In the field of high-performance computing (HPC), there has been recent exploration into the use of deep reinforcement learning for cluster scheduling (DRL scheduling), which has demonstrated promising outcomes. However, a significant challenge arises from the lack of interpretability in deep neural networks (DNN), rendering them as black-box models to system managers. This lack of model interpretability hinders the practical deployment of DRL scheduling. In this work, we present a framework called IRL (Interpretable Reinforcement Learning) to address the issue of interpretability of DRL scheduling. The core idea is to interpret DNN (i.e., the DRL policy) as a decision tree by utilizing imitation learning. Unlike DNN, decision tree models are non-parametric and easily comprehensible to humans. To extract an effective and efficient decision tree, IRL incorporates the Dataset Aggregation (DAgger) algorithm and introduces the notion of critical state to prune the derived decision tree. Through trace-based experiments, we demonstrate that IRL is capable of converting a black-box DNN policy into an interpretable rulebased decision tree while maintaining comparable scheduling performance. Additionally, IRL can contribute to the setting of rewards in DRL scheduling.

The advent of deep reinforcement learning (DRL) has significantly advanced the field of robotics, particularly in the control and coordination of quadruped robots. However, the complexity of real-world tasks often necessitates the deployment of multi-robot systems capable of sophisticated interaction and collaboration. To address this need, we introduce the Multi-agent Quadruped Environment (MQE), a novel platform designed to facilitate the development and evaluation of multi-agent reinforcement learning (MARL) algorithms in realistic and dynamic scenarios. MQE emphasizes complex interactions between robots and objects, hierarchical policy structures, and challenging evaluation scenarios that reflect real-world applications. We present a series of collaborative and competitive tasks within MQE, ranging from simple coordination to complex adversarial interactions, and benchmark state-of-the-art MARL algorithms. Our findings indicate that hierarchical reinforcement learning can simplify task learning, but also highlight the need for advanced algorithms capable of handling the intricate dynamics of multi-agent interactions. MQE serves as a stepping stone towards bridging the gap between simulation and practical deployment, offering a rich environment for future research in multi-agent systems and robot learning. For open-sourced code and more details of MQE, please refer to https://ziyanx02.github.io/multiagent-quadruped-environment/ .

Traditional imitation learning focuses on modeling the behavioral mechanisms of experts, which requires a large amount of interaction history generated by some fixed expert. However, in many streaming applications, such as streaming recommender systems, online decision-makers typically engage in online learning during the decision-making process, meaning that the interaction history generated by online decision-makers includes their behavioral evolution from novice expert to experienced expert. This poses a new challenge for existing imitation learning approaches that can only utilize data from experienced experts. To address this issue, this paper proposes an inverse batched contextual bandit (IBCB) framework that can efficiently perform estimations of environment reward parameters and learned policy based on the expert's behavioral evolution history. Specifically, IBCB formulates the inverse problem into a simple quadratic programming problem by utilizing the behavioral evolution history of the batched contextual bandit with inaccessible rewards. We demonstrate that IBCB is a unified framework for both deterministic and randomized bandit policies. The experimental results indicate that IBCB outperforms several existing imitation learning algorithms on synthetic and real-world data and significantly reduces running time. Additionally, empirical analyses reveal that IBCB exhibits better out-of-distribution generalization and is highly effective in learning the bandit policy from the interaction history of novice experts.

Reinforcement learning (RL) has revolutionized decision-making across a wide range of domains over the past few decades. Yet, deploying RL policies in real-world scenarios presents the crucial challenge of ensuring safety. Traditional safe RL approaches have predominantly focused on incorporating predefined safety constraints into the policy learning process. However, this reliance on predefined safety constraints poses limitations in dynamic and unpredictable real-world settings where such constraints may not be available or sufficiently adaptable. Bridging this gap, we propose a novel approach that concurrently learns a safe RL control policy and identifies the unknown safety constraint parameters of a given environment. Initializing with a parametric signal temporal logic (pSTL) safety specification and a small initial labeled dataset, we frame the problem as a bilevel optimization task, intricately integrating constrained policy optimization, using a Lagrangian-variant of the twin delayed deep deterministic policy gradient (TD3) algorithm, with Bayesian optimization for optimizing parameters for the given pSTL safety specification. Through experimentation in comprehensive case studies, we validate the efficacy of this approach across varying forms of environmental constraints, consistently yielding safe RL policies with high returns. Furthermore, our findings indicate successful learning of STL safety constraint parameters, exhibiting a high degree of conformity with true environmental safety constraints. The performance of our model closely mirrors that of an ideal scenario that possesses complete prior knowledge of safety constraints, demonstrating its proficiency in accurately identifying environmental safety constraints and learning safe policies that adhere to those constraints.

Robust reinforcement learning agents using high-dimensional observations must be able to identify relevant state features amidst many exogeneous distractors. A representation that captures controllability identifies these state elements by determining what affects agent control. While methods such as inverse dynamics and mutual information capture controllability for a limited number of timesteps, capturing long-horizon elements remains a challenging problem. Myopic controllability can capture the moment right before an agent crashes into a wall, but not the control-relevance of the wall while the agent is still some distance away. To address this we introduce action-bisimulation encoding, a method inspired by the bisimulation invariance pseudometric, that extends single-step controllability with a recursive invariance constraint. By doing this, action-bisimulation learns a multi-step controllability metric that smoothly discounts distant state features that are relevant for control. We demonstrate that action-bisimulation pretraining on reward-free, uniformly random data improves sample efficiency in several environments, including a photorealistic 3D simulation domain, Habitat. Additionally, we provide theoretical analysis and qualitative results demonstrating the information captured by action-bisimulation.

One of the main challenges in multi-agent reinforcement learning is scalability as the number of agents increases. This issue is further exacerbated if the problem considered is temporally dependent. State-of-the-art solutions today mainly follow centralized training with decentralized execution paradigm in order to handle the scalability concerns. In this paper, we propose time-dependent multi-agent transformers which can solve the temporally dependent multi-agent problem efficiently with a centralized approach via the use of transformers that proficiently handle the large input. We highlight the efficacy of this method on two problems and use tools from statistics to verify the probability that the trajectories generated under the policy satisfy the task. The experiments show that our approach has superior performance against the literature baseline algorithms in both cases.

In actor-critic framework for fully decentralized multi-agent reinforcement learning (MARL), one of the key components is the MARL policy evaluation (PE) problem, where a set of $N$ agents work cooperatively to evaluate the value function of the global states for a given policy through communicating with their neighbors. In MARL-PE, a critical challenge is how to lower the sample and communication complexities, which are defined as the number of training samples and communication rounds needed to converge to some $\epsilon$-stationary point. To lower communication complexity in MARL-PE, a "natural'' idea is to perform multiple local TD-update steps between each consecutive rounds of communication to reduce the communication frequency. However, the validity of the local TD-update approach remains unclear due to the potential "agent-drift'' phenomenon resulting from heterogeneous rewards across agents in general. This leads to an interesting open question: Can the local TD-update approach entail low sample and communication complexities? In this paper, we make the first attempt to answer this fundamental question. We focus on the setting of MARL-PE with average reward, which is motivated by many multi-agent network optimization problems. Our theoretical and experimental results confirm that allowing multiple local TD-update steps is indeed an effective approach in lowering the sample and communication complexities of MARL-PE compared to consensus-based MARL-PE algorithms. Specifically, the local TD-update steps between two consecutive communication rounds can be as large as $\mathcal{O}(1/\epsilon^{1/2}\log{(1/\epsilon)})$ in order to converge to an $\epsilon$-stationary point of MARL-PE. Moreover, we show theoretically that in order to reach the optimal sample complexity, the communication complexity of local TD-update approach is $\mathcal{O}(1/\epsilon^{1/2}\log{(1/\epsilon)})$.

As Machine Learning systems become increasingly popular across diverse application domains, including those with direct human implications, the imperative of equity and algorithmic fairness has risen to prominence in the Artificial Intelligence community. On the other hand, in the context of Shared Micromobility Systems, the exploration of fairness-oriented approaches remains limited. Addressing this gap, we introduce a pioneering investigation into the balance between performance optimization and algorithmic fairness in the operation and control of Shared Micromobility Services. Our study leverages the Q-Learning algorithm in Reinforcement Learning, benefiting from its convergence guarantees to ensure the robustness of our proposed approach. Notably, our methodology stands out for its ability to achieve equitable outcomes, as measured by the Gini index, across different station categories--central, peripheral, and remote. Through strategic rebalancing of vehicle distribution, our approach aims to maximize operator performance while simultaneously upholding fairness principles for users. In addition to theoretical insights, we substantiate our findings with a case study or simulation based on synthetic data, validating the efficacy of our approach. This paper underscores the critical importance of fairness considerations in shaping control strategies for Shared Micromobility Services, offering a pragmatic framework for enhancing equity in urban transportation systems.

In this paper, we present an online reinforcement learning algorithm for constrained Markov decision processes with a safety constraint. Despite the necessary attention of the scientific community, considering stochastic stopping time, the problem of learning optimal policy without violating safety constraints during the learning phase is yet to be addressed. To this end, we propose an algorithm based on linear programming that does not require a process model. We show that the learned policy is safe with high confidence. We also propose a method to compute a safe baseline policy, which is central in developing algorithms that do not violate the safety constraints. Finally, we provide simulation results to show the efficacy of the proposed algorithm. Further, we demonstrate that efficient exploration can be achieved by defining a subset of the state-space called proxy set.

This study focuses on two important problems related to applying offline model-based optimization to real-world industrial control problems. The first problem is how to create a reliable probabilistic model that accurately captures the dynamics present in noisy industrial data. The second problem is how to reliably optimize control parameters without actively collecting feedback from industrial systems. Specifically, we introduce a novel cGAN ensemble-based uncertainty-aware surrogate model for reliable offline model-based optimization in industrial control problems. The effectiveness of the proposed method is demonstrated through extensive experiments conducted on two representative cases, namely a discrete control case and a continuous control case. The results of these experiments show that our method outperforms several competitive baselines in the field of offline model-based optimization for industrial control.