We present the problem of reinforcement learning with exogenous termination. We define the Termination Markov Decision Process (TerMDP), an extension of the MDP framework, in which episodes may be interrupted by an external non-Markovian observer. This formulation accounts for numerous real-world situations, such as a human interrupting an autonomous driving agent for reasons of discomfort. We learn the parameters of the TerMDP and leverage the structure of the estimation problem to provide state-wise confidence bounds. We use these to construct a provably-efficient algorithm, which accounts for termination, and bound its regret. Motivated by our theoretical analysis, we design and implement a scalable approach, which combines optimism (w.r.t. termination) and a dynamic discount factor, incorporating the termination probability. We deploy our method on high-dimensional driving and MinAtar benchmarks. Additionally, we test our approach on human data in a driving setting. Our results demonstrate fast convergence and significant improvement over various baseline approaches.
Robust Markov decision processes (MDPs) provide a general framework to model decision problems where the system dynamics are changing or only partially known. Recent work established the equivalence between \texttt{s} rectangular $L_p$ robust MDPs and regularized MDPs, and derived a regularized policy iteration scheme that enjoys the same level of efficiency as standard MDPs. However, there lacks a clear understanding of the policy improvement step. For example, we know the greedy policy can be stochastic but have little clue how each action affects this greedy policy. In this work, we focus on the policy improvement step and derive concrete forms for the greedy policy and the optimal robust Bellman operators. We find that the greedy policy is closely related to some combination of the top $k$ actions, which provides a novel characterization of its stochasticity. The exact nature of the combination depends on the shape of the uncertainty set. Furthermore, our results allow us to efficiently compute the policy improvement step by a simple binary search, without turning to an external optimization subroutine. Moreover, for $L_1, L_2$, and $L_\infty$ robust MDPs, we can even get rid of the binary search and evaluate the optimal robust Bellman operators exactly. Our work greatly extends existing results on solving \texttt{s}-rectangular $L_p$ robust MDPs via regularized policy iteration and can be readily adapted to sample-based model-free algorithms.
In risk-averse reinforcement learning (RL), the goal is to optimize some risk measure of the returns. A risk measure often focuses on the worst returns out of the agent's experience. As a result, standard methods for risk-averse RL often ignore high-return strategies. We prove that under certain conditions this inevitably leads to a local-optimum barrier, and propose a soft risk mechanism to bypass it. We also devise a novel Cross Entropy module for risk sampling, which (1) preserves risk aversion despite the soft risk; (2) independently improves sample efficiency. By separating the risk aversion of the sampler and the optimizer, we can sample episodes with poor conditions, yet optimize with respect to successful strategies. We combine these two concepts in CeSoR - Cross-entropy Soft-Risk optimization algorithm - which can be applied on top of any risk-averse policy gradient (PG) method. We demonstrate improved risk aversion in maze navigation, autonomous driving, and resource allocation benchmarks, including in scenarios where standard risk-averse PG completely fails.
Quantum Computing (QC) stands to revolutionize computing, but is currently still limited. To develop and test quantum algorithms today, quantum circuits are often simulated on classical computers. Simulating a complex quantum circuit requires computing the contraction of a large network of tensors. The order (path) of contraction can have a drastic effect on the computing cost, but finding an efficient order is a challenging combinatorial optimization problem. We propose a Reinforcement Learning (RL) approach combined with Graph Neural Networks (GNN) to address the contraction ordering problem. The problem is extremely challenging due to the huge search space, the heavy-tailed reward distribution, and the challenging credit assignment. We show how a carefully implemented RL-agent that uses a GNN as the basic policy construct can address these challenges and obtain significant improvements over state-of-the-art techniques in three varieties of circuits, including the largest scale networks used in contemporary QC.
The Hidden Markov Model (HMM) is one of the most widely used statistical models for sequential data analysis, and it has been successfully applied in a large variety of domains. One of the key reasons for this versatility is the ability of HMMs to deal with missing data. However, standard HMM learning algorithms rely crucially on the assumption that the positions of the missing observations within the observation sequence are known. In some situations where such assumptions are not feasible, a number of special algorithms have been developed. Currently, these algorithms rely strongly on specific structural assumptions of the underlying chain, such as acyclicity, and are not applicable in the general case. In particular, there are numerous domains within medicine and computational biology, where the missing observation locations are unknown and acyclicity assumptions do not hold, thus presenting a barrier for the application of HMMs in those fields. In this paper we consider a general problem of learning HMMs from data with unknown missing observation locations (i.e., only the order of the non-missing observations are known). We introduce a generative model of the location omissions and propose two learning methods for this model, a (semi) analytic approach, and a Gibbs sampler. We evaluate and compare the algorithms in a variety of scenarios, measuring their reconstruction precision and robustness under model misspecification.
Reasoning and interacting with dynamic environments is a fundamental problem in AI, but it becomes extremely challenging when actions can trigger cascades of cross-dependent events. We introduce a new supervised learning setup called {\em Cascade} where an agent is shown a video of a physically simulated dynamic scene, and is asked to intervene and trigger a cascade of events, such that the system reaches a "counterfactual" goal. For instance, the agent may be asked to "Make the blue ball hit the red one, by pushing the green ball". The agent intervention is drawn from a continuous space, and cascades of events makes the dynamics highly non-linear. We combine semantic tree search with an event-driven forward model and devise an algorithm that learns to search in semantic trees in continuous spaces. We demonstrate that our approach learns to effectively follow instructions to intervene in previously unseen complex scenes. It can also reason about alternative outcomes, when provided an observed cascade of events.
Motivated by a real-world problem of blood coagulation control in Heparin-treated patients, we use Stochastic Differential Equations (SDEs) to formulate a new class of sequential prediction problems -- with an unknown latent space, unknown non-linear dynamics, and irregular sparse observations. We introduce the Neural Eigen-SDE (NESDE) algorithm for sequential prediction with sparse observations and adaptive dynamics. NESDE applies eigen-decomposition to the dynamics model to allow efficient frequent predictions given sparse observations. In addition, NESDE uses a learning mechanism for adaptive dynamics model, which handles changes in the dynamics both between sequences and within sequences. We demonstrate the accuracy and efficacy of NESDE for both synthetic problems and real-world data. In particular, to the best of our knowledge, we are the first to provide a patient-adapted prediction for blood coagulation following Heparin dosing in the MIMIC-IV dataset. Finally, we publish a simulated gym environment based on our prediction model, for experimentation in algorithms for blood coagulation control.
The space of value functions is a fundamental concept in reinforcement learning. Characterizing its geometric properties may provide insights for optimization and representation. Existing works mainly focus on the value space for Markov Decision Processes (MDPs). In this paper, we study the geometry of the robust value space for the more general Robust MDPs (RMDPs) setting, where transition uncertainties are considered. Specifically, since we find it hard to directly adapt prior approaches to RMDPs, we start with revisiting the non-robust case, and introduce a new perspective that enables us to characterize both the non-robust and robust value space in a similar fashion. The key of this perspective is to decompose the value space, in a state-wise manner, into unions of hypersurfaces. Through our analysis, we show that the robust value space is determined by a set of conic hypersurfaces, each of which contains the robust values of all policies that agree on one state. Furthermore, we find that taking only extreme points in the uncertainty set is sufficient to determine the robust value space. Finally, we discuss some other aspects about the robust value space, including its non-convexity and policy agreement on multiple states.
Motivated by online recommendation systems, we propose the problem of finding the optimal policy in multitask contextual bandits when a small fraction $\alpha < 1/2$ of tasks (users) are arbitrary and adversarial. The remaining fraction of good users share the same instance of contextual bandits with $S$ contexts and $A$ actions (items). Naturally, whether a user is good or adversarial is not known in advance. The goal is to robustly learn the policy that maximizes rewards for good users with as few user interactions as possible. Without adversarial users, established results in collaborative filtering show that $O(1/\epsilon^2)$ per-user interactions suffice to learn a good policy, precisely because information can be shared across users. This parallelization gain is fundamentally altered by the presence of adversarial users: unless there are super-polynomial number of users, we show a lower bound of $\tilde{\Omega}(\min(S,A) \cdot \alpha^2 / \epsilon^2)$ {\it per-user} interactions to learn an $\epsilon$-optimal policy for the good users. We then show we can achieve an $\tilde{O}(\min(S,A)\cdot \alpha/\epsilon^2)$ upper-bound, by employing efficient robust mean estimators for both uni-variate and high-dimensional random variables. We also show that this can be improved depending on the distributions of contexts.
The classical Policy Iteration (PI) algorithm alternates between greedy one-step policy improvement and policy evaluation. Recent literature shows that multi-step lookahead policy improvement leads to a better convergence rate at the expense of increased complexity per iteration. However, prior to running the algorithm, one cannot tell what is the best fixed lookahead horizon. Moreover, per a given run, using a lookahead of horizon larger than one is often wasteful. In this work, we propose for the first time to dynamically adapt the multi-step lookahead horizon as a function of the state and of the value estimate. We devise two PI variants and analyze the trade-off between iteration count and computational complexity per iteration. The first variant takes the desired contraction factor as the objective and minimizes the per-iteration complexity. The second variant takes as input the computational complexity per iteration and minimizes the overall contraction factor. We then devise a corresponding DQN-based algorithm with an adaptive tree search horizon. We also include a novel enhancement for on-policy learning: per-depth value function estimator. Lastly, we demonstrate the efficacy of our adaptive lookahead method in a maze environment and in Atari.