The multi-armed bandit (MAB) model is one of the most classical models to study decision-making in an uncertain environment. In this model, a player chooses one of $K$ possible arms of a bandit machine to play at each time step, where the corresponding arm returns a random reward to the player, potentially from a specific unknown distribution. The target of the player is to collect as many rewards as possible during the process. Despite its simplicity, the MAB model offers an excellent playground for studying the trade-off between exploration versus exploitation and designing effective algorithms for sequential decision-making under uncertainty. Although many asymptotically optimal algorithms have been established, the finite-time behaviors of the stochastic dynamics of the MAB model appear much more challenging to analyze, due to the intertwine between the decision-making and the rewards being collected. In this paper, we employ techniques in statistical physics to analyze the MAB model, which facilitates the characterization of the distribution of cumulative regrets at a finite short time, the central quantity of interest in an MAB algorithm, as well as the intricate dynamical behaviors of the model. Our analytical results, in good agreement with simulations, point to the emergence of an interesting multimodal regret distribution, with large regrets resulting from excess exploitation of sub-optimal arms due to an initial unlucky output from the optimal one.

This paper investigates conservative exploration in reinforcement learning where the performance of the learning agent is guaranteed to be above a certain threshold throughout the learning process. It focuses on the tabular episodic Markov Decision Process (MDP) setting that has finite states and actions. With the knowledge of an existing safe baseline policy, an algorithm termed as StepMix is proposed to balance the exploitation and exploration while ensuring that the conservative constraint is never violated in each episode with high probability. StepMix features a unique design of a mixture policy that adaptively and smoothly interpolates between the baseline policy and the optimistic policy. Theoretical analysis shows that StepMix achieves near-optimal regret order as in the constraint-free setting, indicating that obeying the stringent episode-wise conservative constraint does not compromise the learning performance. Besides, a randomization-based EpsMix algorithm is also proposed and shown to achieve the same performance as StepMix. The algorithm design and theoretical analysis are further extended to the setting where the baseline policy is not given a priori but must be learned from an offline dataset, and it is proved that similar conservative guarantee and regret can be achieved if the offline dataset is sufficiently large. Experiment results corroborate the theoretical analysis and demonstrate the effectiveness of the proposed conservative exploration strategies.

A fairly reliable trend in deep reinforcement learning is that the performance scales with the number of parameters, provided a complimentary scaling in amount of training data. As the appetite for large models increases, it is imperative to address, sooner than later, the potential problem of running out of high-quality demonstrations. In this case, instead of collecting only new data via costly human demonstrations or risking a simulation-to-real transfer with uncertain effects, it would be beneficial to leverage vast amounts of readily-available low-quality data. Since classical control algorithms such as behavior cloning or temporal difference learning cannot be used on reward-free or action-free data out-of-the-box, this solution warrants novel training paradigms for continuous control. We propose a simple algorithm called Diffused Value Function (DVF), which learns a joint multi-step model of the environment-robot interaction dynamics using a diffusion model. This model can be efficiently learned from state sequences (i.e., without access to reward functions nor actions), and subsequently used to estimate the value of each action out-of-the-box. We show how DVF can be used to efficiently capture the state visitation measure for multiple controllers, and show promising qualitative and quantitative results on challenging robotics benchmarks.

Offline reinforcement learning (ORL) has gained attention as a means of training reinforcement learning models using pre-collected static data. To address the issue of limited data and improve downstream ORL performance, recent work has attempted to expand the dataset's coverage through data augmentation. However, most of these methods are tied to a specific policy (policy-dependent), where the generated data can only guarantee to support the current downstream ORL policy, limiting its usage scope on other downstream policies. Moreover, the quality of synthetic data is often not well-controlled, which limits the potential for further improving the downstream policy. To tackle these issues, we propose HIgh-quality POlicy-DEcoupled (HIPODE), a novel data augmentation method for ORL. On the one hand, HIPODE generates high-quality synthetic data by selecting states near the dataset distribution with potentially high value among candidate states using the negative sampling technique. On the other hand, HIPODE is policy-decoupled, thus can be used as a common plug-in method for any downstream ORL process. We conduct experiments on the widely studied TD3BC and CQL algorithms, and the results show that HIPODE outperforms the state-of-the-art policy-decoupled data augmentation method and most prevalent model-based ORL methods on D4RL benchmarks.

Artificial intelligence (AI) has emerged as a powerful tool for addressing complex and dynamic tasks in communication systems, where traditional rule-based algorithms often struggle. However, most AI applications to networking tasks are designed and trained for specific, limited conditions, hindering the algorithms from learning and adapting to generic situations, such as those met across radio access networks (RAN). This paper proposes design principles for sustainable and scalable AI integration in communication systems, focusing on creating AI algorithms that can generalize across network environments, intents, and control tasks. This approach enables a limited number of AI-driven RAN functions to tackle larger problems, improve system performance, and simplify lifecycle management. To achieve sustainability and automation, we introduce a scalable learning architecture that supports all deployed AI applications in the system. This architecture separates centralized learning functionalities from distributed actuation and inference functions, enabling efficient data collection and management, computational and storage resources optimization, and cost reduction. We illustrate these concepts by designing a generalized link adaptation algorithm, demonstrating the benefits of our proposed approach.

As intelligent trading agents based on reinforcement learning (RL) gain prevalence, it becomes more important to ensure that RL agents obey laws, regulations, and human behavioral expectations. There is substantial literature concerning the aversion of obvious catastrophes like crashing a helicopter or bankrupting a trading account, but little around the avoidance of subtle non-normative behavior for which there are examples, but no programmable definition. Such behavior may violate legal or regulatory, rather than physical or monetary, constraints. In this article, I consider a series of experiments in which an intelligent stock trading agent maximizes profit but may also inadvertently learn to spoof the market in which it participates. I first inject a hand-coded spoofing agent to a multi-agent market simulation and learn to recognize spoofing activity sequences. Then I replace the hand-coded spoofing trader with a simple profit-maximizing RL agent and observe that it independently discovers spoofing as the optimal strategy. Finally, I introduce a method to incorporate the recognizer as normative guide, shaping the agent's perceived rewards and altering its selected actions. The agent remains profitable while avoiding spoofing behaviors that would result in even higher profit. After presenting the empirical results, I conclude with some recommendations. The method should generalize to the reduction of any unwanted behavior for which a recognizer can be learned.

In spite of the large literature on reinforcement learning (RL) algorithms for partially observable Markov decision processes (POMDPs), a complete theoretical understanding is still lacking. In a partially observable setting, the history of data available to the agent increases over time so most practical algorithms either truncate the history to a finite window or compress it using a recurrent neural network leading to an agent state that is non-Markovian. In this paper, it is shown that in spite of the lack of the Markov property, recurrent Q-learning (RQL) converges in the tabular setting. Moreover, it is shown that the quality of the converged limit depends on the quality of the representation which is quantified in terms of what is known as an approximate information state (AIS). Based on this characterization of the approximation error, a variant of RQL with AIS losses is presented. This variant performs better than a strong baseline for RQL that does not use AIS losses. It is demonstrated that there is a strong correlation between the performance of RQL over time and the loss associated with the AIS representation.

Combinatorial optimization problems require an exhaustive search to find the optimal solution. A convenient approach to solving combinatorial optimization tasks in the form of Mixed Integer Linear Programs is Branch-and-Bound. Branch-and-Bound solver splits a task into two parts dividing the domain of an integer variable, then it solves them recursively, producing a tree of nested sub-tasks. The efficiency of the solver depends on the branchning heuristic used to select a variable for splitting. In the present work, we propose a reinforcement learning method that can efficiently learn the branching heuristic. We view the variable selection task as a tree Markov Decision Process, prove that the Bellman operator adapted for the tree Markov Decision Process is contracting in mean, and propose a modified learning objective for the reinforcement learning agent. Our agent requires less training data and produces smaller trees compared to previous reinforcement learning methods.

Learning in MDPs with highly complex state representations is currently possible due to multiple advancements in reinforcement learning algorithm design. However, this incline in complexity, and furthermore the increase in the dimensions of the observation came at the cost of volatility that can be taken advantage of via adversarial attacks (i.e. moving along worst-case directions in the observation space). To solve this policy instability problem we propose a novel method to detect the presence of these non-robust directions via local quadratic approximation of the deep neural policy loss. Our method provides a theoretical basis for the fundamental cut-off between safe observations and adversarial observations. Furthermore, our technique is computationally efficient, and does not depend on the methods used to produce the worst-case directions. We conduct extensive experiments in the Arcade Learning Environment with several different adversarial attack techniques. Most significantly, we demonstrate the effectiveness of our approach even in the setting where non-robust directions are explicitly optimized to circumvent our proposed method.

In reinforcement learning (RL), we are concerned with learning good policies for sequential decisionmaking problems modeled as Markov Decision Processes (MDPs) [29, 35]. MDPs assume that the transition model of the environment is fixed across training and testing, but this is often violated in practical applications. For example, when deploying a simulator-trained robot in reality, a notable challenge is the substantial disparity between the simulated environment and the intricate complexities of the real world, leading to potential subpar performance upon deployment. Such a mismatch may significantly degrade the performance of the trained policy in testing. To deal with this issue, the robust MDP (RMDP) framework has been introduced in [16, 24, 44], aiming to learn policies that are robust to perturbation of the transition model within an uncertainty set.