More Like this

1. Adaptive Discretization for Episodic Reinforcement Learning in Metric Spaces

   https://doi.org/10.1145/3410048.3410059  

   Sinclair, Sean R.; Banerjee, Siddhartha; Lee Yu, Christina (July 2020, ACM SIGMETRICS Performance Evaluation Review)

   null (Ed.)

   We present an efficient algorithm for model-free episodic reinforcement learning on large (potentially continuous) state-action spaces. Our algorithm is based on a novel Q-learning policy with adaptive data-driven discretization. The central idea is to maintain a finer partition of the state-action space in regions which are frequently visited in historical trajectories, and have higher payoff estimates. We demonstrate how our adaptive partitions take advantage of the shape of the optimal Q-function and the joint space, without sacrificing the worst-case performance. In particular, we recover the regret guarantees of prior algorithms for continuous state-action spaces, which additionally require either an optimal discretization as input, and/or access to a simulation oracle. Moreover, experiments demonstrate how our algorithm automatically adapts to the underlying structure of the problem, resulting in much better performance compared both to heuristics and Q-learning with uniform discretization. 

   more » « less
2. Offline Stochastic Shortest Path: Learning, Evaluation and Towards Optimality

   Yin, Ming; Chen, Wenjing; Wang, Mengdi; Wang, Yu-Xiang (May 2022, Uncertainty in artificial intelligence)

   Goal-oriented Reinforcement Learning, where the agent needs to reach the goal state while simultaneously minimizing the cost, has received significant attention in real-world applications. Its theoretical formulation, stochastic shortest path (SSP), has been intensively researched in the online setting. Nevertheless, it remains understudied when such an online interaction is prohibited and only historical data is provided. In this paper, we consider the offline stochastic shortest path problem when the state space and the action space are finite. We design the simple value iteration-based algorithms for tackling both offline policy evaluation (OPE) and offline policy learning tasks. Notably, our analysis of these simple algorithms yields strong instance-dependent bounds which can imply worst-case bounds that are near-minimax optimal. We hope our study could help illuminate the fundamental statistical limits of the offline SSP problem and motivate further studies beyond the scope of current consideration. 

   more » « less
3. Learning Off-policy for Online Planning

   Sikchi, Harshit; Zhou, Wenxuan; Held, David (January 2021, Conference on Robot Learning (CoRL))

   Reinforcement learning (RL) in low-data and risk-sensitive domains requires performant and flexible deployment policies that can readily incorporate constraints during deployment. One such class of policies are the semi-parametric H-step lookahead policies, which select actions using trajectory optimization over a dynamics model for a fixed horizon with a terminal value function. In this work, we investigate a novel instantiation of H-step lookahead with a learned model and a terminal value function learned by a model-free off-policy algorithm, named Learning Off-Policy with Online Planning (LOOP). We provide a theoretical analysis of this method, suggesting a tradeoff between model errors and value function errors and empirically demonstrate this tradeoff to be beneficial in deep reinforcement learning. Furthermore, we identify the "Actor Divergence" issue in this framework and propose Actor Regularized Control (ARC), a modified trajectory optimization procedure. We evaluate our method on a set of robotic tasks for Offline and Online RL and demonstrate improved performance. We also show the flexibility of LOOP to incorporate safety constraints during deployment with a set of navigation environments. We demonstrate that LOOP is a desirable framework for robotics applications based on its strong performance in various important RL settings. 

   more » « less
4. Performance Bounds for Model and Policy Transfer in Hidden-parameter MDPs

   Fu, H; Yao, J; Gottesman, O; Doshi-Velez, F; Konidaris, GD (May 2023, Proceedings of the Eleventh International Conference on Learning Representations)

   In the Hidden-Parameter MDP (HiP-MDP) framework, a family of reinforcement learning tasks is generated by varying hidden parameters specifying the dynamics and reward function for each individual task. The HiP-MDP is a natural model for families of tasks in which meta- and lifelong-reinforcement learning approaches can succeed. Given a learned context encoder that infers the hidden parameters from previous experience, most existing algorithms fall into two categories: model transfer and policy transfer, depending on which function the hidden parameters are used to parameterize. We characterize the robustness of model and policy transfer algorithms with respect to hidden parameter estimation error. We first show that the value function of HiP-MDPs is Lipschitz continuous under certain conditions. We then derive regret bounds for both settings through the lens of Lipschitz continuity. Finally, we empirically corroborate our theoretical analysis by varying the hyper-parameters governing the Lipschitz constants of two continuous control problems; the resulting performance is consistent with our theoretical results. 

   more » « less
5. A Novel Approach to Error Resilience in Online Reinforcement Learning

   Chandramouli, Amarnath.; Chatterjee, Abhijit. (July 2023, International On-Line Testing Symposium)

   Online reinforcement learning (RL) based systems are being increasingly deployed in a variety of safety-critical applications ranging from drone control to medical robotics. These systems typically use RL onboard rather than relying on remote operation from high-performance datacenters. Due to the dynamic nature of the environments they work in, onboard RL hardware is vulnerable to soft errors from radiation, thermal effects and electrical noise that corrupt the results of computations. Existing approaches to on-line error resilience in machine learning systems have relied on availability of the large training datasets to configure resilience parameters, which is not necessarily feasible for online RL systems. Similarly, other approaches involving specialized hardware or modifications to training algorithms are difficult to implement for onboard RL applications. In contrast, we present a novel error resilience approach for online RL that makes use of running statistics collected across the (real-time) RL training process to configure error detection thresholds without the need to access a reference training dataset. In this methodology, statistical concentration bounds leveraging running statistics are used to diagnose neuron outputs as erroneous. These erroneous neurons are then set to zero (suppressed). Our approach is compared against the state of the art and validated on several RL algorithms involving the use of multiple concentration bounds on CPU as well as GPU hardware. 

   more » « less