Abstract:With the increasing demand for large-scale training of machine learning models, consensus-based distributed optimization methods have recently been advocated as alternatives to the popular parameter server framework. In this paradigm, each worker maintains a local estimate of the optimal parameter vector, and iteratively updates it by waiting and averaging all estimates obtained from its neighbors, and then corrects it on the basis of its local dataset. However, the synchronization phase can be time consuming due to the need to wait for \textit{stragglers}, i.e., slower workers. An efficient way to mitigate this effect is to let each worker wait only for updates from the fastest neighbors before updating its local parameter. The remaining neighbors are called \textit{backup workers.} To minimize the globally training time over the network, we propose a fully distributed algorithm to dynamically determine the number of backup workers for each worker. We show that our algorithm achieves a linear speedup for convergence (i.e., convergence performance increases linearly with respect to the number of workers). We conduct extensive experiments on MNIST and CIFAR-10 to verify our theoretical results.
Abstract:We consider the problem of service placement at the network edge, in which a decision maker has to choose between $N$ services to host at the edge to satisfy the demands of customers. Our goal is to design adaptive algorithms to minimize the average service delivery latency for customers. We pose the problem as a Markov decision process (MDP) in which the system state is given by describing, for each service, the number of customers that are currently waiting at the edge to obtain the service. However, solving this $N$-services MDP is computationally expensive due to the curse of dimensionality. To overcome this challenge, we show that the optimal policy for a single-service MDP has an appealing threshold structure, and derive explicitly the Whittle indices for each service as a function of the number of requests from customers based on the theory of Whittle index policy. Since request arrival and service delivery rates are usually unknown and possibly time-varying, we then develop efficient learning augmented algorithms that fully utilize the structure of optimal policies with a low learning regret. The first of these is UCB-Whittle, and relies upon the principle of optimism in the face of uncertainty. The second algorithm, Q-learning-Whittle, utilizes Q-learning iterations for each service by using a two time scale stochastic approximation. We characterize the non-asymptotic performance of UCB-Whittle by analyzing its learning regret, and also analyze the convergence properties of Q-learning-Whittle. Simulation results show that the proposed policies yield excellent empirical performance.
Abstract:We provide an adversarial approach to estimating Riesz representers of linear functionals within arbitrary function spaces. We prove oracle inequalities based on the localized Rademacher complexity of the function space used to approximate the Riesz representer and the approximation error. These inequalities imply fast finite sample mean-squared-error rates for many function spaces of interest, such as high-dimensional sparse linear functions, neural networks and reproducing kernel Hilbert spaces. Our approach offers a new way of estimating Riesz representers with a plethora of recently introduced machine learning techniques. We show how our estimator can be used in the context of de-biasing structural/causal parameters in semi-parametric models, for automated orthogonalization of moment equations and for estimating the stochastic discount factor in the context of asset pricing.
Abstract:Negative control is a strategy for learning the causal relationship between treatment and outcome in the presence of unmeasured confounding. The treatment effect can nonetheless be identified if two auxiliary variables are available: a negative control treatment (which has no effect on the actual outcome), and a negative control outcome (which is not affected by the actual treatment). These auxiliary variables can also be viewed as proxies for a traditional set of control variables, and they bear resemblance to instrumental variables. I propose a new family of non-parametric algorithms for learning treatment effects with negative controls. I consider treatment effects of the population, of sub-populations, and of alternative populations. I allow for data that may be discrete or continuous, and low-, high-, or infinite-dimensional. I impose the additional structure of the reproducing kernel Hilbert space (RKHS), a popular non-parametric setting in machine learning. I prove uniform consistency and provide finite sample rates of convergence. I evaluate the estimators in simulations.
Abstract:In this paper, we propose an algorithm for estimating the parameters of a time-homogeneous hidden Markov model from aggregate observations. This problem arises when only the population level counts of the number of individuals at each time step are available, from which one seeks to learn the individual hidden Markov model. Our algorithm is built upon expectation-maximization and the recently proposed aggregate inference algorithm, the Sinkhorn belief propagation. As compared with existing methods such as expectation-maximization with non-linear belief propagation, our algorithm exhibits convergence guarantees. Moreover, our learning framework naturally reduces to the standard Baum-Welch learning algorithm when observations corresponding to a single individual are recorded. We further extend our learning algorithm to handle HMMs with continuous observations. The efficacy of our algorithm is demonstrated on a variety of datasets.
Abstract:The Reward-Biased Maximum Likelihood Estimate (RBMLE) for adaptive control of Markov chains was proposed in (Kumar and Becker, 1982) to overcome the central obstacle of what is called the "closed-identifiability problem" of adaptive control, the "dual control problem" by Feldbaum (Feldbaum, 1960a,b), or the "exploration vs. exploitation problem". It exploited the key observation that since the maximum likelihood parameter estimator can asymptotically identify the closed-transition probabilities under a certainty equivalent approach (Borkar and Varaiya, 1979), the limiting parameter estimates must necessarily have an optimal reward that is less than the optimal reward for the true but unknown system. Hence it proposed a bias in favor of parameters with larger optimal rewards, providing a carefully structured solution to above problem. It thereby proposed an optimistic approach of favoring parameters with larger optimal rewards, now known as "optimism in the face of uncertainty." The RBMLE approach has been proved to be longterm average reward optimal in a variety of contexts including controlled Markov chains, linear quadratic Gaussian systems, some nonlinear systems, and diffusions. However, modern attention is focused on the much finer notion of "regret," or finite-time performance for all time, espoused by (Lai and Robbins, 1985). Recent analysis of RBMLE for multi-armed stochastic bandits (Liu et al., 2020) and linear contextual bandits (Hung et al., 2020) has shown that it has state-of-the-art regret and exhibits empirical performance comparable to or better than the best current contenders. Motivated by this, we examine the finite-time performance of RBMLE for reinforcement learning tasks of optimal control of unknown Markov Decision Processes. We show that it has a regret of $O(\log T)$ after $T$ steps, similar to state-of-art algorithms.
Abstract:The deep neural networks (DNNs) have achieved great success in learning complex patterns with strong predictive power, but they are often thought of as "black box" models without a sufficient level of transparency and interpretability. It is important to demystify the DNNs with rigorous mathematics and practical tools, especially when they are used for mission-critical applications. This paper aims to unwrap the black box of deep ReLU networks through local linear representation, which utilizes the activation pattern and disentangles the complex network into an equivalent set of local linear models (LLMs). We develop a convenient LLM-based toolkit for interpretability, diagnostics, and simplification of a pre-trained deep ReLU network. We propose the local linear profile plot and other visualization methods for interpretation and diagnostics, and an effective merging strategy for network simplification. The proposed methods are demonstrated by simulation examples, benchmark datasets, and a real case study in home lending credit risk assessment.
Abstract:We consider a class of filtering problems for large populations where each individual is modeled by the same hidden Markov model (HMM). In this paper, we focus on aggregate inference problems in HMMs with discrete state space and continuous observation space. The continuous observations are aggregated in a way such that the individuals are indistinguishable from measurements. We propose an aggregate inference algorithm called continuous observation collective forward-backward algorithm. It extends the recently proposed collective forward-backward algorithm for aggregate inference in HMMs with discrete observations to the case of continuous observations. The efficacy of this algorithm is illustrated through several numerical experiments.
Abstract:We study a variant of the classical multi-armed bandit problem (MABP) which we call as Multi-Armed Bandits with dependent arms. More specifically, multiple arms are grouped together to form a cluster, and the reward distributions of arms belonging to the same cluster are known functions of an unknown parameter that is a characteristic of the cluster. Thus, pulling an arm $i$ not only reveals information about its own reward distribution, but also about all those arms that share the same cluster with arm $i$. This "correlation" amongst the arms complicates the exploration-exploitation trade-off that is encountered in the MABP because the observation dependencies allow us to test simultaneously multiple hypotheses regarding the optimality of an arm. We develop learning algorithms based on the UCB principle which utilize these additional side observations appropriately while performing exploration-exploitation trade-off. We show that the regret of our algorithms grows as $O(K\log T)$, where $K$ is the number of clusters. In contrast, for an algorithm such as the vanilla UCB that is optimal for the classical MABP and does not utilize these dependencies, the regret scales as $O(M\log T)$ where $M$ is the number of arms.
Abstract:We propose a novel framework for non-parametric policy evaluation in static and dynamic settings. Under the assumption of selection on observables, we consider treatment effects of the population, of sub-populations, and of alternative populations that may have alternative covariate distributions. We further consider the decomposition of a total effect into a direct effect and an indirect effect (as mediated by a particular mechanism). Under the assumption of sequential selection on observables, we consider the effects of sequences of treatments. Across settings, we allow for treatments that may be discrete, continuous, or even text. Across settings, we allow for estimation of not only counterfactual mean outcomes but also counterfactual distributions of outcomes. We unify analyses across settings by showing that all of these causal learning problems reduce to the re-weighting of a prediction, i.e. causal adjustment. We implement the re-weighting as an inner product in a function space called a reproducing kernel Hilbert space (RKHS), with a closed form solution that can be computed in one line of code. We prove uniform consistency and provide finite sample rates of convergence. We evaluate our estimators in simulations devised by other authors. We use our new estimators to evaluate continuous and heterogeneous treatment effects of the US Jobs Corps training program for disadvantaged youth.