title
stringlengths 6
244
| abstract
stringlengths 19
5.09k
| label
stringclasses 10
values |
|---|---|---|
Estimation and Inference for Multi-Kink Quantile Regression | The Multi-Kink Quantile Regression (MKQR) model is an important tool for
analyzing data with heterogeneous conditional distributions, especially when
quantiles of response variable are of interest, due to its robustness to
outliers and heavy-tailed errors in the response. It assumes different linear
quantile regression forms in different regions of the domain of the threshold
covariate but are still continuous at kink points. In this paper, we
investigate parameter estimation, kink point detection and statistical
inference in MKQR models. We propose an iterative segmented quantile regression
algorithm for estimating both the regression coefficients and the locations of
kink points. The proposed algorithm is much more computationally efficient than
the grid search algorithm and not sensitive to the selection of initial values.
Asymptotic properties, such as selection consistency of the number of kink
points, asymptotic normality of the estimators of both regression coefficients
and kink effects, are established to justify the proposed method theoretically.
A score test, based on partial subgradients, is developed to verify whether the
kink effects exist or not. Test-inversion confidence intervals for kink
location parameters are also constructed. Intensive simulation studies
conducted show the proposed methods work very well when sample size is finite.
Finally, we apply the MKQR models together with the proposed methods to the
dataset about secondary industrial structure of China and the dataset about
triceps skinfold thickness of Gambian females, which leads to some very
interesting findings. A new R package MultiKink is developed to implement the
proposed methods.
| stat |
Robustified Multivariate Regression and Classification Using
Distributionally Robust Optimization under the Wasserstein Metric | We develop Distributionally Robust Optimization (DRO) formulations for
Multivariate Linear Regression (MLR) and Multiclass Logistic Regression (MLG)
when both the covariates and responses/labels may be contaminated by outliers.
The DRO framework uses a probabilistic ambiguity set defined as a ball of
distributions that are close to the empirical distribution of the training set
in the sense of the Wasserstein metric. We relax the DRO formulation into a
regularized learning problem whose regularizer is a norm of the coefficient
matrix. We establish out-of-sample performance guarantees for the solutions to
our model, offering insights on the role of the regularizer in controlling the
prediction error. Experimental results show that our approach improves the
predictive error by 7% -- 37% for MLR, and a metric of robustness by 100% for
MLG.
| stat |
Coarse- and fine-scale geometric information content of Multiclass
Classification and implied Data-driven Intelligence | Under any Multiclass Classification (MCC) setting defined by a collection of
labeled point-cloud specified by a feature-set, we extract only stochastic
partial orderings from all possible triplets of point-cloud without explicitly
measuring the three cloud-to-cloud distances. We demonstrate that such a
collective of partial ordering can efficiently compute a label embedding tree
geometry on the Label-space. This tree in turn gives rise to a predictive
graph, or a network with precisely weighted linkages. Such two multiscale
geometries are taken as the coarse scale information content of MCC. They
indeed jointly shed lights on explainable knowledge on why and how labeling
comes about and facilitates error-free prediction with potential multiple
candidate labels supported by data. For revealing within-label heterogeneity,
we further undergo labeling naturally found clusters within each point-cloud,
and likewise derive multiscale geometry as its fine-scale information content
contained in data. This fine-scale endeavor shows that our computational
proposal is indeed scalable to a MCC setting having a large label-space.
Overall the computed multiscale collective of data-driven patterns and
knowledge will serve as a basis for constructing visible and explainable
subject matter intelligence regarding the system of interest.
| stat |
Multi-Complementary and Unlabeled Learning for Arbitrary Losses and
Models | A weakly-supervised learning framework named as complementary-label learning
has been proposed recently, where each sample is equipped with a single
complementary label that denotes one of the classes the sample does not belong
to. However, the existing complementary-label learning methods cannot learn
from the easily accessible unlabeled samples and samples with multiple
complementary labels, which are more informative. In this paper, to remove
these limitations, we propose the novel multi-complementary and unlabeled
learning framework that allows unbiased estimation of classification risk from
samples with any number of complementary labels and unlabeled samples, for
arbitrary loss functions and models. We first give an unbiased estimator of the
classification risk from samples with multiple complementary labels, and then
further improve the estimator by incorporating unlabeled samples into the risk
formulation. The estimation error bounds show that the proposed methods are in
the optimal parametric convergence rate. Finally, the experiments on both
linear and deep models show the effectiveness of our methods.
| stat |
A Conceptual Introduction to Markov Chain Monte Carlo Methods | Markov Chain Monte Carlo (MCMC) methods have become a cornerstone of many
modern scientific analyses by providing a straightforward approach to
numerically estimate uncertainties in the parameters of a model using a
sequence of random samples. This article provides a basic introduction to MCMC
methods by establishing a strong conceptual understanding of what problems MCMC
methods are trying to solve, why we want to use them, and how they work in
theory and in practice. To develop these concepts, I outline the foundations of
Bayesian inference, discuss how posterior distributions are used in practice,
explore basic approaches to estimate posterior-based quantities, and derive
their link to Monte Carlo sampling and MCMC. Using a simple toy problem, I then
demonstrate how these concepts can be used to understand the benefits and
drawbacks of various MCMC approaches. Exercises designed to highlight various
concepts are also included throughout the article.
| stat |
Sharp hypotheses and bispatial inference | A fundamental class of inferential problems are those characterised by there
having been a substantial degree of pre-data (or prior) belief that the value
of a model parameter was equal or lay close to a specified value, which may,
for example, be the value that indicates the absence of an effect. Standard
ways of tackling problems of this type, including the Bayesian method, are
often highly inadequate in practice. To address this issue, an inferential
framework called bispatial inference is put forward, which can be viewed as
both a generalisation and radical reinterpretation of existing approaches to
inference that are based on P values. It is shown that to obtain an appropriate
post-data density function for a given parameter, it is often convenient to
combine a special type of bispatial inference, which is constructed around
one-sided P values, with a previously outlined form of fiducial inference.
Finally, by using what are called post-data opinion curves, this
bispatial-fiducial theory is naturally extended to deal with the general
scenario in which any number of parameters may be unknown. The application of
the theory is illustrated in various examples, which are especially relevant to
the analysis of clinical trial data.
| stat |
Prior Convictions: Black-Box Adversarial Attacks with Bandits and Priors | We study the problem of generating adversarial examples in a black-box
setting in which only loss-oracle access to a model is available. We introduce
a framework that conceptually unifies much of the existing work on black-box
attacks, and we demonstrate that the current state-of-the-art methods are
optimal in a natural sense. Despite this optimality, we show how to improve
black-box attacks by bringing a new element into the problem: gradient priors.
We give a bandit optimization-based algorithm that allows us to seamlessly
integrate any such priors, and we explicitly identify and incorporate two
examples. The resulting methods use two to four times fewer queries and fail
two to five times less often than the current state-of-the-art.
| stat |
Newcomb-Benford's law as a fast ersatz of discrepancy measures | Thanks to the increasing availability in computing power, high-dimensional
engineering problems seem to be at reach. But the curse of dimensionality will
always prevent us to try out extensively all the hypotheses. There is a vast
literature on efficient methods to construct a Design of Experiments (DoE) such
as low discrepancy sequences and optimized designs. Classically, the
performance of these methods is assessed using a discrepancy metric. Having a
fast discrepancy measure is of prime importance if ones want to optimize a
design. This work proposes a new methodology to assess the quality of a random
sampling by using a flavor of Newcomb-Benford's law. The performance of the new
metric is compared to classical discrepancy measures and showed to offer
similar information at a fraction of the computational cost of traditional
discrepancy measures.
| stat |
A New Smoothing Technique based on the Parallel Concatenation of
Forward/Backward Bayesian Filters: Turbo Smoothing | Recently, a novel method for developing filtering algorithms, based on the
parallel concatenation of Bayesian filters and called turbo filtering, has been
proposed. In this manuscript we show how the same conceptual approach can be
exploited to devise a new smoothing method, called turbo smoothing. A turbo
smoother combines a turbo filter, employed in its forward pass, with the
parallel concatenation of two backward information filters used in its backward
pass. As a specific application of our general theory, a detailed derivation of
two turbo smoothing algorithms for conditionally linear Gaussian systems is
illustrated. Numerical results for a specific dynamic system evidence that
these algorithms can achieve a better complexity-accuracy tradeoff than other
smoothing techniques recently appeared in the literature.
| stat |
Optimal Variance Control of the Score Function Gradient Estimator for
Importance Weighted Bounds | This paper introduces novel results for the score function gradient estimator
of the importance weighted variational bound (IWAE). We prove that in the limit
of large $K$ (number of importance samples) one can choose the control variate
such that the Signal-to-Noise ratio (SNR) of the estimator grows as $\sqrt{K}$.
This is in contrast to the standard pathwise gradient estimator where the SNR
decreases as $1/\sqrt{K}$. Based on our theoretical findings we develop a novel
control variate that extends on VIMCO. Empirically, for the training of both
continuous and discrete generative models, the proposed method yields superior
variance reduction, resulting in an SNR for IWAE that increases with $K$
without relying on the reparameterization trick. The novel estimator is
competitive with state-of-the-art reparameterization-free gradient estimators
such as Reweighted Wake-Sleep (RWS) and the thermodynamic variational objective
(TVO) when training generative models.
| stat |
Sequential online subsampling for thinning experimental designs | We consider a design problem where experimental conditions (design points
$X_i$) are presented in the form of a sequence of i.i.d.\ random variables,
generated with an unknown probability measure $\mu$, and only a given
proportion $\alpha\in(0,1)$ can be selected. The objective is to select good
candidates $X_i$ on the fly and maximize a concave function $\Phi$ of the
corresponding information matrix. The optimal solution corresponds to the
construction of an optimal bounded design measure $\xi_\alpha^*\leq
\mu/\alpha$, with the difficulty that $\mu$ is unknown and $\xi_\alpha^*$ must
be constructed online. The construction proposed relies on the definition of a
threshold $\tau$ on the directional derivative of $\Phi$ at the current
information matrix, the value of $\tau$ being fixed by a certain quantile of
the distribution of this directional derivative. Combination with recursive
quantile estimation yields a nonlinear two-time-scale stochastic approximation
method. It can be applied to very long design sequences since only the current
information matrix and estimated quantile need to be stored. Convergence to an
optimum design is proved. Various illustrative examples are presented.
| stat |
Enhanced Monte Carlo Estimation of the Fisher Information Matrix with
Independent Perturbations for Complex Problems | The Fisher information matrix provides a way to measure the amount of
information given observed data based on parameters of interest. Many
applications of the FIM exist in statistical modeling, system identification,
and parameter estimation. We sometimes use the Monte Carlo-based method to
estimate the FIM because its analytical form is often impossible or difficult
to be computed in real-world models. In this paper, we review the basic method
based on simultaneous perturbations and present an enhanced resampling-based
method with independent simultaneous perturbations to estimate the Fisher
information matrix. We conduct theoretical and numerical analysis to show its
accuracy via variance reduction from $O(1/N)$ to $O(1/(nN))$, where $n$ is the
sample size of the data and $N$ is a measure of the Monte Carlo averaging. We
also consider the trade-off between accuracy and computational cost.
| stat |
To Bag is to Prune | It is notoriously difficult to build a bad Random Forest (RF). Concurrently,
RF is perhaps the only standard ML algorithm that blatantly overfits in-sample
without any consequence out-of-sample. Standard arguments cannot rationalize
this paradox. I propose a new explanation: bootstrap aggregation and model
perturbation as implemented by RF automatically prune a (latent) true
underlying tree. More generally, randomized ensembles of greedily optimized
learners implicitly perform optimal early stopping out-of-sample. So there is
no need to tune the stopping point. By construction, variants of Boosting and
MARS are also eligible for automatic tuning. I empirically demonstrate the
property, with simulated and real data, by reporting that these new completely
overfitting ensembles yield an out-of-sample performance equivalent to that of
their tuned counterparts -- or better.
| stat |
Spectral estimation from simulations via sketching | Sketching is a stochastic dimension reduction method that preserves geometric
structures of data and has applications in high-dimensional regression, low
rank approximation and graph sparsification. In this work, we show that
sketching can be used to compress simulation data and still accurately estimate
time autocorrelation and power spectral density. For a given compression ratio,
the accuracy is much higher than using previously known methods. In addition to
providing theoretical guarantees, we apply sketching to a molecular dynamics
simulation of methanol and find that the estimate of spectral density is 90%
accurate using only 10% of the data.
| stat |
A Mean Field Theory of Quantized Deep Networks: The Quantization-Depth
Trade-Off | Reducing the precision of weights and activation functions in neural network
training, with minimal impact on performance, is essential for the deployment
of these models in resource-constrained environments. We apply mean-field
techniques to networks with quantized activations in order to evaluate the
degree to which quantization degrades signal propagation at initialization. We
derive initialization schemes which maximize signal propagation in such
networks and suggest why this is helpful for generalization. Building on these
results, we obtain a closed form implicit equation for $L_{\max}$, the maximal
trainable depth (and hence model capacity), given $N$, the number of
quantization levels in the activation function. Solving this equation
numerically, we obtain asymptotically: $L_{\max}\propto N^{1.82}$.
| stat |
Above-ground biomass change estimation using national forest inventory
data with Sentinel-2 and Landsat 8 | This study aimed at estimating total forest above-ground net change (Delta
AGB, Mt) over five years (2014-2019) based on model-assisted estimation
utilizing freely available satellite imagery. The study was conducted for a
boreal forest area (approx. 1.4 Mill hectares) in Norway where bi-temporal
national forest inventory (NFI), Sentinel-2, and Landsat data were available.
Biomass change was modelled based on a direct approach. The precision of
estimates using only the NFI data in a basic expansion estimator were compared
to four different alternative model-assisted estimates using 1) Sentinel-2 or
Landsat data, and 2) using bi- or uni-temporal remotely sensed data. We found
that the use of remotely sensed data improved the precision of the purely
field-based estimates by a factor of up to three. The most precise estimates
were found for the model-assisted estimation using bi-temporal Sentinel-2
(standard error; SE= 1.7 Mt). However, the decrease in precision when using
Landsat data was small (SE= 1.92 Mt). In addition, we found that Delta AGB
could be precisely estimated also when remotely sensed data were available only
at the end of the monitoring period. We conclude that satellite optical data
can considerably improve Delta AGB estimates, even in those cases where
repeated and coincident NFI data are available. The free availability, global
coverage, frequent update, and long-term time horizon make data from programs
such as Sentinel-2 and Landsat a valuable data source for a consistent and
durable monitoring of forest carbon dynamics.
| stat |
Quantile Treatment Effects and Bootstrap Inference under
Covariate-Adaptive Randomization | In this paper, we study the estimation and inference of the quantile
treatment effect under covariate-adaptive randomization. We propose two
estimation methods: (1) the simple quantile regression and (2) the inverse
propensity score weighted quantile regression. For the two estimators, we
derive their asymptotic distributions uniformly over a compact set of quantile
indexes, and show that, when the treatment assignment rule does not achieve
strong balance, the inverse propensity score weighted estimator has a smaller
asymptotic variance than the simple quantile regression estimator. For the
inference of method (1), we show that the Wald test using a weighted bootstrap
standard error under-rejects. But for method (2), its asymptotic size equals
the nominal level. We also show that, for both methods, the asymptotic size of
the Wald test using a covariate-adaptive bootstrap standard error equals the
nominal level. We illustrate the finite sample performance of the new
estimation and inference methods using both simulated and real datasets.
| stat |
Animal Movement Models with Mechanistic Selection Functions | A suite of statistical methods are used to study animal movement. Most of
these methods treat animal telemetry data in one of three ways: as discrete
processes, as continuous processes, or as point processes. We briefly review
each of these approaches and then focus in on the latter. In the context of
point processes, so-called resource selection analyses are among the most
common way to statistically treat animal telemetry data. However, most resource
selection analyses provide inference based on approximations of point process
models. The forms of these models have been limited to a few types of
specifications that provide inference about relative resource use and, less
commonly, probability of use. For more general spatio-temporal point process
models, the most common type of analysis often proceeds with a data
augmentation approach that is used to create a binary data set that can be
analyzed with conditional logistic regression. We show that the conditional
logistic regression likelihood can be generalized to accommodate a variety of
alternative specifications related to resource selection. We then provide an
example of a case where a spatio-temporal point process model coincides with
that implied by a mechanistic model for movement expressed as a partial
differential equation derived from first principles of movement. We demonstrate
that inference from this form of point process model is intuitive (and could be
useful for management and conservation) by analyzing a set of telemetry data
from a mountain lion in Colorado, USA, to understand the effects of spatially
explicit environmental conditions on movement behavior of this species.
| stat |
Sliced Inverse Regression for Spatial Data | Sliced inverse regression is one of the most popular sufficient dimension
reduction methods. Originally, it was designed for independent and identically
distributed data and recently extend to the case of serially and spatially
dependent data. In this work we extend it to the case of spatially dependent
data where the response might depend also on neighbouring covariates when the
observations are taken on a grid-like structure as it is often the case in
econometric spatial regression applications. We suggest guidelines on how to
decide upon the dimension of the subspace of interest and also which spatial
lag might be of interest when modeling the response. These guidelines are
supported by a conducted simulation study.
| stat |
Combining Online Learning Guarantees | We show how to take any two parameter-free online learning algorithms with
different regret guarantees and obtain a single algorithm whose regret is the
minimum of the two base algorithms. Our method is embarrassingly simple: just
add the iterates. This trick can generate efficient algorithms that adapt to
many norms simultaneously, as well as providing diagonal-style algorithms that
still maintain dimension-free guarantees. We then proceed to show how a variant
on this idea yields a black-box procedure for generating optimistic online
learning algorithms. This yields the first optimistic regret guarantees in the
unconstrained setting and generically increases adaptivity. Further, our
optimistic algorithms are guaranteed to do no worse than their non-optimistic
counterparts regardless of the quality of the optimistic estimates provided to
the algorithm.
| stat |
A Bayesian algorithm for retrosynthesis | The identification of synthetic routes that end with a desired product has
been an inherently time-consuming process that is largely dependent on expert
knowledge regarding a limited fraction of the entire reaction space. At
present, emerging machine-learning technologies are overturning the process of
retrosynthetic planning. The objective of this study is to discover synthetic
routes backwardly from a given desired molecule to commercially available
compounds. The problem is reduced to a combinatorial optimization task with the
solution space subject to the combinatorial complexity of all possible pairs of
purchasable reactants. We address this issue within the framework of Bayesian
inference and computation. The workflow consists of two steps: a deep neural
network is trained that forwardly predicts a product of the given reactants
with a high level of accuracy, following which this forward model is inverted
into the backward one via Bayes' law of conditional probability. Using the
backward model, a diverse set of highly probable reaction sequences ending with
a given synthetic target is exhaustively explored using a Monte Carlo search
algorithm. The Bayesian retrosynthesis algorithm could successfully rediscover
80.3% and 50.0% of known synthetic routes of single-step and two-step reactions
within top-10 accuracy, respectively, thereby outperforming state-of-the-art
algorithms in terms of the overall accuracy. Remarkably, the Monte Carlo
method, which was specifically designed for the presence of diverse multiple
routes, often revealed a ranked list of hundreds of reaction routes to the same
synthetic target. We investigated the potential applicability of such diverse
candidates based on expert knowledge from synthetic organic chemistry.
| stat |
Distilling Black-Box Travel Mode Choice Model for Behavioral
Interpretation | Machine learning has proved to be very successful for making predictions in
travel behavior modeling. However, most machine-learning models have complex
model structures and offer little or no explanation as to how they arrive at
these predictions. Interpretations about travel behavior models are essential
for decision makers to understand travelers' preferences and plan policy
interventions accordingly. Therefore, this paper proposes to apply and extend
the model distillation approach, a model-agnostic machine-learning
interpretation method, to explain how a black-box travel mode choice model
makes predictions for the entire population and subpopulations of interest.
Model distillation aims at compressing knowledge from a complex model (teacher)
into an understandable and interpretable model (student). In particular, the
paper integrates model distillation with market segmentation to generate more
insights by accounting for heterogeneity. Furthermore, the paper provides a
comprehensive comparison of student models with the benchmark model (decision
tree) and the teacher model (gradient boosting trees) to quantify the fidelity
and accuracy of the students' interpretations.
| stat |
Multi-level Training and Bayesian Optimization for Economical
Hyperparameter Optimization | Hyperparameters play a critical role in the performances of many machine
learning methods. Determining their best settings or Hyperparameter
Optimization (HPO) faces difficulties presented by the large number of
hyperparameters as well as the excessive training time. In this paper, we
develop an effective approach to reducing the total amount of required training
time for HPO. In the initialization, the nested Latin hypercube design is used
to select hyperparameter configurations for two types of training, which are,
respectively, heavy training and light training. We propose a truncated
additive Gaussian process model to calibrate approximate performance
measurements generated by light training, using accurate performance
measurements generated by heavy training. Based on the model, a sequential
model-based algorithm is developed to generate the performance profile of the
configuration space as well as find optimal ones. Our proposed approach
demonstrates competitive performance when applied to optimize synthetic
examples, support vector machines, fully connected networks and convolutional
neural networks.
| stat |
CS Sparse K-means: An Algorithm for Cluster-Specific Feature Selection
in High-Dimensional Clustering | Feature selection is an important and challenging task in high dimensional
clustering. For example, in genomics, there may only be a small number of genes
that are differentially expressed, which are informative to the overall
clustering structure. Existing feature selection methods, such as Sparse
K-means, rarely tackle the problem of accounting features that can only
separate a subset of clusters. In genomics, it is highly likely that a gene can
only define one subtype against all the other subtypes or distinguish a pair of
subtypes but not others. In this paper, we propose a K-means based clustering
algorithm that discovers informative features as well as which cluster pairs
are separable by each selected features. The method is essentially an EM
algorithm, in which we introduce lasso-type constraints on each cluster pair in
the M step, and make the E step possible by maximizing the raw cross-cluster
distance instead of minimizing the intra-cluster distance. The results were
demonstrated on simulated data and a leukemia gene expression dataset.
| stat |
Multi-subject MEG/EEG source imaging with sparse multi-task regression | Magnetoencephalography and electroencephalography (M/EEG) are non-invasive
modalities that measure the weak electromagnetic fields generated by neural
activity. Estimating the location and magnitude of the current sources that
generated these electromagnetic fields is a challenging ill-posed regression
problem known as \emph{source imaging}. When considering a group study, a
common approach consists in carrying out the regression tasks independently for
each subject. An alternative is to jointly localize sources for all subjects
taken together, while enforcing some similarity between them. By pooling all
measurements in a single multi-task regression, one makes the problem better
posed, offering the ability to identify more sources and with greater
precision. The Minimum Wasserstein Estimates (MWE) promotes focal activations
that do not perfectly overlap for all subjects, thanks to a regularizer based
on Optimal Transport (OT) metrics. MWE promotes spatial proximity on the
cortical mantel while coping with the varying noise levels across subjects. On
realistic simulations, MWE decreases the localization error by up to 4 mm per
source compared to individual solutions. Experiments on the Cam-CAN dataset
show a considerable improvement in spatial specificity in population imaging.
Our analysis of a multimodal dataset shows how multi-subject source
localization closes the gap between MEG and fMRI for brain mapping.
| stat |
Bayesian Doubly Robust Causal Inference via Loss Functions | Frequentist inference has a well-established supporting theory for doubly
robust causal inference based on the potential outcomes framework, which is
realized via outcome regression (OR) and propensity score (PS) models. The
Bayesian counterpart, however, is not obvious as the PS model loses its
balancing property in joint modeling. In this paper, we propose a natural and
formal Bayesian solution by bridging loss-type Bayesian inference with a
utility function derived from the notion of a pseudo-population via the change
of measure. Consistency of the posterior distribution is shown with correctly
specified and misspecified OR models. Simulation studies suggest that our
proposed method can estimate the true causal effect more efficiently and
achieve the frequentist coverage if either the OR model is correctly specified
or fit with a flexible function of the confounders, compared to the previous
Bayesian approach via the Bayesian bootstrap. Finally, we apply this novel
Bayesian method to assess the impact of speed cameras on the reduction of car
collisions in England.
| stat |
Bayesian Multiresolution Modeling Of Georeferenced Data | Current implementations of multiresolution methods are limited in terms of
possible types of responses and approaches to inference. We provide a
multiresolution approach for spatial analysis of non-Gaussian responses using
latent Gaussian models and Bayesian inference via integrated nested Laplace
approximation (INLA). The approach builds on `LatticeKrig', but uses a
reparameterization of the model parameters that is intuitive and interpretable
so that modeling and prior selection can be guided by expert knowledge about
the different spatial scales at which dependence acts. The priors can be used
to make inference robust and integration over model parameters allows for more
accurate posterior estimates of uncertainty. The extended LatticeKrig (ELK)
model is compared to a standard implementation of LatticeKrig (LK), and a
standard Mat\'ern model, and we find modest improvement in spatial
oversmoothing and prediction for the ELK model for counts of secondary
education completion for women in Kenya collected in the 2014 Kenya demographic
health survey. Through a simulation study with Gaussian responses and a
realistic mix of short and long scale dependencies, we demonstrate that the
differences between the three approaches for prediction increases with distance
to nearest observation.
| stat |
DeepClean -- self-supervised artefact rejection for intensive care
waveform data using deep generative learning | Waveform physiological data is important in the treatment of critically ill
patients in the intensive care unit. Such recordings are susceptible to
artefacts, which must be removed before the data can be re-used for alerting or
reprocessed for other clinical or research purposes. Accurate removal of
artefacts reduces bias and uncertainty in clinical assessment, as well as the
false positive rate of intensive care unit alarms, and is therefore a key
component in providing optimal clinical care. In this work, we present
DeepClean; a prototype self-supervised artefact detection system using a
convolutional variational autoencoder deep neural network that avoids costly
and painstaking manual annotation, requiring only easily-obtained 'good' data
for training. For a test case with invasive arterial blood pressure, we
demonstrate that our algorithm can detect the presence of an artefact within a
10-second sample of data with sensitivity and specificity around 90%.
Furthermore, DeepClean was able to identify regions of artefact within such
samples with high accuracy and we show that it significantly outperforms a
baseline principle component analysis approach in both signal reconstruction
and artefact detection. DeepClean learns a generative model and therefore may
also be used for imputation of missing data.
| stat |
Bayesian joint inference for multiple directed acyclic graphs | In many applications, data often arise from multiple groups that may share
similar characteristics. A joint estimation method that models several groups
simultaneously can be more efficient than estimating parameters in each group
separately. We focus on unraveling the dependence structures of data based on
directed acyclic graphs and propose a Bayesian joint inference method for
multiple graphs. To encourage similar dependence structures across all groups,
a Markov random field prior is adopted. We establish the joint selection
consistency of the fractional posterior in high dimensions, and benefits of the
joint inference are shown under the common support assumption. This is the
first Bayesian method for joint estimation of multiple directed acyclic graphs.
The performance of the proposed method is demonstrated using simulation
studies, and it is shown that our joint inference outperforms other
competitors. We apply our method to an fMRI data for simultaneously inferring
multiple brain functional networks.
| stat |
Modelling Competing Legal Arguments using Bayesian Model Comparison and
Averaging | Bayesian models of legal arguments generally aim to produce a single
integrated model, combining each of the legal arguments under consideration.
This combined approach implicitly assumes that variables and their
relationships can be represented without any contradiction or misalignment, and
in a way that makes sense with respect to the competing argument narratives.
This paper describes a novel approach to compare and 'average' Bayesian models
of legal arguments that have been built independently and with no attempt to
make them consistent in terms of variables, causal assumptions or
parametrisation. The approach involves assessing whether competing models of
legal arguments are explained or predict facts uncovered before or during the
trial process. Those models that are more heavily disconfirmed by the facts are
given lower weight, as model plausibility measures, in the Bayesian model
comparison and averaging framework adopted. In this way a plurality of
arguments is allowed yet a single judgement based on all arguments is possible
and rational.
| stat |
Estimations of means and variances in a Markov linear model | Multivariate regression models and ANOVA are probably the most frequently
applied methods of all statistical analyses. We study the case where the
predictors are qualitative variables, and the response variable is
quantitative. In this case, we propose an alternative to the classic approaches
that does not assume homoscedasticity but assumes that a Markov chain can
describe the covariates' correlations. This approach transforms the dependent
covariates using a change of measure to independent covariates. The transformed
estimates allow a pairwise comparison of the mean and variance of the
contribution of different values of the covariates. We show that under standard
moment conditions, the estimators are asymptotically normally distributed. We
test our method with data from simulations and apply it to several classic data
sets.
| stat |
Instrumental Variable Quantile Regression with Misclassification | This paper considers the instrumental variable quantile regression model
(Chernozhukov and Hansen, 2005, 2013) with a binary endogenous treatment. It
offers two identification results when the treatment status is not directly
observed. The first result is that, remarkably, the reduced-form quantile
regression of the outcome variable on the instrumental variable provides a
lower bound on the structural quantile treatment effect under the stochastic
monotonicity condition (Small and Tan, 2007; DiNardo and Lee, 2011). This
result is relevant, not only when the treatment variable is subject to
misclassification, but also when any measurement of the treatment variable is
not available. The second result is for the structural quantile function when
the treatment status is measured with error; I obtain the sharp identified set
by deriving moment conditions under widely-used assumptions on the measurement
error. Furthermore, I propose an inference method in the presence of other
covariates.
| stat |
Disentangling Derivatives, Uncertainty and Error in Gaussian Process
Models | Gaussian Processes (GPs) are a class of kernel methods that have shown to be
very useful in geoscience applications. They are widely used because they are
simple, flexible and provide very accurate estimates for nonlinear problems,
especially in parameter retrieval. An addition to a predictive mean function,
GPs come equipped with a useful property: the predictive variance function
which provides confidence intervals for the predictions. The GP formulation
usually assumes that there is no input noise in the training and testing
points, only in the observations. However, this is often not the case in Earth
observation problems where an accurate assessment of the instrument error is
usually available. In this paper, we showcase how the derivative of a GP model
can be used to provide an analytical error propagation formulation and we
analyze the predictive variance and the propagated error terms in a temperature
prediction problem from infrared sounding data.
| stat |
Gaussian process regression for forest attribute estimation from
airborne laser scanning data | While the analysis of airborne laser scanning (ALS) data often provides
reliable estimates for certain forest stand attributes -- such as total volume
or basal area -- there is still room for improvement, especially in estimating
species-specific attributes. Moreover, while information on the estimate
uncertainty would be useful in various economic and environmental analyses on
forests, a computationally feasible framework for uncertainty quantifying in
ALS is still missing. In this article, the species-specific stand attribute
estimation and uncertainty quantification (UQ) is approached using Gaussian
process regression (GPR), which is a nonlinear and nonparametric machine
learning method. Multiple species-specific stand attributes are estimated
simultaneously: tree height, stem diameter, stem number, basal area, and stem
volume. The cross-validation results show that GPR yields on average an
improvement of 4.6\% in estimate RMSE over a state-of-the-art k-nearest
neighbors (kNN) implementation, negligible bias and well performing UQ
(credible intervals), while being computationally fast. The performance
advantage over kNN and the feasibility of credible intervals persists even when
smaller training sets are used.
| stat |
Reproducible Research: A Retrospective | Rapid advances in computing technology over the past few decades have spurred
two extraordinary phenomena in science: large-scale and high-throughput data
collection coupled with the creation and implementation of complex statistical
algorithms for data analysis. Together, these two phenomena have brought about
tremendous advances in scientific discovery but have also raised two serious
concerns, one relatively new and one quite familiar. The complexity of modern
data analyses raises questions about the reproducibility of the analyses,
meaning the ability of independent analysts to re-create the results claimed by
the original authors using the original data and analysis techniques. While
seemingly a straightforward concept, reproducibility of analyses is typically
thwarted by the lack of availability of the data and computer code that were
used in the analyses. A much more general concern is the replicability of
scientific findings, which concerns the frequency with which scientific claims
are confirmed by completely independent investigations. While the concepts of
reproduciblity and replicability are related, it is worth noting that they are
focused on quite different goals and address different aspects of scientific
progress. In this review, we will discuss the origins of reproducible research,
characterize the current status of reproduciblity in public health research,
and connect reproduciblity to current concerns about replicability of
scientific findings. Finally, we describe a path forward for improving both the
reproducibility and replicability of public health research in the future.
| stat |
Hierarchical network design for nitrogen dioxide measurement in urban
environments, part 2: network-based sensor calibration | We present a management and data correction framework for low-cost
electrochemical sensors for nitrogen dioxide (NO2) deployed within a
hierarchical network of low-cost and regulatory-grade instruments. The
framework is founded on the idea that it is possible in a suitably configured
network to identify a source of reliable proxy data for each sensor site that
has a similar probability distribution of measurement values over a suitable
time period. Previous work successfully applied these ideas to a sensor system
with a simple linear 2-parameter (slope and offset) response. Applying these
ideas to electrochemical sensors for NO2 presents significant additional
difficulties for which we demonstrate solutions. The three NO2 sensor response
parameters (offset, ozone (O3) response slope, and NO2 response slope) are
known to vary significantly as a consequence of ambient humidity and
temperature variations. Here we demonstrate that these response parameters can
be estimated by minimising the Kullback-Leibler divergence between
sensor-estimated and proxy NO2 distributions over a 3-day window. We then
estimate an additional offset term by using co-location data. This offset term
is dependent on climate and spatially correlated and can thus be projected
across the network. Co-location data also estimates the time-, space- and
concentration-dependent error distribution between sensors and regulatory-grade
instruments. We show how the parameter variations can be used to indicate both
sensor failure and failure of the proxy assumption. We apply the procedures to
a network of 56 sensors distributed across the Inland Empire and Los Angeles
County regions, demonstrating the need for reliable data from dense networks of
monitors to supplement the existing regulatory networks.
| stat |
Scalable changepoint and anomaly detection in cross-correlated data with
an application to condition monitoring | Motivated by a condition monitoring application arising from subsea
engineering we derive a novel, scalable approach to detecting anomalous mean
structure in a subset of correlated multivariate time series. Given the need to
analyse such series efficiently we explore a computationally efficient
approximation of the maximum likelihood solution to the resulting modelling
framework, and develop a new dynamic programming algorithm for solving the
resulting Binary Quadratic Programme when the precision matrix of the time
series at any given time-point is banded. Through a comprehensive simulation
study, we show that the resulting methods perform favourably compared to
competing methods both in the anomaly and change detection settings, even when
the sparsity structure of the precision matrix estimate is misspecified. We
also demonstrate its ability to correctly detect faulty time-periods of a pump
within the motivating application.
| stat |
A High-dimensional M-estimator Framework for Bi-level Variable Selection | In high-dimensional data analysis, bi-level sparsity is often assumed when
covariates function group-wisely and sparsity can appear either at the group
level or within certain groups. In such case, an ideal model should be able to
encourage the bi-level variable selection consistently. The bi-level variable
selection has become even more challenging when data have heavy-tailed
distribution or outliers exist in random errors and covariates. In this paper,
we study a framework of high-dimensional M-estimation for bi-level variable
selection. This framework encourages bi-level sparsity through a
computationally efficient two-stage procedure. It produces strong robust
parameter estimators if nonconvex redescending loss functions are applied. In
theory, we provide sufficient conditions under which our two-stage penalized
M-estimator possesses simultaneous local estimation consistency and the
bi-level variable selection consistency if certain nonconvex penalty functions
are used at the group level. Both our simulation studies and real data analysis
demonstrate satisfactory finite sample performance of the proposed estimators
under different irregular settings.
| stat |
A new correlation coefficient between categorical, ordinal and interval
variables with Pearson characteristics | A prescription is presented for a new and practical correlation coefficient,
$\phi_K$, based on several refinements to Pearson's hypothesis test of
independence of two variables. The combined features of $\phi_K$ form an
advantage over existing coefficients. First, it works consistently between
categorical, ordinal and interval variables. Second, it captures non-linear
dependency. Third, it reverts to the Pearson correlation coefficient in case of
a bi-variate normal input distribution. These are useful features when studying
the correlation between variables with mixed types. Particular emphasis is paid
to the proper evaluation of statistical significance of correlations and to the
interpretation of variable relationships in a contingency table, in particular
in case of low statistics samples and significant dependencies. Three practical
applications are discussed. The presented algorithms are easy to use and
available through a public Python library.
| stat |
The performance of the partially overlapping samples t-tests at the
limits | This note includes an assessment of the partially overlapping samples t-tests
in scenarios where elements of the test are at their extremes, including where
only one sample contains independent observations
| stat |
Expected Hypothetical Completion Probability | Using high-resolution player tracking data made available by the National
Football League (NFL) for their 2019 Big Data Bowl competition, we introduce
the Expected Hypothetical Completion Probability (EHCP), a objective framework
for evaluating plays. At the heart of EHCP is the question "on a given passing
play, did the quarterback throw the pass to the receiver who was most likely to
catch it?" To answer this question, we first built a Bayesian non-parametric
catch probability model that automatically accounts for complex interactions
between inputs like the receiver's speed and distances to the ball and nearest
defender. While building such a model is, in principle, straightforward, using
it to reason about a hypothetical pass is challenging because many of the model
inputs corresponding to a hypothetical are necessarily unobserved. To wit, it
is impossible to observe how close an un-targeted receiver would be to his
nearest defender had the pass been thrown to him instead of the receiver who
was actually targeted. To overcome this fundamental difficulty, we propose
imputing the unobservable inputs and averaging our model predictions across
these imputations to derive EHCP. In this way, EHCP can track how the
completion probability evolves for each receiver over the course of a play in a
way that accounts for the uncertainty about missing inputs.
| stat |
On Information Gain and Regret Bounds in Gaussian Process Bandits | Consider the sequential optimization of an expensive to evaluate and possibly
non-convex objective function $f$ from noisy feedback, that can be considered
as a continuum-armed bandit problem. Upper bounds on the regret performance of
several learning algorithms (GP-UCB, GP-TS, and their variants) are known under
both a Bayesian (when $f$ is a sample from a Gaussian process (GP)) and a
frequentist (when $f$ lives in a reproducing kernel Hilbert space) setting. The
regret bounds often rely on the maximal information gain $\gamma_T$ between $T$
observations and the underlying GP (surrogate) model. We provide general bounds
on $\gamma_T$ based on the decay rate of the eigenvalues of the GP kernel,
whose specialisation for commonly used kernels, improves the existing bounds on
$\gamma_T$, and subsequently the regret bounds relying on $\gamma_T$ under
numerous settings. For the Mat\'ern family of kernels, where the lower bounds
on $\gamma_T$, and regret under the frequentist setting, are known, our results
close a huge polynomial in $T$ gap between the upper and lower bounds (up to
logarithmic in $T$ factors).
| stat |
Limited Functional Form, Misspecification, and Unreliable
Interpretations in Psychology and Social Science | The replicability crisis has drawn attention to numerous weaknesses in
psychology and social science research practice. In this work we focus on three
issues that deserve more attention: The use of models with limited functional
form, the use of misspecified causal models, and unreliable interpretation of
results. We demonstrate a number of possible consequences via simulation, and
provide recommendations for researchers to improve their research practice. We
believe it is extremely important to encourage psychologists and social
scientists to engage with the debate surrounding areas of possible analytical
and statistical improvements, particularly given that these shortfalls have the
potential to seriously hinder scientific progress. Every research question and
hypothesis may present its own unique challenges, and it is only through an
awareness and understanding of varied statistical methods for predictive and
causal modeling, that researchers will have the tools with which to
appropriately address them.
| stat |
Modeling Variables with a Detection Limit using a Truncated Normal
Distribution with Censoring | When data are collected subject to a detection limit, observations below the
detection limit may be considered censored. In addition, the domain of such
observations may be restricted; for example, values may be required to be
non-negative. We propose a regression method for censored observations that
also accounts for domain restriction. The method finds maximum likelihood
estimates assuming an underlying truncated normal distribution. We show that
our method, tcensReg, outperforms other methods commonly used for data with
detection limits such as Tobit regression and single imputation of the
detection limit or half detection limit with respect to bias and mean squared
error under a range of simulation settings. We apply our method to analyze
vision quality data collected from ophthalmology clinical trials comparing
different types of intraocular lenses implanted during cataract surgery.
| stat |
A continuation method in Bayesian inference | We present a continuation method that entails generating a sequence of
transition probability density functions from the prior to the posterior in the
context of Bayesian inference for parameter estimation problems. The
characterization of transition distributions, by tempering the likelihood
function, results in a homogeneous nonlinear partial integro-differential
equation whose existence and uniqueness of solutions are addressed. The
posterior probability distribution comes as the interpretation of the final
state of the path of transition distributions. A computationally stable scaling
domain for the likelihood is explored for the approximation of the expected
deviance, where we manage to hold back all the evaluations of the forward
predictive model at the prior stage. It follows the computational tractability
of the posterior distribution and opens access to the posterior distribution
for direct samplings. To get a solution formulation of the expected deviance,
we derive a partial differential equation governing the moments generating
function of the log-likelihood. We show also that a spectral formulation of the
expected deviance can be obtained for low-dimensional problems under certain
conditions. The computational efficiency of the proposed method is demonstrated
through three differents numerical examples that focus on analyzing the
computational bias generated by the method, assessing the continuation method
in the Bayesian inference with non-Gaussian noise, and evaluating its ability
to invert a multimodal parameter of interest.
| stat |
Methods for Population Adjustment with Limited Access to Individual
Patient Data: A Review and Simulation Study | Population-adjusted indirect comparisons estimate treatment effects when
access to individual patient data is limited and there are cross-trial
differences in effect modifiers. Popular methods include matching-adjusted
indirect comparison (MAIC) and simulated treatment comparison (STC). There is
limited formal evaluation of these methods and whether they can be used to
accurately compare treatments. Thus, we undertake a comprehensive simulation
study to compare standard unadjusted indirect comparisons, MAIC and STC across
162 scenarios. This simulation study assumes that the trials are investigating
survival outcomes and measure continuous covariates, with the log hazard ratio
as the measure of effect. MAIC yields unbiased treatment effect estimates under
no failures of assumptions. The typical usage of STC produces bias because it
targets a conditional treatment effect where the target estimand should be a
marginal treatment effect. The incompatibility of estimates in the indirect
comparison leads to bias as the measure of effect is non-collapsible. Standard
indirect comparisons are systematically biased, particularly under stronger
covariate imbalance and interaction effects. Standard errors and coverage rates
are often valid in MAIC but the robust sandwich variance estimator
underestimates variability where effective sample sizes are small. Interval
estimates for the standard indirect comparison are too narrow and STC suffers
from bias-induced undercoverage. MAIC provides the most accurate estimates and,
with lower degrees of covariate overlap, its bias reduction outweighs the loss
in effective sample size and precision under no failures of assumptions. An
important future objective is the development of an alternative formulation to
STC that targets a marginal treatment effect.
| stat |
Causal structure learning from time series: Large regression
coefficients may predict causal links better in practice than small p-values | In this article, we describe the algorithms for causal structure learning
from time series data that won the Causality 4 Climate competition at the
Conference on Neural Information Processing Systems 2019 (NeurIPS). We examine
how our combination of established ideas achieves competitive performance on
semi-realistic and realistic time series data exhibiting common challenges in
real-world Earth sciences data. In particular, we discuss a) a rationale for
leveraging linear methods to identify causal links in non-linear systems, b) a
simulation-backed explanation as to why large regression coefficients may
predict causal links better in practice than small p-values and thus why
normalising the data may sometimes hinder causal structure learning.
For benchmark usage, we detail the algorithms here and provide
implementations at https://github.com/sweichwald/tidybench . We propose the
presented competition-proven methods for baseline benchmark comparisons to
guide the development of novel algorithms for structure learning from time
series.
| stat |
A Bayesian Approach to Directed Acyclic Graphs with a Candidate Graph | Directed acyclic graphs represent the dependence structure among variables.
When learning these graphs from data, different amounts of information may be
available for different edges. Although many methods have been developed to
learn the topology of these graphs, most of them do not provide a measure of
uncertainty in the inference. We propose a Bayesian method, baycn (BAYesian
Causal Network), to estimate the posterior probability of three states for each
edge: present with one direction ($X \rightarrow Y$), present with the opposite
direction ($X \leftarrow Y$), and absent. Unlike existing Bayesian methods, our
method requires that the prior probabilities of these states be specified, and
therefore provides a benchmark for interpreting the posterior probabilities. We
develop a fast Metropolis-Hastings Markov chain Monte Carlo algorithm for the
inference. Our algorithm takes as input the edges of a candidate graph, which
may be the output of another graph inference method and may contain false
edges. In simulation studies our method achieves high accuracy with small
variation across different scenarios and is comparable or better than existing
Bayesian methods. We apply baycn to genomic data to distinguish the direct and
indirect targets of genetic variants.
| stat |
Robust Group Synchronization via Cycle-Edge Message Passing | We propose a general framework for group synchronization with adversarial
corruption and sufficiently small noise. Specifically, we apply a novel message
passing procedure that uses cycle consistency information in order to estimate
the corruption levels of group ratios and consequently infer the corrupted
group ratios and solve the synchronization problem. We first explain why the
group cycle consistency information is essential for effectively solving group
synchronization problems. We then establish exact recovery and linear
convergence guarantees for the proposed message passing procedure under a
deterministic setting with adversarial corruption. These guarantees hold as
long as the ratio of corrupted cycles per edge is bounded by a reasonable
constant. We also establish the stability of the proposed procedure to
sub-Gaussian noise. We further show that under a uniform corruption model, the
recovery results are sharp in terms of an information-theoretic bound.
| stat |
Impact of model misspecification in shared frailty survival models | Survival models incorporating random effects to account for unmeasured
heterogeneity are being increasingly used in biostatistical and applied
research. Specifically, unmeasured covariates whose lack of inclusion in the
model would lead to biased, inefficient results are commonly modelled by
including a subject-specific (or cluster-specific) frailty term that follows a
given distribution (e.g. Gamma or log-Normal). Despite that, in the context of
parametric frailty models little is known about the impact of misspecifying the
baseline hazard, the frailty distribution, or both. Therefore, our aim is to
quantify the impact of such misspecification in a wide variety of clinically
plausible scenarios via Monte Carlo simulation, using open source software
readily available to applied researchers. We generate clustered survival data
assuming various baseline hazard functions, including mixture distributions
with turning points, and assess the impact of sample size, variance of the
frailty, baseline hazard function, and frailty distribution. Models compared
include standard parametric distributions and more flexible spline-based
approaches; we also included semiparametric Cox models. The resulting bias can
be clinically relevant. In conclusion, we highlight the importance of fitting
models that are flexible enough and the importance of assessing model fit. We
illustrate our conclusions with two applications using data on diabetic
retinopathy and bladder cancer. Our results show the importance of assessing
model fit with respect to the baseline hazard function and the distribution of
the frailty: misspecifying the former leads to biased relative and absolute
risk estimates while misspecifying the latter affects absolute risk estimates
and measures of heterogeneity.
| stat |
Which Factorization Machine Modeling is Better: A Theoretical Answer
with Optimal Guarantee | Factorization machine (FM) is a popular machine learning model to capture the
second order feature interactions. The optimal learning guarantee of FM and its
generalized version is not yet developed. For a rank $k$ generalized FM of $d$
dimensional input, the previous best known sampling complexity is
$\mathcal{O}[k^{3}d\cdot\mathrm{polylog}(kd)]$ under Gaussian distribution.
This bound is sub-optimal comparing to the information theoretical lower bound
$\mathcal{O}(kd)$. In this work, we aim to tighten this bound towards optimal
and generalize the analysis to sub-gaussian distribution. We prove that when
the input data satisfies the so-called $\tau$-Moment Invertible Property, the
sampling complexity of generalized FM can be improved to
$\mathcal{O}[k^{2}d\cdot\mathrm{polylog}(kd)/\tau^{2}]$. When the second order
self-interaction terms are excluded in the generalized FM, the bound can be
improved to the optimal $\mathcal{O}[kd\cdot\mathrm{polylog}(kd)]$ up to the
logarithmic factors. Our analysis also suggests that the positive semi-definite
constraint in the conventional FM is redundant as it does not improve the
sampling complexity while making the model difficult to optimize. We evaluate
our improved FM model in real-time high precision GPS signal calibration task
to validate its superiority.
| stat |
Mean absolute deviations about the mean, the cut norm and taxicab
correspondence analysis | Optimization has two faces, minimization of a loss function or maximization
of a gain function. We show that the mean absolute deviations about the mean,
d, maximizes a gain function based on the power set of the individuals, and it
is equal to twice the value of its cut-norm. This property is generalized to
double-centered and triple-centered data sets. Furthermore, we show that among
the three well known dispersion measures, standard deviation, least absolute
deviation and d, d is the most robust based on the relative contribution
criterion. More importantly, we show that the computation of each principal
dimension of taxicab correspondence analysis corresponds to balanced 2-blocks
seriation. Examples are provided.
| stat |
Dynamics of ternary statistical experiments with equilibrium state | We study the scenarios of the dynamics of ternary statistical experiments,
modeled employing difference equations. The important features are a balance
condition and the existence of a steady-state (equilibrium). We give a
classification of scenarios of the model evolution which are significantly
different between them, depending on the domain of the values of the model
basic parameters.
| stat |
On Determining the Distribution of a Goodness-of-Fit Test Statistic | We consider the problem of goodness-of-fit testing for a model that has at
least one unknown parameter that cannot be eliminated by transformation.
Examples of such problems can be as simple as testing whether a sample consists
of independent Gamma observations, or whether a sample consists of independent
Generalised Pareto observations given a threshold. Over time the approach to
determining the distribution of a test statistic for such a problem has moved
towards on-the-fly calculation post observing a sample. Modern approaches
include the parametric bootstrap and posterior predictive checks. We argue that
these approaches are merely approximations to integrating over the posterior
predictive distribution that flows naturally from a given model. Further, we
attempt to demonstrate that shortcomings which may be present in the parametric
bootstrap, especially in small samples, can be reduced through the use of
objective Bayes techniques, in order to more reliably produce a test with the
correct size.
| stat |
Scalable modeling of nonstationary covariance functions with non-folding
B-spline deformation | We propose a method for nonstationary covariance function modeling, based on
the spatial deformation method of Sampson and Guttorp [1992], but using a
low-rank, scalable deformation function written as a linear combination of the
tensor product of B-spline basis. This approach addresses two important
weaknesses in current computational aspects. First, it allows one to constrain
estimated 2D deformations to be non-folding (bijective) in 2D. This requirement
of the model has, up to now,been addressed only by arbitrary levels of spatial
smoothing. Second, basis functions with compact support enable the application
to large datasets of spatial monitoring sites of environmental data. An
application to rainfall data in southeastern Brazil illustrates the method
| stat |
A simple yet efficient algorithm for multiple kernel learning under
elastic-net constraints | This papers introduces an algorithm for the solution of multiple kernel
learning (MKL) problems with elastic-net constraints on the kernel weights. The
algorithm compares very favourably in terms of time and space complexity to
existing approaches and can be implemented with simple code that does not rely
on external libraries (except a conventional SVM solver).
| stat |
On the Ergodicity, Bias and Asymptotic Normality of Randomized Midpoint
Sampling Method | The randomized midpoint method, proposed by [SL19], has emerged as an optimal
discretization procedure for simulating the continuous time Langevin
diffusions. Focusing on the case of strong-convex and smooth potentials, in
this paper, we analyze several probabilistic properties of the randomized
midpoint discretization method for both overdamped and underdamped Langevin
diffusions. We first characterize the stationary distribution of the discrete
chain obtained with constant step-size discretization and show that it is
biased away from the target distribution. Notably, the step-size needs to go to
zero to obtain asymptotic unbiasedness. Next, we establish the asymptotic
normality for numerical integration using the randomized midpoint method and
highlight the relative advantages and disadvantages over other discretizations.
Our results collectively provide several insights into the behavior of the
randomized midpoint discretization method, including obtaining confidence
intervals for numerical integrations.
| stat |
A post hoc test on the Sharpe ratio | We describe a post hoc test for the Sharpe ratio, analogous to Tukey's test
for pairwise equality of means. The test can be applied after rejection of the
hypothesis that all population Signal-Noise ratios are equal. The test is
applicable under a simple correlation structure among asset returns.
Simulations indicate the test maintains nominal type I rate under a wide range
of conditions and is moderately powerful under reasonable alternatives.
| stat |
Zeroth-Order Algorithms for Nonconvex Minimax Problems with Improved
Complexities | In this paper, we study zeroth-order algorithms for minimax optimization
problems that are nonconvex in one variable and strongly-concave in the other
variable. Such minimax optimization problems have attracted significant
attention lately due to their applications in modern machine learning tasks. We
first design and analyze the Zeroth-Order Gradient Descent Ascent
(\texttt{ZO-GDA}) algorithm, and provide improved results compared to existing
works, in terms of oracle complexity. Next, we propose the Zeroth-Order
Gradient Descent Multi-Step Ascent (\texttt{ZO-GDMSA}) algorithm that
significantly improves the oracle complexity of \texttt{ZO-GDA}. We also
provide stochastic version of \texttt{ZO-GDA} and \texttt{ZO-GDMSA} to handle
stochastic nonconvex minimax problems, and provide oracle complexity results.
| stat |
Locally Optimized Random Forests | Standard supervised learning procedures are validated against a test set that
is assumed to have come from the same distribution as the training data.
However, in many problems, the test data may have come from a different
distribution. We consider the case of having many labeled observations from one
distribution, $P_1$, and making predictions at unlabeled points that come from
$P_2$. We combine the high predictive accuracy of random forests (Breiman,
2001) with an importance sampling scheme, where the splits and predictions of
the base-trees are done in a weighted manner, which we call Locally Optimized
Random Forests. These weights correspond to a non-parametric estimate of the
likelihood ratio between the training and test distributions. To estimate these
ratios with an unlabeled test set, we make the covariate shift assumption,
where the differences in distribution are only a function of the training
distributions (Shimodaira, 2000.) This methodology is motivated by the problem
of forecasting power outages during hurricanes. The extreme nature of the most
devastating hurricanes means that typical validation set ups will overly favor
less extreme storms. Our method provides a data-driven means of adapting a
machine learning method to deal with extreme events.
| stat |
Improving precipitation forecasts using extreme quantile regression | Aiming to estimate extreme precipitation forecast quantiles, we propose a
nonparametric regression model that features a constant extreme value index.
Using local linear quantile regression and an extrapolation technique from
extreme value theory, we develop an estimator for conditional quantiles
corresponding to extreme high probability levels. We establish uniform
consistency and asymptotic normality of the estimators. In a simulation study,
we examine the performance of our estimator on finite samples in comparison
with a method assuming linear quantiles. On a precipitation data set in the
Netherlands, these estimators have greater predictive skill compared to the
upper member of ensemble forecasts provided by a numerical weather prediction
model.
| stat |
Bayesian Poisson Log-normal Model with Regularized Time Structure for
Mortality Projection of Multi-population | The improvement of mortality projection is a pivotal topic in the diverse
branches related to insurance, demography, and public policy. Motivated by the
thread of Lee-Carter related models, we propose a Bayesian model to estimate
and predict mortality rates for multi-population. This new model features in
information borrowing among populations and properly reflecting variations of
data. It also provides a solution to a long-time overlooked problem: model
selection for dependence structures of population-specific time parameters. By
introducing a Dirac spike function, simultaneous model selection and estimation
for population-specific time effects can be achieved without much extra
computation cost. We use the Japanese mortality data from Human Mortality
Database to illustrate the desirable properties of our model.
| stat |
Diagnostic checking in FARIMA models with uncorrelated but
non-independent error terms | This work considers the problem of modified portmanteau tests for testing the
adequacy of FARIMA models under the assumption that the errors are uncorrelated
but not necessarily independent (i.e. weak FARIMA). We first study the joint
distribution of the least squares estimator and the noise empirical
autocovariances. We then derive the asymp-totic distribution of residual
empirical autocovariances and autocorrelations. We deduce the asymptotic
distribution of the Ljung-Box (or Box-Pierce) modified portmanteau statistics
for weak FARIMA models. We also propose another method based on a
self-normalization approach to test the adequacy of FARIMA models. Finally some
simulation studies are presented to corroborate our theoretical work. An
application to the Standard \& Poor's 500 and Nikkei returns also illustrate
the practical relevance of our theoretical results. AMS 2000 subject
classifications: Primary 62M10, 62F03, 62F05; secondary 91B84, 62P05.
| stat |
Hyperparameter Selection for Subsampling Bootstraps | Massive data analysis becomes increasingly prevalent, subsampling methods
like BLB (Bag of Little Bootstraps) serves as powerful tools for assessing the
quality of estimators for massive data. However, the performance of the
subsampling methods are highly influenced by the selection of tuning parameters
( e.g., the subset size, number of resamples per subset ). In this article we
develop a hyperparameter selection methodology, which can be used to select
tuning parameters for subsampling methods. Specifically, by a careful
theoretical analysis, we find an analytically simple and elegant relationship
between the asymptotic efficiency of various subsampling estimators and their
hyperparameters. This leads to an optimal choice of the hyperparameters. More
specifically, for an arbitrarily specified hyperparameter set, we can improve
it to be a new set of hyperparameters with no extra CPU time cost, but the
resulting estimator's statistical efficiency can be much improved. Both
simulation studies and real data analysis demonstrate the superior advantage of
our method.
| stat |
Fast and Robust Shortest Paths on Manifolds Learned from Data | We propose a fast, simple and robust algorithm for computing shortest paths
and distances on Riemannian manifolds learned from data. This amounts to
solving a system of ordinary differential equations (ODEs) subject to boundary
conditions. Here standard solvers perform poorly because they require
well-behaved Jacobians of the ODE, and usually, manifolds learned from data
imply unstable and ill-conditioned Jacobians. Instead, we propose a fixed-point
iteration scheme for solving the ODE that avoids Jacobians. This enhances the
stability of the solver, while reduces the computational cost. In experiments
involving both Riemannian metric learning and deep generative models we
demonstrate significant improvements in speed and stability over both
general-purpose state-of-the-art solvers as well as over specialized solvers.
| stat |
Sparse Parallel Training of Hierarchical Dirichlet Process Topic Models | To scale non-parametric extensions of probabilistic topic models such as
Latent Dirichlet allocation to larger data sets, practitioners rely
increasingly on parallel and distributed systems. In this work, we study
data-parallel training for the hierarchical Dirichlet process (HDP) topic
model. Based upon a representation of certain conditional distributions within
an HDP, we propose a doubly sparse data-parallel sampler for the HDP topic
model. This sampler utilizes all available sources of sparsity found in natural
language - an important way to make computation efficient. We benchmark our
method on a well-known corpus (PubMed) with 8m documents and 768m tokens, using
a single multi-core machine in under four days.
| stat |
Uncertainty Quantification and Deep Ensembles | Deep Learning methods are known to suffer from calibration issues: they
typically produce over-confident estimates. These problems are exacerbated in
the low data regime. Although the calibration of probabilistic models is well
studied, calibrating extremely over-parametrized models in the low-data regime
presents unique challenges. We show that deep-ensembles do not necessarily lead
to improved calibration properties. In fact, we show that standard ensembling
methods, when used in conjunction with modern techniques such as mixup
regularization, can lead to less calibrated models. In this text, we examine
the interplay between three of the most simple and commonly used approaches to
leverage deep learning when data is scarce: data-augmentation, ensembling, and
post-processing calibration methods. We demonstrate that, although standard
ensembling techniques certainly help to boost accuracy, the calibration of
deep-ensembles relies on subtle trade-offs. Our main finding is that
calibration methods such as temperature scaling need to be slightly tweaked
when used with deep-ensembles and, crucially, need to be executed after the
averaging process. Our simulations indicate that, in the low data regime, this
simple strategy can halve the Expected Calibration Error (ECE) on a range of
benchmark classification problems when compared to standard deep-ensembles.
| stat |
Banded Matrix Operators for Gaussian Markov Models in the Automatic
Differentiation Era | Banded matrices can be used as precision matrices in several models including
linear state-space models, some Gaussian processes, and Gaussian Markov random
fields. The aim of the paper is to make modern inference methods (such as
variational inference or gradient-based sampling) available for Gaussian models
with banded precision. We show that this can efficiently be achieved by
equipping an automatic differentiation framework, such as TensorFlow or
PyTorch, with some linear algebra operators dedicated to banded matrices. This
paper studies the algorithmic aspects of the required operators, details their
reverse-mode derivatives, and show that their complexity is linear in the
number of observations.
| stat |
The P-LOOP Estimator: Covariate Adjustment for Paired Experiments | In paired experiments, participants are grouped into pairs with similar
characteristics, and one observation from each pair is randomly assigned to
treatment. Because of both the pairing and the randomization, the treatment and
control groups should be well balanced; however, there may still be small
chance imbalances. It may be possible to improve the precision of the treatment
effect estimate by adjusting for these imbalances. Building on related work for
completely randomized experiments, we propose the P-LOOP (paired leave-one-out
potential outcomes) estimator for paired experiments. We leave out each pair
and then impute its potential outcomes using any prediction algorithm. The
imputation method is flexible; for example, we could use lasso or random
forests. While similar methods exist for completely randomized experiments,
covariate adjustment methods in paired experiments are relatively understudied.
A unique trade-off exists for paired experiments, where it can be unclear
whether to factor in pair assignments when making adjustments. We address this
issue in the P-LOOP estimator by automatically deciding whether to account for
the pairing when imputing the potential outcomes. By addressing this trade-off,
the method has the potential to improve precision over existing methods.
| stat |
The Winning Solution to the IEEE CIG 2017 Game Data Mining Competition | Machine learning competitions such as those organized by Kaggle or KDD
represent a useful benchmark for data science research. In this work, we
present our winning solution to the Game Data Mining competition hosted at the
2017 IEEE Conference on Computational Intelligence and Games (CIG 2017). The
contest consisted of two tracks, and participants (more than 250, belonging to
both industry and academia) were to predict which players would stop playing
the game, as well as their remaining lifetime. The data were provided by a
major worldwide video game company, NCSoft, and came from their successful
massively multiplayer online game Blade and Soul. Here, we describe the long
short-term memory approach and conditional inference survival ensemble model
that made us win both tracks of the contest, as well as the validation
procedure that we followed in order to prevent overfitting. In particular,
choosing a survival method able to deal with censored data was crucial to
accurately predict the moment in which each player would leave the game, as
censoring is inherent in churn. The selected models proved to be robust against
evolving conditions---since there was a change in the business model of the
game (from subscription-based to free-to-play) between the two sample datasets
provided---and efficient in terms of time cost. Thanks to these features and
also to their a ability to scale to large datasets, our models could be readily
implemented in real business settings.
| stat |
Monotonic Alpha-divergence Minimisation | In this paper, we introduce a novel iterative algorithm which carries out
$\alpha$-divergence minimisation by ensuring a systematic decrease in the
$\alpha$-divergence at each step. In its most general form, our framework
allows us to simultaneously optimise the weights and components parameters of a
given mixture model. Notably, our approach permits to build on various methods
previously proposed for $\alpha$-divergence minimisation such as gradient or
power descent schemes. Furthermore, we shed a new light on an integrated
Expectation Maximization algorithm. We provide empirical evidence that our
methodology yields improved results, all the while illustrating the numerical
benefits of having introduced some flexibility through the parameter $\alpha$
of the $\alpha$-divergence.
| stat |
Probabilistic Riemannian submanifold learning with wrapped Gaussian
process latent variable models | Latent variable models (LVMs) learn probabilistic models of data manifolds
lying in an \emph{ambient} Euclidean space. In a number of applications, a
priori known spatial constraints can shrink the ambient space into a
considerably smaller manifold. Additionally, in these applications the
Euclidean geometry might induce a suboptimal similarity measure, which could be
improved by choosing a different metric. Euclidean models ignore such
information and assign probability mass to data points that can never appear as
data, and vastly different likelihoods to points that are similar under the
desired metric. We propose the wrapped Gaussian process latent variable model
(WGPLVM), that extends Gaussian process latent variable models to take values
strictly on a given ambient Riemannian manifold, making the model blind to
impossible data points. This allows non-linear, probabilistic inference of
low-dimensional Riemannian submanifolds from data. Our evaluation on diverse
datasets show that we improve performance on several tasks, including encoding,
visualization and uncertainty quantification.
| stat |
Consistent regression of biophysical parameters with kernel methods | This paper introduces a novel statistical regression framework that allows
the incorporation of consistency constraints. A linear and nonlinear
(kernel-based) formulation are introduced, and both imply closed-form
analytical solutions. The models exploit all the information from a set of
drivers while being maximally independent of a set of auxiliary, protected
variables. We successfully illustrate the performance in the estimation of
chlorophyll content.
| stat |
Escaping the Curse of Dimensionality in Similarity Learning: Efficient
Frank-Wolfe Algorithm and Generalization Bounds | Similarity and metric learning provides a principled approach to construct a
task-specific similarity from weakly supervised data. However, these methods
are subject to the curse of dimensionality: as the number of features grows
large, poor generalization is to be expected and training becomes intractable
due to high computational and memory costs. In this paper, we propose a
similarity learning method that can efficiently deal with high-dimensional
sparse data. This is achieved through a parameterization of similarity
functions by convex combinations of sparse rank-one matrices, together with the
use of a greedy approximate Frank-Wolfe algorithm which provides an efficient
way to control the number of active features. We show that the convergence rate
of the algorithm, as well as its time and memory complexity, are independent of
the data dimension. We further provide a theoretical justification of our
modeling choices through an analysis of the generalization error, which depends
logarithmically on the sparsity of the solution rather than on the number of
features. Our experiments on datasets with up to one million features
demonstrate the ability of our approach to generalize well despite the high
dimensionality as well as its superiority compared to several competing
methods.
| stat |
Scalable Data Augmentation for Deep Learning | Scalable Data Augmentation (SDA) provides a framework for training deep
learning models using auxiliary hidden layers. Scalable MCMC is available for
network training and inference. SDA provides a number of computational
advantages over traditional algorithms, such as avoiding backtracking, local
modes and can perform optimization with stochastic gradient descent (SGD) in
TensorFlow. Standard deep neural networks with logit, ReLU and SVM activation
functions are straightforward to implement. To illustrate our architectures and
methodology, we use P\'{o}lya-Gamma logit data augmentation for a number of
standard datasets. Finally, we conclude with directions for future research.
| stat |
Beyond Perturbation Stability: LP Recovery Guarantees for MAP Inference
on Noisy Stable Instances | Several works have shown that perturbation stable instances of the MAP
inference problem in Potts models can be solved exactly using a natural linear
programming (LP) relaxation. However, most of these works give few (or no)
guarantees for the LP solutions on instances that do not satisfy the relatively
strict perturbation stability definitions. In this work, we go beyond these
stability results by showing that the LP approximately recovers the MAP
solution of a stable instance even after the instance is corrupted by noise.
This "noisy stable" model realistically fits with practical MAP inference
problems: we design an algorithm for finding "close" stable instances, and show
that several real-world instances from computer vision have nearby instances
that are perturbation stable. These results suggest a new theoretical
explanation for the excellent performance of this LP relaxation in practice.
| stat |
Fast Generating A Large Number of Gumbel-Max Variables | The well-known Gumbel-Max Trick for sampling elements from a categorical
distribution (or more generally a nonnegative vector) and its variants have
been widely used in areas such as machine learning and information retrieval.
To sample a random element $i$ (or a Gumbel-Max variable $i$) in proportion to
its positive weight $v_i$, the Gumbel-Max Trick first computes a Gumbel random
variable $g_i$ for each positive weight element $i$, and then samples the
element $i$ with the largest value of $g_i+\ln v_i$. Recently, applications
including similarity estimation and graph embedding require to generate $k$
independent Gumbel-Max variables from high dimensional vectors. However, it is
computationally expensive for a large $k$ (e.g., hundreds or even thousands)
when using the traditional Gumbel-Max Trick. To solve this problem, we propose
a novel algorithm, \emph{FastGM}, that reduces the time complexity from
$O(kn^+)$ to $O(k \ln k + n^+)$, where $n^+$ is the number of positive elements
in the vector of interest. Instead of computing $k$ independent Gumbel random
variables directly, we find that there exists a technique to generate these
variables in descending order. Using this technique, our method FastGM computes
variables $g_i+\ln v_i$ for all positive elements $i$ in descending order. As a
result, FastGM significantly reduces the computation time because we can stop
the procedure of Gumbel random variables computing for many elements especially
for those with small weights. Experiments on a variety of real-world datasets
show that FastGM is orders of magnitude faster than state-of-the-art methods
without sacrificing accuracy and incurring additional expenses.
| stat |
Detecting a Structural Change in Functional Time Series Using Local
Wilcoxon Statistic | Functional data analysis (FDA) is a part of modern multivariate statistics
that analyses data providing information about curves, surfaces or anything
else varying over a certain continuum. In economics and empirical finance we
often have to deal with time series of functional data, where we cannot easily
decide, whether they are to be considered as homogeneous or heterogeneous. At
present a discussion on adequate tests of homogenity for functional data is
carried. We propose a novel statistic for detetecting a structural change in
functional time series based on a local Wilcoxon statistic induced by a local
depth function proposed by Paindaveine and Van Bever (2013).
| stat |
A Normality Test for High-dimensional Data based on a Nearest Neighbor
Approach | Many statistical methodologies for high-dimensional data assume the
population is normal. Although a few multivariate normality tests have been
proposed, to the best of our knowledge, none of them can properly control the
type I error when the dimension is larger than the number of observations. In
this work, we propose a novel nonparametric test that utilizes the nearest
neighbor information. The proposed method guarantees the asymptotic type I
error control under the high-dimensional setting. Simulation studies verify the
empirical size performance of the proposed test when the dimension grows with
the sample size and at the same time exhibit a superior power performance of
the new test compared with alternative methods. We also illustrate our approach
through two popularly used data sets in high-dimensional classification and
clustering literatures where deviation from the normality assumption may lead
to invalid conclusions.
| stat |
On Signal-to-Noise Ratio Issues in Variational Inference for Deep
Gaussian Processes | We show that the gradient estimates used in training Deep Gaussian Processes
(DGPs) with importance-weighted variational inference are susceptible to
signal-to-noise ratio (SNR) issues. Specifically, we show both theoretically
and empirically that the SNR of the gradient estimates for the latent
variable's variational parameters decreases as the number of importance samples
increases. As a result, these gradient estimates degrade to pure noise if the
number of importance samples is too large. To address this pathology, we show
how doubly-reparameterized gradient estimators, originally proposed for
training variational autoencoders, can be adapted to the DGP setting and that
the resultant estimators completely remedy the SNR issue, thereby providing
more reliable training. Finally, we demonstrate that our fix can lead to
improvements in the predictive performance of the model's predictive posterior.
| stat |
Modelling and simulation of dependence structures in nonlife insurance
with Bernstein copulas | In this paper we review Bernstein and grid-type copulas for arbitrary
dimensions and general grid resolutions in connection with discrete random
vectors possessing uniform margins. We further suggest a pragmatic way to fit
the dependence structure of multivariate data to Bernstein copulas via
grid-type copulas and empirical contingency tables. Finally, we discuss a Monte
Carlo study for the simulation and PML estimation for aggregate dependent
losses form observed windstorm and flooding data.
| stat |
Debiased Kernel Methods | I propose a practical procedure based on bias correction and sample splitting
to calculate confidence intervals for functionals of generic kernel methods,
i.e. nonparametric estimators learned in a reproducing kernel Hilbert space
(RKHS). For example, an analyst may desire confidence intervals for functionals
of kernel ridge regression. I propose a bias correction that mirrors kernel
ridge regression. The framework encompasses (i) evaluations over discrete
domains, (ii) derivatives over continuous domains, (iii) treatment effects of
discrete treatments, and (iv) incremental treatment effects of continuous
treatments. For the target quantity, whether it is (i)-(iv), I prove root-n
consistency, Gaussian approximation, and semiparametric efficiency by finite
sample arguments. I show that the classic assumptions of RKHS learning theory
also imply inference.
| stat |
Implications on Feature Detection when using the Benefit-Cost Ratio | In many practical machine learning applications, there are two objectives:
one is to maximize predictive accuracy and the other is to minimize costs of
the resulting model. These costs of individual features may be financial costs,
but can also refer to other aspects, like for example evaluation time. Feature
selection addresses both objectives, as it reduces the number of features and
can improve the generalization ability of the model. If costs differ between
features, the feature selection needs to trade-off the individual benefit and
cost of each feature. A popular trade-off choice is the ratio of both, the BCR
(benefit-cost ratio). In this paper we analyze implications of using this
measure with special focus to the ability to distinguish relevant features from
noise. We perform a simulation study for different cost and data settings and
obtain detection rates of relevant features and empirical distributions of the
trade-off ratio. Our simulation study exposed a clear impact of the cost
setting on the detection rate. In situations with large cost differences and
small effect sizes, the BCR missed relevant features and preferred cheap noise
features. We conclude that a trade-off between predictive performance and costs
without a controlling hyperparameter can easily overemphasize very cheap noise
features. While the simple benefit-cost ratio offers an easy solution to
incorporate costs, it is important to be aware of its risks. Avoiding costs
close to 0, rescaling large cost differences, or using a hyperparameter
trade-off are ways to counteract the adverse effects exposed in this paper.
| stat |
Improving D-Optimality in Nonlinear Situations | Experimental designs based on the classical D-optimal criterion minimize the
volume of the linear-approximation inference regions for the parameters using
local sensitivity coefficients. For nonlinear models, these designs can be
unreliable because the linearized inference regions do not always provide a
true indication of the exact parameter inference regions. In this article, we
apply the profile-based sensitivity coefficients developed by Sulieman et.al.
[12] in designing D-optimal experiments for parameter estimation in some
selected nonlinear models. Profile-based sensitivity coefficients are defined
by the total derivative of the model function with respect to the parameters.
They have been shown to account for both parameter co-dependencies and model
nonlinearity up to second order-derivative. This work represents a first
attempt to construct experiments using profile-based sensitivity coefficients.
Two common nonlinear models are used to illustrate the computational aspects of
the profile-based designs and simulation studies are conducted to demonstrate
the efficiency of the constructed experiments.
| stat |
A layered multiple importance sampling scheme for focused optimal
Bayesian experimental design | We develop a new computational approach for "focused" optimal Bayesian
experimental design with nonlinear models, with the goal of maximizing expected
information gain in targeted subsets of model parameters. Our approach
considers uncertainty in the full set of model parameters, but employs a design
objective that can exploit learning trade-offs among different parameter
subsets. We introduce a new layered multiple importance sampling scheme that
provides consistent estimates of expected information gain in this focused
setting. This sampling scheme yields significant reductions in estimator bias
and variance for a given computational effort, making optimal design more
tractable for a wide range of computationally intensive problems.
| stat |
Detecting British Columbia Coastal Rainfall Patterns by Clustering
Gaussian Processes | Functional data analysis is a statistical framework where data are assumed to
follow some functional form. This method of analysis is commonly applied to
time series data, where time, measured continuously or in discrete intervals,
serves as the location for a function's value. Gaussian processes are a
generalization of the multivariate normal distribution to function space and,
in this paper, they are used to shed light on coastal rainfall patterns in
British Columbia (BC). Specifically, this work addressed the question over how
one should carry out an exploratory cluster analysis for the BC, or any
similar, coastal rainfall data. An approach is developed for clustering
multiple processes observed on a comparable interval, based on how similar
their underlying covariance kernel is. This approach provides interesting
insights into the BC data, and these insights can be framed in terms of El
Ni\~{n}o and La Ni\~{n}a; however, the result is not simply one cluster
representing El Ni\~{n}o years and another for La Ni\~{n}a years. From one
perspective, the results show that clustering annual rainfall can potentially
be used to identify extreme weather patterns.
| stat |
Kernelized Wasserstein Natural Gradient | Many machine learning problems can be expressed as the optimization of some
cost functional over a parametric family of probability distributions. It is
often beneficial to solve such optimization problems using natural gradient
methods. These methods are invariant to the parametrization of the family, and
thus can yield more effective optimization. Unfortunately, computing the
natural gradient is challenging as it requires inverting a high dimensional
matrix at each iteration. We propose a general framework to approximate the
natural gradient for the Wasserstein metric, by leveraging a dual formulation
of the metric restricted to a Reproducing Kernel Hilbert Space. Our approach
leads to an estimator for gradient direction that can trade-off accuracy and
computational cost, with theoretical guarantees. We verify its accuracy on
simple examples, and show the advantage of using such an estimator in
classification tasks on Cifar10 and Cifar100 empirically.
| stat |
Bayesian Nonparametric Federated Learning of Neural Networks | In federated learning problems, data is scattered across different servers
and exchanging or pooling it is often impractical or prohibited. We develop a
Bayesian nonparametric framework for federated learning with neural networks.
Each data server is assumed to provide local neural network weights, which are
modeled through our framework. We then develop an inference approach that
allows us to synthesize a more expressive global network without additional
supervision, data pooling and with as few as a single communication round. We
then demonstrate the efficacy of our approach on federated learning problems
simulated from two popular image classification datasets.
| stat |
Generalization Bounds for Sparse Random Feature Expansions | Random feature methods have been successful in various machine learning
tasks, are easy to compute, and come with theoretical accuracy bounds. They
serve as an alternative approach to standard neural networks since they can
represent similar function spaces without a costly training phase. However, for
accuracy, random feature methods require more measurements than trainable
parameters, limiting their use for data-scarce applications or problems in
scientific machine learning. This paper introduces the sparse random feature
expansion to obtain parsimonious random feature models. Specifically, we
leverage ideas from compressive sensing to generate random feature expansions
with theoretical guarantees even in the data-scarce setting. In particular, we
provide uniform bounds on the approximation error and generalization bounds for
functions in a certain class (that is dense in a reproducing kernel Hilbert
space) depending on the number of samples and the distribution of features. The
error bounds improve with additional structural conditions, such as coordinate
sparsity, compact clusters of the spectrum, or rapid spectral decay. In
particular, by introducing sparse features, i.e. features with random sparse
weights, we provide improved bounds for low order functions. We show that the
sparse random feature expansions outperforms shallow networks in several
scientific machine learning tasks.
| stat |
Qualitative Analysis of Monte Carlo Dropout | In this report, we present qualitative analysis of Monte Carlo (MC) dropout
method for measuring model uncertainty in neural network (NN) models. We first
consider the sources of uncertainty in NNs, and briefly review Bayesian Neural
Networks (BNN), the group of Bayesian approaches to tackle uncertainties in
NNs. After presenting mathematical formulation of MC dropout, we proceed to
suggesting potential benefits and associated costs for using MC dropout in
typical NN models, with the results from our experiments.
| stat |
On Poisson-exponential-Tweedie models for ultra-overdispersed data | We introduce a new class of Poisson-exponential-Tweedie (PET) mixture in the
framework of generalized linear models for ultra-overdispersed count data. The
mean-variance relationship is of the form $m+m^{2}+\phi m^{p}$, where $\phi$
and $p$ are the dispersion and Tweedie power parameters, respectively. The
proposed model is equivalent to the exponential-Poisson-Tweedie models arising
from geometric sums of Poisson-Tweedie random variables. In this respect, the
PET models encompass the geometric versions of Hermite, Neyman Type A,
P\'{o}lya-Aeppli, negative binomial and Poisson inverse Gaussian models. The
algorithms we shall propose allow us to estimate the real power parameter,
which works as an automatic distribution selection. Instead of the classical
Poisson, zero-shifted geometric is presented as the reference count
distribution. Practical properties are incorporated into the PET of new
relative indexes of dispersion and zero-inflation phenomena. Simulation studies
demonstrate that the proposed model highlights unbiased and consistent
estimators for large samples. Illustrative practical applications are analyzed
on count datasets; in particular, PET models for data without covariates and
PET regression models. The PET models are compared to Poisson-Tweedie models
showing that parameters of both models are adopted to data.
| stat |
Convex Hierarchical Clustering for Graph-Structured Data | Convex clustering is a recent stable alternative to hierarchical clustering.
It formulates the recovery of progressively coalescing clusters as a
regularized convex problem. While convex clustering was originally designed for
handling Euclidean distances between data points, in a growing number of
applications, the data is directly characterized by a similarity matrix or
weighted graph. In this paper, we extend the robust hierarchical clustering
approach to these broader classes of similarities. Having defined an
appropriate convex objective, the crux of this adaptation lies in our ability
to provide: (a) an efficient recovery of the regularization path and (b) an
empirical demonstration of the use of our method. We address the first
challenge through a proximal dual algorithm, for which we characterize both the
theoretical efficiency as well as the empirical performance on a set of
experiments. Finally, we highlight the potential of our method by showing its
application to several real-life datasets, thus providing a natural extension
to the current scope of applications of convex clustering.
| stat |
Sparse Recovery With Multiple Data Streams: A Sequential Adaptive
Testing Approach | Multistage design has been used in a wide range of scientific fields. By
allocating sensing resources adaptively, one can effectively eliminate null
locations and localize signals with a smaller study budget. We formulate a
decision-theoretic framework for simultaneous multi-stage adaptive testing and
study how to minimize the total number of measurements while meeting
pre-specified constraints on both the false positive rate (FPR) and missed
discovery rate (MDR). The new procedure, which effectively pools information
across individual tests using a simultaneous multistage adaptive ranking and
thresholding (SMART) approach, can achieve precise error rates control and lead
to great savings in total study costs. Numerical studies confirm the
effectiveness of SMART for FPR and MDR control and show that it achieves
substantial power gain over existing methods. The SMART procedure is
demonstrated through the analysis of high-throughput screening data and spatial
imaging data.
| stat |
Reflection on modern methods: a note on variance estimation when using
inverse probability weighting to handle attrition in cohort studies | The inverse probability weighting (IPW) method is used to handle attrition in
association analyses derived from cohort studies. It consists in weighting the
respondents at a given follow-up by their inverse probability to participate.
Weights are estimated first and then used in a weighted association model. When
the IPW method is used, instead of using a so-called na{\"i}ve variance
estimator, the literature recommends using a robust variance estimator.
However, the latter may overestimate the variance because the weights are
considered known rather than estimated. In this note, we develop, by a
linearization technique, an estimator accounting for the weight estimation
phase and explain how it differs from na{\"i}ve and robust variance estimators.
We compare the three variance estimators through simulations under several MAR
and MNAR scenarios. We found that both the robust and linearized variance
estimators were approximately unbiased, even in MNAR scenarios. The naive
variance estimator severely underestimated the variance. We encourage
researchers to be careful with variance estimation when using the IPW method,
avoiding na{\"i}ve estimator and opting for a robust or linearized estimator. R
and SAS codes are provided to implement them in their own studies.
| stat |
Ensemble approximate control variate estimators: Applications to
multi-fidelity importance sampling | The recent growth in multi-fidelity uncertainty quantification has given rise
to a large set of variance reduction techniques that leverage information from
model ensembles to provide variance reduction for estimates of the statistics
of a high-fidelity model. In this paper we provide two contributions: (1) we
utilize an ensemble estimator to account for uncertainties in the optimal
weights of approximate control variate (ACV) approaches and derive lower bounds
on the number of samples required to guarantee variance reduction; and (2) we
extend an existing multi-fidelity importance sampling (MFIS) scheme to leverage
control variates. As such we make significant progress towards both increasing
the practicality of approximate control variates$-$for instance, by accounting
for the effect of pilot samples$-$and using multi-fidelity approaches more
effectively for estimating low-probability events. The numerical results
indicate our hybrid MFIS-ACV estimator achieves up to 50% improvement in
variance reduction over the existing state-of-the-art MFIS estimator, which had
already shown outstanding convergence rate compared to the Monte Carlo method,
on several problems of computational mechanics.
| stat |
Evaluating Public Supports to the Investment Activities of Business
Firms: A Multilevel Meta-Regression Analysis of Italian Studies | We conduct an extensive meta-regression analysis of counterfactual programme
evaluations from Italy, considering both published and grey literature on
enterprise and innovation policies. We specify a multilevel model for the
probability of finding positive effect estimates, also assessing correlation
possibly induced by co-authorship networks. We find that the probability of
positive effects is considerable, especially for weaker firms and outcomes that
are directly targeted by public programmes. However, these policies are less
likely to trigger change in the long run.
| stat |
Robust Imitation Learning from Noisy Demonstrations | Robust learning from noisy demonstrations is a practical but highly
challenging problem in imitation learning. In this paper, we first
theoretically show that robust imitation learning can be achieved by optimizing
a classification risk with a symmetric loss. Based on this theoretical finding,
we then propose a new imitation learning method that optimizes the
classification risk by effectively combining pseudo-labeling with co-training.
Unlike existing methods, our method does not require additional labels or
strict assumptions about noise distributions. Experimental results on
continuous-control benchmarks show that our method is more robust compared to
state-of-the-art methods.
| stat |
Invertible Generative Modeling using Linear Rational Splines | Normalizing flows attempt to model an arbitrary probability distribution
through a set of invertible mappings. These transformations are required to
achieve a tractable Jacobian determinant that can be used in high-dimensional
scenarios. The first normalizing flow designs used coupling layer mappings
built upon affine transformations. The significant advantage of such models is
their easy-to-compute inverse. Nevertheless, making use of affine
transformations may limit the expressiveness of such models. Recently,
invertible piecewise polynomial functions as a replacement for affine
transformations have attracted attention. However, these methods require
solving a polynomial equation to calculate their inverse. In this paper, we
explore using linear rational splines as a replacement for affine
transformations used in coupling layers. Besides having a straightforward
inverse, inference and generation have similar cost and architecture in this
method. Moreover, simulation results demonstrate the competitiveness of this
approach's performance compared to existing methods.
| stat |
Robust Bayesian Inference for Discrete Outcomes with the Total Variation
Distance | Models of discrete-valued outcomes are easily misspecified if the data
exhibit zero-inflation, overdispersion or contamination. Without additional
knowledge about the existence and nature of this misspecification, model
inference and prediction are adversely affected. Here, we introduce a robust
discrepancy-based Bayesian approach using the Total Variation Distance (TVD).
In the process, we address and resolve two challenges: First, we study
convergence and robustness properties of a computationally efficient estimator
for the TVD between a parametric model and the data-generating mechanism.
Second, we provide an efficient inference method adapted from Lyddon et al.
(2019) which corresponds to formulating an uninformative nonparametric prior
directly over the data-generating mechanism. Lastly, we empirically demonstrate
that our approach is robust and significantly improves predictive performance
on a range of simulated and real world data.
| stat |
Recovery of Sparse Signals from a Mixture of Linear Samples | Mixture of linear regressions is a popular learning theoretic model that is
used widely to represent heterogeneous data. In the simplest form, this model
assumes that the labels are generated from either of two different linear
models and mixed together. Recent works of Yin et al. and Krishnamurthy et al.,
2019, focus on an experimental design setting of model recovery for this
problem. It is assumed that the features can be designed and queried with to
obtain their label. When queried, an oracle randomly selects one of the two
different sparse linear models and generates a label accordingly. How many such
oracle queries are needed to recover both of the models simultaneously? This
question can also be thought of as a generalization of the well-known
compressed sensing problem (Cand\`es and Tao, 2005, Donoho, 2006). In this
work, we address this query complexity problem and provide efficient algorithms
that improves on the previously best known results.
| stat |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.