Designing reward functions is a core component of reinforcement learning but can be challenging for truly complex behavior. Reinforcement Learning from Human Feedback (RLHF) has been used to alleviate this challenge by replacing a hand-coded reward function with a reward function learned from preferences. However, it can be exceedingly inefficient to learn these rewards as they are often learned tabula rasa. We investigate whether Large Language Models (LLMs) can reduce this query inefficiency by converting an iterative series of human preferences into code representing the rewards. We propose In-Context Preference Learning (ICPL), a method that uses the grounding of an LLM to accelerate learning reward functions from preferences. ICPL takes the environment context and task description, synthesizes a set of reward functions, and then repeatedly updates the reward functions using human rankings of videos of the resultant policies. Using synthetic preferences, we demonstrate that ICPL is orders of magnitude more efficient than RLHF and is even competitive with methods that use ground-truth reward functions instead of preferences. Finally, we perform a series of human preference-learning trials and observe that ICPL extends beyond synthetic settings and can work effectively with humans-in-the-loop. Additional information and videos are provided at https://sites.google.com/view/few-shot-icpl/home.

In reinforcement learning (RL), temporal difference (TD) error is known to be related to the firing rate of dopamine neurons. It has been observed that each dopamine neuron does not behave uniformly, but each responds to the TD error in an optimistic or pessimistic manner, interpreted as a kind of distributional RL. To explain such a biological data, a heuristic model has also been designed with learning rates asymmetric for the positive and negative TD errors. However, this heuristic model is not theoretically-grounded and unknown whether it can work as a RL algorithm. This paper therefore introduces a novel theoretically-grounded model with optimism and pessimism, which is derived from control as inference. In combination with ensemble learning, a distributional value function as a critic is estimated from regularly introduced optimism and pessimism. Based on its central value, a policy in an actor is improved. This proposed algorithm, so-called DROP (distributional and regular optimism and pessimism), is compared on dynamic tasks. Although the heuristic model showed poor learning performance, DROP showed excellent one in all tasks with high generality. In other words, it was suggested that DROP is a new model that can elicit the potential contributions of optimism and pessimism.

Reinforcement learning (RL) has been successfully applied to solve the problem of finding obstacle-free paths for autonomous agents operating in stochastic and uncertain environments. However, when the underlying stochastic dynamics of the environment experiences drastic distribution shifts, the optimal policy obtained in the trained environment may be sub-optimal or may entirely fail in helping find goal-reaching paths for the agent. Approaches like domain randomization and robust RL can provide robust policies, but typically assume minor (bounded) distribution shifts. For substantial distribution shifts, retraining (either with a warm-start policy or from scratch) is an alternative approach. In this paper, we develop a novel approach called {\em Evolutionary Robust Policy Optimization} (ERPO), an adaptive re-training algorithm inspired by evolutionary game theory (EGT). ERPO learns an optimal policy for the shifted environment iteratively using a temperature parameter that controls the trade off between exploration and adherence to the old optimal policy. The policy update itself is an instantiation of the replicator dynamics used in EGT. We show that under fairly common sparsity assumptions on rewards in such environments, ERPO converges to the optimal policy in the shifted environment. We empirically demonstrate that for path finding tasks in a number of environments, ERPO outperforms several popular RL and deep RL algorithms (PPO, A3C, DQN) in many scenarios and popular environments. This includes scenarios where the RL algorithms are allowed to train from scratch in the new environment, when they are retrained on the new environment, or when they are used in conjunction with domain randomization. ERPO shows faster policy adaptation, higher average rewards, and reduced computational costs in policy adaptation.

A vast amount of scholarly work is published daily, yet much of it remains inaccessible to the general public due to dense jargon and complex language. To address this challenge in science communication, we introduce a reinforcement learning framework that fine-tunes a language model to rewrite scholarly abstracts into more comprehensible versions. Guided by a carefully balanced combination of word- and sentence-level accessibility rewards, our language model effectively substitutes technical terms with more accessible alternatives, a task which models supervised fine-tuned or guided by conventional readability measures struggle to accomplish. Our best model adjusts the readability level of scholarly abstracts by approximately six U.S. grade levels -- in other words, from a postgraduate to a high school level. This translates to roughly a 90% relative boost over the supervised fine-tuning baseline, all while maintaining factual accuracy and high-quality language. An in-depth analysis of our approach shows that balanced rewards lead to systematic modifications in the base model, likely contributing to smoother optimization and superior performance. We envision this work as a step toward bridging the gap between scholarly research and the general public, particularly younger readers and those without a college degree.

In this study, we investigate the effect of SSL objective modifications within the SPR framework, focusing on specific adjustments such as terminal state masking and prioritized replay weighting, which were not explicitly addressed in the original design. While these modifications are specific to RL, they are not universally applicable across all RL algorithms. Therefore, we aim to assess their impact on performance and explore other SSL objectives that do not accommodate these adjustments like Barlow Twins and VICReg. We evaluate six SPR variants on the Atari 100k benchmark, including versions both with and without these modifications. Additionally, we test the performance of these objectives on the DeepMind Control Suite, where such modifications are absent. Our findings reveal that incorporating specific SSL modifications within SPR significantly enhances performance, and this influence extends to subsequent frameworks like SR-SPR and BBF, highlighting the critical importance of SSL objective selection and related adaptations in achieving data efficiency in self-predictive reinforcement learning.

Off-policy evaluation (OPE) is one of the most fundamental problems in reinforcement learning (RL) to estimate the expected long-term payoff of a given target policy with only experiences from another behavior policy that is potentially unknown. The distribution correction estimation (DICE) family of estimators have advanced the state of the art in OPE by breaking the curse of horizon. However, the major bottleneck of applying DICE estimators lies in the difficulty of solving the saddle-point optimization involved, especially with neural network implementations. In this paper, we tackle this challenge by establishing a linear representation of value function and stationary distribution correction ratio, i.e., primal and dual variables in the DICE framework, using the spectral decomposition of the transition operator. Such primal-dual representation not only bypasses the non-convex non-concave optimization in vanilla DICE, therefore enabling an computational efficient algorithm, but also paves the way for more efficient utilization of historical data. We highlight that our algorithm, SpectralDICE, is the first to leverage the linear representation of primal-dual variables that is both computation and sample efficient, the performance of which is supported by a rigorous theoretical sample complexity guarantee and a thorough empirical evaluation on various benchmarks.

Understanding cognitive processes in multi-agent interactions is a primary goal in cognitive science. It can guide the direction of artificial intelligence (AI) research toward social decision-making in multi-agent systems, which includes uncertainty from character heterogeneity. In this paper, we introduce an episodic future thinking (EFT) mechanism for a reinforcement learning (RL) agent, inspired by cognitive processes observed in animals. To enable future thinking functionality, we first develop a multi-character policy that captures diverse characters with an ensemble of heterogeneous policies. Here, the character of an agent is defined as a different weight combination on reward components, representing distinct behavioral preferences. The future thinking agent collects observation-action trajectories of the target agents and uses the pre-trained multi-character policy to infer their characters. Once the character is inferred, the agent predicts the upcoming actions of target agents and simulates the potential future scenario. This capability allows the agent to adaptively select the optimal action, considering the predicted future scenario in multi-agent interactions. To evaluate the proposed mechanism, we consider the multi-agent autonomous driving scenario with diverse driving traits and multiple particle environments. Simulation results demonstrate that the EFT mechanism with accurate character inference leads to a higher reward than existing multi-agent solutions. We also confirm that the effect of reward improvement remains valid across societies with different levels of character diversity.

Recovering a spectrum of diverse policies from a set of expert trajectories is an important research topic in imitation learning. After determining a latent style for a trajectory, previous diverse policies recovering methods usually employ a vanilla behavioral cloning learning objective conditioned on the latent style, treating each state-action pair in the trajectory with equal importance. Based on an observation that in many scenarios, behavioral styles are often highly relevant with only a subset of state-action pairs, this paper presents a new principled method in diverse polices recovery. In particular, after inferring or assigning a latent style for a trajectory, we enhance the vanilla behavioral cloning by incorporating a weighting mechanism based on pointwise mutual information. This additional weighting reflects the significance of each state-action pair's contribution to learning the style, thus allowing our method to focus on state-action pairs most representative of that style. We provide theoretical justifications for our new objective, and extensive empirical evaluations confirm the effectiveness of our method in recovering diverse policies from expert data.

Multi-Agent Reinforcement Learning (MARL) struggles with sample inefficiency and poor generalization [1]. These challenges are partially due to a lack of structure or inductive bias in the neural networks typically used in learning the policy. One such form of structure that is commonly observed in multi-agent scenarios is symmetry. The field of Geometric Deep Learning has developed Equivariant Graph Neural Networks (EGNN) that are equivariant (or symmetric) to rotations, translations, and reflections of nodes. Incorporating equivariance has been shown to improve learning efficiency and decrease error [ 2 ]. In this paper, we demonstrate that EGNNs improve the sample efficiency and generalization in MARL. However, we also show that a naive application of EGNNs to MARL results in poor early exploration due to a bias in the EGNN structure. To mitigate this bias, we present Exploration-enhanced Equivariant Graph Neural Networks or E2GN2. We compare E2GN2 to other common function approximators using common MARL benchmarks MPE and SMACv2. E2GN2 demonstrates a significant improvement in sample efficiency, greater final reward convergence, and a 2x-5x gain in over standard GNNs in our generalization tests. These results pave the way for more reliable and effective solutions in complex multi-agent systems.

Federated learning (FL) is susceptible to a range of security threats. Although various defense mechanisms have been proposed, they are typically non-adaptive and tailored to specific types of attacks, leaving them insufficient in the face of multiple uncertain, unknown, and adaptive attacks employing diverse strategies. This work formulates adversarial federated learning under a mixture of various attacks as a Bayesian Stackelberg Markov game, based on which we propose the meta-Stackelberg defense composed of pre-training and online adaptation. {The gist is to simulate strong attack behavior using reinforcement learning (RL-based attacks) in pre-training and then design meta-RL-based defense to combat diverse and adaptive attacks.} We develop an efficient meta-learning approach to solve the game, leading to a robust and adaptive FL defense. Theoretically, our meta-learning algorithm, meta-Stackelberg learning, provably converges to the first-order $\varepsilon$-meta-equilibrium point in $O(\varepsilon^{-2})$ gradient iterations with $O(\varepsilon^{-4})$ samples per iteration. Experiments show that our meta-Stackelberg framework performs superbly against strong model poisoning and backdoor attacks of uncertain and unknown types.