AISTATS 2020 论文列表

In this paper, we consider distributed algorithms for solving the empirical risk minimization problem under the master/worker communication model. We develop a distributed asynchronous quasi-Newton algorithm that can achieve superlinear convergence. To our knowledge, this is the first distributed asynchronous algorithm with superlinear convergence guarantees. Our algorithm is communication-efficient in the sense that at every iteration the master node and workers communicate vectors of size $O(p)$, where $p$ is the dimension of the decision variable. The proposed method is based on a distributed asynchronous averaging scheme of decision vectors and gradients in a way to effectively capture the local Hessian information of the objective function. Our convergence theory supports asynchronous computations subject to both bounded delays and unbounded delays with a bounded time-average. Unlike in the majority of asynchronous optimization literature, we do not require choosing smaller stepsize when delays are huge. We provide numerical experiments that match our theoretical results and showcase significant improvement comparing to state-of-the-art distributed algorithms.

We consider the problem of optimizing an unknown (typically non-convex) function with a bounded norm in some Reproducing Kernel Hilbert Space (RKHS), based on noisy bandit feedback. We consider a novel variant of this problem in which the point evaluations are not only corrupted by random noise, but also adversarial corruptions. We introduce an algorithm Fast-Slow GP-UCB based on Gaussian process methods, randomized selection between two instances labeled ’fast’ (but non-robust) and ’slow’ (but robust), enlarged confidence bounds, and the principle of optimism under uncertainty. We present a novel theoret- ical analysis upper bounding the cumulative regret in terms of the corruption level, the time horizon, and the underlying kernel, and we argue that certain dependencies cannot be improved. We observe that distinct algorithmic ideas are required depending on whether one is required to perform well in both the corrupted and non-corrupted settings, and whether the corruption level is known or not.

Deterministic models are approximations of reality that are easy to interpret and often easier to build than stochastic alternatives. Unfortunately, as nature is capricious, observational data can never be fully explained by deterministic models in practice. Observation and process noise need to be added to adapt deterministic models to behave stochastically, such that they are capable of explaining and extrapolating from noisy data. We investigate and address computational inefficiencies that arise from adding process noise to deterministic simulators that fail to return for certain inputs; a property we describe as ’brittle’. We show how to train a conditional normalizing flow to propose perturbations such that the simulator succeeds with high probability, increasing computational efficiency.

We develop a convex analytic framework for ReLU neural networks which elucidates the inner workings of hidden neurons and their function space characteristics. We show that rectified linear units in neural networks act as convex regularizers, where simple solutions are encouraged via extreme points of a certain convex set. For one dimensional regression and classification, we prove that finite two-layer ReLU networks with norm regularization yield linear spline interpolation. In the more general higher dimensional case, we show that the training problem for two-layer networks can be cast as a convex optimization problem with infinitely many constraints. We then provide a family of convex relaxations to approximate the solution, and a cutting-plane algorithm to improve the relaxations. We derive conditions for the exactness of the relaxations and provide simple closed form formulas for the optimal neural network weights in certain cases. Our results show that the hidden neurons of a ReLU network can be interpreted as convex autoencoders of the input layer. We also establish a connection to $\ell_0$-$\ell_1$ equivalence for neural networks analogous to the minimal cardinality solutions in compressed sensing. Extensive experimental results show that the proposed approach yields interpretable and accurate models.

Maximum a posteriori (MAP) inference is a fundamental computational paradigm for statistical inference. In the setting of graphical models, MAP inference entails solving a combinatorial optimization problem to find the most likely configuration of the discrete-valued model. Linear programming (LP) relaxations in the Sherali-Adams hierarchy are widely used to attempt to solve this problem, and smooth message passing algorithms have been proposed to solve regularized versions of these LPs with great success. This paper leverages recent work in entropy-regularized LPs to analyze convergence rates of a class of edge-based smooth message passing algorithms to epsilon-optimality in the relaxation. With an appropriately chosen regularization constant, we present a theoretical guarantee on the number of iterations sufficient to recover the true integral MAP solution when the LP is tight and the solution is unique.

We present tight convergence rate bounds for gradient descent and MM algorithms for maximum likelihood (ML) estimation and maximum a posteriori probability (MAP) estimation of a popular Bayesian inference method, for Bradley-Terry models of ranking data. Our results show that MM algorithms have the same convergence rate, up to a constant factor, as gradient descent algorithms with optimal constant step size. For the ML estimation objective, the convergence is linear with the rate crucially determined by the algebraic connectivity of the matrix of item pair co-occurrences in observed comparison data. For the MAP estimation objective, we show that the convergence rate is also linear, with the rate determined by a parameter of the prior distribution in a way that can make convergence arbitrarily slow for small values of this parameter. The limit of small values of this parameter corresponds to a flat, non-informative prior distribution.

We analyze the convergence rate of the randomized Newton-like method introduced by Qu et. al. (2016) for smooth and convex objectives, which uses random coordinate blocks of a Hessian-over-approximation matrix M instead of the true Hessian. The convergence analysis of the algorithm is challenging because of its complex dependence on the structure of M. However, we show that when the coordinate blocks are sampled with probability proportional to their determinant, the convergence rate depends solely on the eigenvalue distribution of matrix M, and has an analytically tractable form. To do so, we derive a fundamental new expectation formula for determinantal point processes. We show that determinantal sampling allows us to reason about the optimal subset size of blocks in terms of the spectrum of M. Additionally, we provide a numerical evaluation of our analysis, demonstrating cases where determinantal sampling is superior or on par with uniform sampling.

Multiple hypothesis testing, a situation when we wish to consider many hypotheses, is a core problem in statistical inference that arises in almost every scientific field. In this setting, controlling the false discovery rate (FDR), which is the expected proportion of type I error, is an important challenge for making meaningful inferences. In this paper, we consider a setting where an ordered (possibly infinite) sequence of hypotheses arrives in a stream, and for each hypothesis we observe a p-value along with a set of features specific to that hypothesis. The decision whether or not to reject the current hypothesis must be made immediately at each time step, before the next hypothesis is observed. This model provides a general way of leveraging the side (contextual) information in the data to help maximize the number of discoveries while controlling the FDR.We propose a new class of powerful online testing procedures, where the rejection thresholds are learned sequentially by incorporating contextual information and previous results. We prove that any rule in this class controls online FDR under some standard assumptions. We then focus on a subclass of these procedures, based on weighting the rejection thresholds, to derive a practical algorithm that learns a parametric weight function in an online fashion to gain more discoveries. We also theoretically prove that our proposed procedures, under some easily verifiable assumptions, would lead to an increase of statistical power over a popular online testing procedure proposed by Javanmard and Montanari (2018). Finally, we demonstrate the superior performance of our procedure, by comparing it to state-of-the-art online multiple testing procedures, on both synthetic data and real data generated from different applications.

Clinical trials in the medical domain are constrained by budgets. The number of patients that can be recruited is therefore limited. When a patient population is heterogeneous, this creates difficulties in learning subgroup specific responses to a particular drug and especially for a variety of dosages. In addition, patient recruitment can be difficult by the fact that clinical trials do not aim to provide a benefit to any given patient in the trial. In this paper, we propose C3T-Budget, a contextual constrained clinical trial algorithm for dose-finding under both budget and safety constraints. The algorithm aims to maximize drug efficacy within the clinical trial while also learning about the drug being tested. C3T-Budget recruits patients with consideration of the remaining budget, the remaining time, and the characteristics of each group, such as the population distribution, estimated expected efficacy, and estimation credibility. In addition, the algorithm aims to avoid unsafe dosages. These characteristics are further illustrated in a simulated clinical trial study, which corroborates the theoretical analysis and demonstrates an efficient budget usage as well as a balanced learning-treatment trade-off.

We consider contextual combinatorial volatile multi-armed bandit (CCV-MAB), in which at each round, the learner observes a set of available base arms and their contexts, and then, selects a super arm that contains $K$ base arms in order to maximize its cumulative reward. Under the semi-bandit feedback setting and assuming that the contexts lie in a space ${\cal X}$ endowed with the Euclidean norm and that the expected base arm outcomes (expected rewards) are Lipschitz continuous in the contexts (expected base arm outcomes), we propose an algorithm called Adaptive Contextual Combinatorial Upper Confidence Bound (ACC-UCB). This algorithm, which adaptively discretizes ${\cal X}$ to form estimates of base arm outcomes and uses an $\alpha$-approximation oracle as a subroutine to select a super arm in each round, achieves $\tilde{O} ( T^{(\bar{D}+1)/(\bar{D}+2) + \epsilon} )$ regret for any $\epsilon>0$, where $\bar{D}$ represents the approximate optimality dimension related to ${\cal X}$. This dimension captures both the benignness of the base arm arrivals and the structure of the expected reward. In addition, we provide a recipe for obtaining more optimistic regret bounds by taking into account the volatility of the base arms and show that ACC-UCB achieves significant performance gains compared to the state-of-the-art for worker selection in mobile crowdsourcing.

We present a framework for building unsupervised representations of entities and their compositions, where each entity is viewed as a probability distribution rather than a vector embedding. In particular, this distribution is supported over the contexts which co-occur with the entity and are embedded in a suitable low-dimensional space. This enables us to consider representation learning from the perspective of Optimal Transport and take advantage of its tools such as Wasserstein distance and barycenters. We elaborate how the method can be applied for obtaining unsupervised representations of text and illustrate the performance (quantitatively as well as qualitatively) on tasks such as measuring sentence similarity, word entailment and similarity, where we empirically observe significant gains (e.g., 4.1% relative improvement over Sent2vec, GenSen).The key benefits of the proposed approach include: (a) capturing uncertainty and polysemy via modeling the entities as distributions, (b) utilizing the underlying geometry of the particular task (with the ground cost), (c) simultaneously providing interpretability with the notion of optimal transport between contexts and (d) easy applicability on top of existing point embedding methods. The code, as well as pre-built histograms, are available under https://github.com/context-mover/.

Modern machine learning models with very high accuracy have been shown to be vulnerable to small, adversarially chosen perturbations of the input. Given black-box access to a high-accuracy classifier f, we show how to construct a new classifier g that has high accuracy and is also robust to adversarial L2-bounded perturbations. Our algorithm builds upon the framework of randomized smoothing that has been recently shown to outperform all previous defenses against L2-bounded adversaries. Using techniques like random partitions and doubling dimension, we are able to bound the adversarial error of g in terms of the optimum error. In this paper we focus on our conceptual contribution, but we do present two examples to illustrate our framework. We will argue that, under some assumptions, our bounds are optimal for these cases.

While learning in an unknown Markov Decision Process (MDP), an agent should trade off exploration to discover new information about the MDP, and exploitation of the current knowledge to maximize the reward. Although the agent will eventually learn a good or optimal policy, there is no guarantee on the quality of the intermediate policies. This lack of control is undesired in real-world applications where a minimum requirement is that the executed policies are guaranteed to perform at least as well as an existing baseline. In this paper, we introduce the notion of conservative exploration for average reward and finite horizon problems. We present two optimistic algorithms that guarantee (w.h.p.) that the conservative constraint is never violated during learning. We derive regret bounds showing that being conservative does not hinder the learning ability of these algorithms.

Work in machine learning and statistics commonly focuses on building models that capture the vast majority of data, possibly ignoring a segment of the population as outliers. However, there may not exist a good, simple model for the distribution, so we seek to find a small subset where there exists such a model. We give a computationally efficient algorithm with theoretical analysis for the conditional linear regression task, which is the joint task of identifying a significant portion of the data distribution, described by a k-DNF, along with a linear predictor on that portion with a small loss. In contrast to work in robust statistics on small subsets, our loss bounds do not feature a dependence on the density of the portion we fit, and compared to previous work on conditional linear regression, our algorithm’s running time scales polynomially with the sparsity of the linear predictor. We also demonstrate empirically that our algorithm can leverage this advantage to obtain a k-DNF with a better linear predictor in practice.

The principal contribution of this paper is a conceptual framework for off-policy reinforcement learning, based on conditional expectations of importance sampling ratios. This framework yields new perspectives and understanding of existing off-policy algorithms, and reveals a broad space of unexplored algorithms. We theoretically analyse this space, and concretely investigate several algorithms that arise from this framework.

Differentially private (DP) machine learning has recently become popular. The privacy loss of DP algorithms is commonly reported using (e.d)-DP. In this paper, we propose a numerical accountant for evaluating the privacy loss for algorithms with continuous one dimensional output. This accountant can be applied to the subsampled multidimensional Gaussian mechanism which underlies the popular DP stochastic gradient descent. The proposed method is based on a numerical approximation of an integral formula which gives the exact (e.d)-values. The approximation is carried out by discretising the integral and by evaluating discrete convolutions using the fast Fourier transform algorithm. We give theoretical error bounds which show the convergence of the approximation and guarantee its accuracy to an arbitrary degree. We give both theoretical error bounds and numerical error estimates for the approximation. Experimental comparisons with state-of-the-art techniques demonstrate significant improvements in bound tightness and/or computation time.

Stable matching, a classical model for two-sided markets, has long been studied assuming known preferences. In reality agents often have to learn about their preferences through exploration. With the advent of massive online markets powered by data-driven matching platforms, it has become necessary to better understand the interplay between learning and market objectives. We propose a statistical learning model in which one side of the market does not have a priori knowledge about its preferences for the other side and is required to learn these from stochastic rewards. Our model extends the standard multi-armed bandits framework to multiple players, with the added feature that arms have preferences over players. We study both centralized and decentralized approaches to this problem and show surprising exploration-exploitation trade-offs compared to the single player multi-armed bandits setting.

Due to the imminent need to alleviate the communication burden in multi-agent and federated learning, the investigation of communication-efficient distributed optimization algorithms for empirical risk minimization has flourished recently. A large fraction of existing algorithms are developed for the master/slave setting, relying on the presence of a central parameter server. This paper focuses on distributed optimization in the network setting (also known as the decentralized setting), where each agent is only allowed to aggregate information from its neighbors over a graph. By properly adjusting the global gradient estimate via a tracking term, we first develop a communication-efficient approximate Newton-type method, called Network-DANE, which generalizes the attractive DANE algorithm to decentralized networks. Our key algorithmic ideas can be applied, in a systematic manner, to obtain decentralized versions of other master/slave distributed algorithms. Notably, we develop Network-SVRG/SARAH, which employ stochastic variance reduction at each agent to accelerate local computations. We establish linear convergence of Network-DANE and Network-SVRG for strongly convex losses, and Network-SARAH for quadratic losses, which shed light on the impact of data homogeneity, network connectivity, and local averaging upon the rate of convergence. Numerical evidence is provided to demonstrate the appealing performance of our algorithms over competitive baselines, in terms of both communication and computation efficiency.

Large-scale machine learning training suffers from two prior challenges, specifically for nuclear-norm constrained problems with distributed systems: the synchronization slowdown due to the straggling workers, and high communication costs. In this work, we propose an asynchronous Stochastic Frank Wolfe (SFW-asyn) method, which, for the first time, solves the two problems simultaneously, while successfully maintaining the same convergence rate as the vanilla SFW. We implement our algorithm in python (with MPI) to run on Amazon EC2, and demonstrate that SFW-asyn yields speed-ups almost linear to the number of machines compared to the vanilla SFW.