Most reinforcement learning algorithms with formal regret guarantees assume all mistakes are reversible and rely on essentially trying all possible options. This approach leads to poor outcomes when some mistakes are irreparable or even catastrophic. We propose a variant of the contextual bandit problem where the goal is to minimize the chance of catastrophe. Specifically, we assume that the payoff each round represents the chance of avoiding catastrophe that round, and try to maximize the product of payoffs (the overall chance of avoiding catastrophe). To give the agent some chance of success, we allow a limited number of queries to a mentor and assume a Lipschitz continuous payoff function. We present an algorithm whose regret and rate of querying the mentor both approach 0 as the time horizon grows, assuming a continuous 1D state space and a relatively "simple" payoff function. We also provide a matching lower bound: without the simplicity assumption: any algorithm either constantly asks for help or is nearly guaranteed to cause catastrophe. Finally, we identify the key obstacle to generalizing our algorithm to a multi-dimensional state space.

Optimizing large-scale wireless networks, including optimal resource management, power allocation, and throughput maximization, is inherently challenging due to their non-observable system dynamics and heterogeneous and complex nature. Herein, a novel ensemble Q-learning algorithm that addresses the performance and complexity challenges of the traditional Q-learning algorithm for optimizing wireless networks is presented. Ensemble learning with synthetic Markov Decision Processes is tailored to wireless networks via new models for approximating large state-space observable wireless networks. In particular, digital cousins are proposed as an extension of the traditional digital twin concept wherein multiple Q-learning algorithms on multiple synthetic Markovian environments are run in parallel and their outputs are fused into a single Q-function. Convergence analyses of key statistics and Q-functions and derivations of upper bounds on the estimation bias and variance are provided. Numerical results across a variety of real-world wireless networks show that the proposed algorithm can achieve up to 50% less average policy error with up to 40% less runtime complexity than the state-of-the-art reinforcement learning algorithms. It is also shown that theoretical results properly predict trends in the experimental results.

Developing agents that can leverage planning abilities during their decision and learning processes is critical to the advancement of Artificial Intelligence. Recent works have demonstrated the effectiveness of combining tree-based search methods and self-play learning mechanisms. Yet, these methods typically face scaling challenges due to the sequential nature of their search. While practical engineering solutions can partly overcome this, they still demand extensive computational resources, which hinders their applicability. In this paper, we introduce SMX, a model-based planning algorithm that utilises scalable Sequential Monte Carlo methods to create an effective self-learning mechanism. Grounded in the theoretical framework of control as inference, SMX benefits from robust theoretical underpinnings. Its sampling-based search approach makes it adaptable to environments with both discrete and continuous action spaces. Furthermore, SMX allows for high parallelisation and can run on hardware accelerators to optimise computing efficiency. SMX demonstrates a statistically significant improvement in performance compared to AlphaZero, as well as demonstrating its performance as an improvement operator for a model-free policy, matching or exceeding top model-free methods across both continuous and discrete environments.

Doubly robust methods hold considerable promise for off-policy evaluation in Markov decision processes (MDPs) under sequential ignorability: They have been shown to converge as $1/\sqrt{T}$ with the horizon $T$, to be statistically efficient in large samples, and to allow for modular implementation where preliminary estimation tasks can be executed using standard reinforcement learning techniques. Existing results, however, make heavy use of a strong distributional overlap assumption whereby the stationary distributions of the target policy and the data-collection policy are within a bounded factor of each other -- and this assumption is typically only credible when the state space of the MDP is bounded. In this paper, we re-visit the task of off-policy evaluation in MDPs under a weaker notion of distributional overlap, and introduce a class of truncated doubly robust (TDR) estimators which we find to perform well in this setting. When the distribution ratio of the target and data-collection policies is square-integrable (but not necessarily bounded), our approach recovers the large-sample behavior previously established under strong distributional overlap. When this ratio is not square-integrable, TDR is still consistent but with a slower-than-$1/\sqrt{T}$; furthermore, this rate of convergence is minimax over a class of MDPs defined only using mixing conditions. We validate our approach numerically and find that, in our experiments, appropriate truncation plays a major role in enabling accurate off-policy evaluation when strong distributional overlap does not hold.

Distributed decision-making in multi-agent systems presents difficult challenges for interactive behavior learning in both cooperative and competitive systems. To mitigate this complexity, MAIDRL presents a semi-centralized Dense Reinforcement Learning algorithm enhanced by agent influence maps (AIMs), for learning effective multi-agent control on StarCraft Multi-Agent Challenge (SMAC) scenarios. In this paper, we extend the DenseNet in MAIDRL and introduce semi-centralized Multi-Agent Dense-CNN Reinforcement Learning, MAIDCRL, by incorporating convolutional layers into the deep model architecture, and evaluate the performance on both homogeneous and heterogeneous scenarios. The results show that the CNN-enabled MAIDCRL significantly improved the learning performance and achieved a faster learning rate compared to the existing MAIDRL, especially on more complicated heterogeneous SMAC scenarios. We further investigate the stability and robustness of our model. The statistics reflect that our model not only achieves higher winning rate in all the given scenarios but also boosts the agent's learning process in fine-grained decision-making.

Engineering change orders (ECOs) in late stages make minimal design fixes to recover from timing shifts due to excessive IR drops. This paper integrates IR-drop-aware timing analysis and ECO timing optimization using reinforcement learning (RL). The method operates after physical design and power grid synthesis, and rectifies IR-drop-induced timing degradation through gate sizing. It incorporates the Lagrangian relaxation (LR) technique into a novel RL framework, which trains a relational graph convolutional network (R-GCN) agent to sequentially size gates to fix timing violations. The R-GCN agent outperforms a classical LR-only algorithm: in an open 45nm technology, it (a) moves the Pareto front of the delay-area tradeoff curve to the left and (b) saves runtime over the classical method by running fast inference using trained models at iso-quality. The RL model is transferable across timing specifications, and transferable to unseen designs with zero-shot learning or fine tuning.

In the field of online sequential decision-making, we address the problem with delays utilizing the framework of online convex optimization (OCO), where the feedback of a decision can arrive with an unknown delay. Unlike previous research that is limited to Euclidean norm and gradient information, we propose three families of delayed algorithms based on approximate solutions to handle different types of received feedback. Our proposed algorithms are versatile and applicable to universal norms. Specifically, we introduce a family of Follow the Delayed Regularized Leader algorithms for feedback with full information on the loss function, a family of Delayed Mirror Descent algorithms for feedback with gradient information on the loss function and a family of Simplified Delayed Mirror Descent algorithms for feedback with the value information of the loss function's gradients at corresponding decision points. For each type of algorithm, we provide corresponding regret bounds under cases of general convexity and relative strong convexity, respectively. We also demonstrate the efficiency of each algorithm under different norms through concrete examples. Furthermore, our theoretical results are consistent with the current best bounds when degenerated to standard settings.

We propose a new algorithm for model-based distributional reinforcement learning (RL), and prove that it is minimax-optimal for approximating return distributions with a generative model (up to logarithmic factors), resolving an open question of Zhang et al. (2023). Our analysis provides new theoretical results on categorical approaches to distributional RL, and also introduces a new distributional Bellman equation, the stochastic categorical CDF Bellman equation, which we expect to be of independent interest. We also provide an experimental study comparing several model-based distributional RL algorithms, with several takeaways for practitioners.

The deployment of ever-larger machine learning models reflects a growing consensus that the more expressive the model$\unicode{x2013}$and the more data one has access to$\unicode{x2013}$the more one can improve performance. As models get deployed in a variety of real world scenarios, they inevitably face strategic environments. In this work, we consider the natural question of how the interplay of models and strategic interactions affects scaling laws. We find that strategic interactions can break the conventional view of scaling laws$\unicode{x2013}$meaning that performance does not necessarily monotonically improve as models get larger and/ or more expressive (even with infinite data). We show the implications of this phenomenon in several contexts including strategic regression, strategic classification, and multi-agent reinforcement learning through examples of strategic environments in which$\unicode{x2013}$by simply restricting the expressivity of one's model or policy class$\unicode{x2013}$one can achieve strictly better equilibrium outcomes. Motivated by these examples, we then propose a new paradigm for model-selection in games wherein an agent seeks to choose amongst different model classes to use as their action set in a game.

Reinforcement learning (RL) has shown its strength in challenging sequential decision-making problems. The reward function in RL is crucial to the learning performance, as it serves as a measure of the task completion degree. In real-world problems, the rewards are predominantly human-designed, which requires laborious tuning, and is easily affected by human cognitive biases. To achieve automatic auxiliary reward generation, we propose a novel representation learning approach that can measure the ``transition distance'' between states. Building upon these representations, we introduce an auxiliary reward generation technique for both single-task and skill-chaining scenarios without the need for human knowledge. The proposed approach is evaluated in a wide range of manipulation tasks. The experiment results demonstrate the effectiveness of measuring the transition distance between states and the induced improvement by auxiliary rewards, which not only promotes better learning efficiency but also increases convergent stability.