environment stay constant at any point in time. As a result,
algorithms proposed for this problem have only been tested on carefully
constructed datasets with this exact property, leaving it unclear whether they
will transfer to natural scenes. Here we provide evidence that objects in
segmented natural movies undergo transitions that are typically small in
magnitude with occasional large jumps, which is characteristic of a temporally
sparse distribution. We leverage this finding and present SlowVAE, a model for
unsupervised representation learning that uses a sparse prior on temporally
adjacent observations to disentangle generative factors without any assumptions
on the number of changing factors. We provide a proof of identifiability and
show that the model reliably learns disentangled representations on several
established benchmark datasets, often surpassing the current state-of-the-art.
We additionally demonstrate transferability towards video datasets with natural
dynamics, Natural Sprites and KITTI Masks, which we contribute as benchmarks
for guiding disentanglement research towards more natural data domains.
| stat |
Distortion estimates for approximate Bayesian inference | Current literature on posterior approximation for Bayesian inference offers
many alternative methods. Does our chosen approximation scheme work well on the
observed data? The best existing generic diagnostic tools treating this kind of
question by looking at performance averaged over data space, or otherwise lack
diagnostic detail. However, if the approximation is bad for most data, but good
at the observed data, then we may discard a useful approximation. We give
graphical diagnostics for posterior approximation at the observed data. We
estimate a "distortion map" that acts on univariate marginals of the
approximate posterior to move them closer to the exact posterior, without
recourse to the exact posterior.
| stat |
Principled learning method for Wasserstein distributionally robust
optimization with local perturbations | Wasserstein distributionally robust optimization (WDRO) attempts to learn a
model that minimizes the local worst-case risk in the vicinity of the empirical
data distribution defined by Wasserstein ball. While WDRO has received
attention as a promising tool for inference since its introduction, its
theoretical understanding has not been fully matured. Gao et al. (2017)
proposed a minimizer based on a tractable approximation of the local worst-case
risk, but without showing risk consistency. In this paper, we propose a
minimizer based on a novel approximation theorem and provide the corresponding
risk consistency results. Furthermore, we develop WDRO inference for locally
perturbed data that include the Mixup (Zhang et al., 2017) as a special case.
We show that our approximation and risk consistency results naturally extend to
the cases when data are locally perturbed. Numerical experiments demonstrate
robustness of the proposed method using image classification datasets. Our
results show that the proposed method achieves significantly higher accuracy
than baseline models on noisy datasets.
| stat |
Logistic-Regression with peer-group effects via inference in higher
order Ising models | Spin glass models, such as the Sherrington-Kirkpatrick, Hopfield and Ising
models, are all well-studied members of the exponential family of discrete
distributions, and have been influential in a number of application domains
where they are used to model correlation phenomena on networks. Conventionally
these models have quadratic sufficient statistics and consequently capture
correlations arising from pairwise interactions. In this work we study
extensions of these to models with higher-order sufficient statistics, modeling
behavior on a social network with peer-group effects. In particular, we model
binary outcomes on a network as a higher-order spin glass, where the behavior
of an individual depends on a linear function of their own vector of covariates
and some polynomial function of the behavior of others, capturing peer-group
effects. Using a {\em single}, high-dimensional sample from such model our goal
is to recover the coefficients of the linear function as well as the strength
of the peer-group effects. The heart of our result is a novel approach for
showing strong concavity of the log pseudo-likelihood of the model, implying
statistical error rate of $\sqrt{d/n}$ for the Maximum Pseudo-Likelihood
Estimator (MPLE), where $d$ is the dimensionality of the covariate vectors and
$n$ is the size of the network (number of nodes). Our model generalizes vanilla
logistic regression as well as the peer-effect models studied in recent works,
and our results extend these results to accommodate higher-order interactions.
| stat |
Double Happiness: Enhancing the Coupled Gains of L-lag Coupling via
Control Variates | The recently proposed L-lag coupling for unbiased Markov chain Monte Carlo
(MCMC) calls for a joint celebration by MCMC practitioners and theoreticians.
For practitioners, it circumvents the thorny issue of deciding the burn-in
period or when to terminate an MCMC sampling process, and opens the door for
safe parallel implementation. For theoreticians, it provides a powerful tool to
establish elegant and easily estimable bounds on the exact error of an MCMC
approximation at any finite number of iterates. A serendipitous observation
about the bias-correcting term leads us to introduce naturally available
control variates into the L-lag coupling estimators. In turn, this extension
enhances the coupled gains of L-lag coupling, because it results in more
efficient unbiased estimators, as well as a better bound on the total variation
error of MCMC iterations, albeit the gains diminish as L increases.
Specifically, the new upper bound is theoretically guaranteed to never exceed
the one given previously. We also argue that L-lag coupling represents a
coupling for the future, breaking from the coupling-from-the-past type of
perfect sampling, by reducing the generally unachievable requirement of being
perfect to one of being unbiased, a worthwhile trade-off for ease of
implementation in most practical situations. The theoretical analysis is
supported by numerical experiments that show tighter bounds and a gain in
efficiency when control variates are introduced.
| stat |
Spectral neighbor joining for reconstruction of latent tree models | A common assumption in multiple scientific applications is that the
distribution of observed data can be modeled by a latent tree graphical model.
An important example is phylogenetics, where the tree models the evolutionary
lineages of a set of observed organisms. Given a set of independent
realizations of the random variables at the leaves of the tree, a key challenge
is to infer the underlying tree topology. In this work we develop Spectral
Neighbor Joining (SNJ), a novel method to recover the structure of latent tree
graphical models. Given a matrix that contains a measure of similarity between
all pairs of observed variables, SNJ computes a spectral measure of cohesion
between groups of observed variables. We prove that SNJ is consistent, and
derive a sufficient condition for correct tree recovery from an estimated
similarity matrix. Combining this condition with a concentration of measure
result on the similarity matrix, we bound the number of samples required to
recover the tree with high probability. We illustrate via extensive simulations
that in comparison to several other reconstruction methods, SNJ requires fewer
samples to accurately recover trees with a large number of leaves or long
edges.
| stat |
Generalisation Guarantees for Continual Learning with Orthogonal
Gradient Descent | In Continual Learning settings, deep neural networks are prone to
Catastrophic Forgetting. Orthogonal Gradient Descent was proposed to tackle the
challenge. However, no theoretical guarantees have been proven yet. We present
a theoretical framework to study Continual Learning algorithms in the Neural
Tangent Kernel regime. This framework comprises closed form expression of the
model through tasks and proxies for Transfer Learning, generalisation and tasks
similarity. In this framework, we prove that OGD is robust to Catastrophic
Forgetting then derive the first generalisation bound for SGD and OGD for
Continual Learning. Finally, we study the limits of this framework in practice
for OGD and highlight the importance of the Neural Tangent Kernel variation for
Continual Learning with OGD.
| stat |
High-dimensional regression in practice: an empirical study of
finite-sample prediction, variable selection and ranking | Penalized likelihood approaches are widely used for high-dimensional
regression. Although many methods have been proposed and the associated theory
is now well-developed, the relative efficacy of different approaches in
finite-sample settings, as encountered in practice, remains incompletely
understood. There is therefore a need for empirical investigations in this area
that can offer practical insight and guidance to users. In this paper we
present a large-scale comparison of penalized regression methods. We
distinguish between three related goals: prediction, variable selection and
variable ranking. Our results span more than 2,300 data-generating scenarios,
including both synthetic and semi-synthetic data (real covariates and simulated
responses), allowing us to systematically consider the influence of various
factors (sample size, dimensionality, sparsity, signal strength and
multicollinearity). We consider several widely-used approaches (Lasso, Adaptive
Lasso, Elastic Net, Ridge Regression, SCAD, the Dantzig Selector and Stability
Selection). We find considerable variation in performance between methods. Our
results support a `no panacea' view, with no unambiguous winner across all
scenarios or goals, even in this restricted setting where all data align well
with the assumptions underlying the methods. The study allows us to make some
recommendations as to which approaches may be most (or least) suitable given
the goal and some data characteristics. Our empirical results complement
existing theory and provide a resource to compare methods across a range of
scenarios and metrics.
| stat |
Dealing with Stochastic Volatility in Time Series Using the R Package
stochvol | The R package stochvol provides a fully Bayesian implementation of
heteroskedasticity modeling within the framework of stochastic volatility. It
utilizes Markov chain Monte Carlo (MCMC) samplers to conduct inference by
obtaining draws from the posterior distribution of parameters and latent
variables which can then be used for predicting future volatilities. The
package can straightforwardly be employed as a stand-alone tool; moreover, it
allows for easy incorporation into other MCMC samplers. The main focus of this
paper is to show the functionality of stochvol. In addition, it provides a
brief mathematical description of the model, an overview of the sampling
schemes used, and several illustrative examples using exchange rate data.
| stat |
Model selection with lasso-zero: adding straw to the haystack to better
find needles | The high-dimensional linear model $y = X \beta^0 + \epsilon$ is considered
and the focus is put on the problem of recovering the support $S^0$ of the
sparse vector $\beta^0.$ We introduce Lasso-Zero, a new $\ell_1$-based
estimator whose novelty resides in an "overfit, then threshold" paradigm and
the use of noise dictionaries concatenated to $X$ for overfitting the response.
To select the threshold, we employ the quantile universal threshold based on a
pivotal statistic that requires neither knowledge nor preliminary estimation of
the noise level. Numerical simulations show that Lasso-Zero performs well in
terms of support recovery and provides an excellent trade-off between high true
positive rate and low false discovery rate compared to competitors. Our
methodology is supported by theoretical results showing that when no noise
dictionary is used, Lasso-Zero recovers the signs of $\beta^0$ under weaker
conditions on $X$ and $S^0$ than the Lasso and achieves sign consistency for
correlated Gaussian designs. The use of noise dictionary improves the procedure
for low signals.
| stat |
Feature-weighted elastic net: using "features of features" for better
prediction | In some supervised learning settings, the practitioner might have additional
information on the features used for prediction. We propose a new method which
leverages this additional information for better prediction. The method, which
we call the feature-weighted elastic net ("fwelnet"), uses these "features of
features" to adapt the relative penalties on the feature coefficients in the
elastic net penalty. In our simulations, fwelnet outperforms the lasso in terms
of test mean squared error and usually gives an improvement in true positive
rate or false positive rate for feature selection. We also apply this method to
early prediction of preeclampsia, where fwelnet outperforms the lasso in terms
of 10-fold cross-validated area under the curve (0.86 vs. 0.80). We also
provide a connection between fwelnet and the group lasso and suggest how
fwelnet might be used for multi-task learning.
| stat |
Improving image classifiers for small datasets by learning rate
adaptations | Our paper introduces an efficient combination of established techniques to
improve classifier performance, in terms of accuracy and training time. We
achieve two-fold to ten-fold speedup in nearing state of the art accuracy, over
different model architectures, by dynamically tuning the learning rate. We find
it especially beneficial in the case of a small dataset, where reliability of
machine reasoning is lower. We validate our approach by comparing our method
versus vanilla training on CIFAR-10. We also demonstrate its practical
viability by implementing on an unbalanced corpus of diagnostic images.
| stat |
Forecasting vegetation condition for drought early warning systems in
pastoral communities in Kenya | Droughts are a recurring hazard in sub-Saharan Africa, that can wreak huge
socioeconomic costs.Acting early based on alerts provided by early warning
systems (EWS) can potentially provide substantial mitigation, reducing the
financial and human cost. However, existing EWS tend only to monitor current,
rather than forecast future, environmental and socioeconomic indicators of
drought, and hence are not always sufficiently timely to be effective in
practice. Here we present a novel method for forecasting satellite-based
indicators of vegetation condition. Specifically, we focused on the 3-month
Vegetation Condition Index (VCI3M) over pastoral livelihood zones in Kenya,
which is the indicator used by the Kenyan National Drought Management
Authority(NDMA). Using data from MODIS and Landsat, we apply linear
autoregression and Gaussian process modeling methods and demonstrate high
forecasting skill several weeks ahead. As a benchmark we predicted the drought
alert marker used by NDMA (VCI3M<35). Both of our models were able to predict
this alert marker four weeks ahead with a hit rate of around 89% and a false
alarm rate of around 4%, or 81% and 6% respectively six weeks ahead. The
methods developed here can thus identify a deteriorating vegetation condition
well and sufficiently in advance to help disaster risk managers act early to
support vulnerable communities and limit the impact of a drought hazard.
| stat |
Understanding the Effect of Incentivized Advertising along the
Conversion Funnel | In an effort to combat ad annoyance in mobile apps, publishers have
introduced a new ad format called "Incentivized Advertising" or "Rewarded
Advertising", whereby users receive rewards in exchange for watching ads. There
is much debate in the industry regarding its' effectiveness. On the one hand,
incentivized advertising is less intrusive and annoying, but on the other hand,
users might be more interested in the rewards rather than the ad content. Using
a large dataset of 1 million impressions from a mobile advertising platform,
and in three separate quasi-experimental approaches, we find that incentivized
advertising leads to lower users' click-through rates, but a higher overall
install rate of the advertised app.
In the second part, we study the mechanism of how incentivized advertising
affects users' behavior. We test the hypothesis that incentivized advertising
causes a temptation effect, whereby users prefer to collect and enjoy their
rewards immediately, instead of pursuing the ads. We find the temptation effect
is stronger when (i) users have to wait longer before receiving the rewards and
when (ii) the value of the reward is relatively larger. We further find support
that incentivized advertising has a positive effect of reducing ad annoyance --
an effect that is stronger for small-screen mobile devices, where advertising
is more annoying. Finally, we take the publisher's perspective and quantify the
overall effect on ad revenue. Our difference-in-differences estimates suggest
switching to incentivized advertising would increase the publisher's revenue by
$3.10 per 1,000 impressions.
| stat |
Asymptotically Optimal Algorithms for Budgeted Multiple Play Bandits | We study a generalization of the multi-armed bandit problem with multiple
plays where there is a cost associated with pulling each arm and the agent has
a budget at each time that dictates how much she can expect to spend. We derive
an asymptotic regret lower bound for any uniformly efficient algorithm in our
setting. We then study a variant of Thompson sampling for Bernoulli rewards and
a variant of KL-UCB for both single-parameter exponential families and bounded,
finitely supported rewards. We show these algorithms are asymptotically
optimal, both in rateand leading problem-dependent constants, including in the
thick margin setting where multiple arms fall on the decision boundary.
| stat |
Automated detection of business-relevant outliers in e-commerce
conversion rate | We evaluate how modern outlier detection methods perform in identifying
outliers in e-commerce conversion rate data. Based on the limitations
identified, we then present a novel method to detect outliers in e-commerce
conversion rate. This unsupervised method is made more business relevant by
letting it automatically adjust the sensitivity based on the activity observed
on the e-commerce platform. We call this outlier detection method the fluid
IQR. Using real e-commerce conversion data acquired from a known store, we
compare the performance of the existing and the new outlier detection methods.
Fluid IQR method outperforms the existing outlier detection methods by a large
margin when it comes to business-relevance. Furthermore, the fluids IQR method
is the most robust outlier detection method in the presence of clusters of
extreme outliers or level shifts. Future research will evaluate how the fluid
IQR method perform in diverse e-business settings.
| stat |
Boulevard: Regularized Stochastic Gradient Boosted Trees and Their
Limiting Distribution | This paper examines a novel gradient boosting framework for regression. We
regularize gradient boosted trees by introducing subsampling and employ a
modified shrinkage algorithm so that at every boosting stage the estimate is
given by an average of trees. The resulting algorithm, titled Boulevard, is
shown to converge as the number of trees grows. We also demonstrate a central
limit theorem for this limit, allowing a characterization of uncertainty for
predictions. A simulation study and real world examples provide support for
both the predictive accuracy of the model and its limiting behavior.
| stat |
Stochastic modeling of in vitro bactericidal potency | We provide a Galton--Watson model for the growth of a bacterial population in
the presence of antibiotics. We assume that bacterial cells either die or
duplicate, and the corresponding probabilities depend on the concentration of
the antibiotic. Assuming that the mean offspring number is given by $m(c) = 2 /
(1 + \alpha c^\beta)$ for some $\alpha, \beta$, where $c$ stands for the
antibiotic concentration we obtain weakly consistent, asymptotically normal
estimator both for $(\alpha, \beta)$ and for the minimal inhibitory
concentration (MIC), a relevant parameter in pharmacology. We apply our method
to real data, where \emph{Chlamydia trachomatis} bacteria was treated by
azithromycin and ciprofloxacin. For the measurements of \emph{Chlamydia} growth
quantitative PCR technique was used. The 2-parameter model fits remarkably well
to the biological data.
| stat |
Merging versus Ensembling in Multi-Study Machine Learning: Theoretical
Insight from Random Effects | A critical decision point when training predictors using multiple studies is
whether these studies should be combined or treated separately. We compare two
multi-study learning approaches in the presence of potential heterogeneity in
predictor-outcome relationships across datasets. We consider 1) merging all of
the datasets and training a single learner, and 2) multi-study ensembling,
which involves training a separate learner on each dataset and combining the
predictions resulting from each learner. In a linear regression setting, we
show analytically and confirm via simulation that merging yields lower
prediction error than ensembling when the predictor-outcome relationships are
relatively homogeneous across studies. However, as cross-study heterogeneity
increases, there exists a transition point beyond which ensembling outperforms
merging. We provide analytic expressions for the transition point in various
scenarios, study asymptotic properties, and illustrate how transition point
theory can be used for deciding when studies should be combined with an
application from metabolomics.
| stat |
Enriched Mixtures of Gaussian Process Experts | Mixtures of experts probabilistically divide the input space into regions,
where the assumptions of each expert, or conditional model, need only hold
locally. Combined with Gaussian process (GP) experts, this results in a
powerful and highly flexible model. We focus on alternative mixtures of GP
experts, which model the joint distribution of the inputs and targets
explicitly. We highlight issues of this approach in multi-dimensional input
spaces, namely, poor scalability and the need for an unnecessarily large number
of experts, degrading the predictive performance and increasing uncertainty. We
construct a novel model to address these issues through a nested partitioning
scheme that automatically infers the number of components at both levels.
Multiple response types are accommodated through a generalised GP framework,
while multiple input types are included through a factorised exponential family
structure. We show the effectiveness of our approach in estimating a
parsimonious probabilistic description of both synthetic data of increasing
dimension and an Alzheimer's challenge dataset.
| stat |
Bayesian Aggregation | A general challenge in statistics is prediction in the presence of multiple
candidate models or learning algorithms. Model aggregation tries to combine all
predictive distributions from individual models, which is more stable and
flexible than single model selection. In this article we describe when and how
to aggregate models under the lens of Bayesian decision theory. Among two
widely used methods, Bayesian model averaging (BMA) and Bayesian stacking, we
compare their predictive performance, and review their theoretical optimality,
probabilistic interpretation, practical implementation, and extensions in
complex models.
| stat |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.