This paper addresses exploration in reinforcement learning (RL) in unknown environments with sparse rewards, focusing on maximum entropy exploration. It introduces a game-theoretic algorithm for visitation entropy maximization with improved sample complexity of O(H^3S^2A/ε^2). For trajectory entropy, the paper presents an algorithm with O(poly(S, A, H)/ε) complexity, showing the statistical advantage of regularized MDPs for exploration. Why it matters: The research offers new techniques to reduce the sample complexity of RL, potentially enhancing the efficiency of AI agents in complex environments.
This paper introduces a decentralized multi-agent decision-making framework for search and action problems under time constraints, treating time as a budgeted resource where actions have costs and rewards. The approach uses probabilistic reasoning to optimize decisions, maximizing reward within the given time. Evaluated in a simulated search, pick, and place scenario inspired by the Mohamed Bin Zayed International Robotics Challenge (MBZIRC), the algorithm outperformed benchmark strategies. Why it matters: The framework's validation in a Gazebo environment signals potential for real-world robotic applications, particularly in time-sensitive and cooperative tasks within the robotics domain in the UAE.
MBZUAI researchers presented FIRE, a new fact-checking framework for LLM outputs, at NAACL 2025. FIRE first assesses the LLM's confidence in its claims before searching the web, reducing computational cost. It also stores knowledge gained from web searches to aid in classifying other claims. Why it matters: This approach improves the efficiency and cost-effectiveness of automatically verifying the accuracy of LLMs, addressing a key limitation in their reliability.
This paper introduces DaringFed, a novel dynamic Bayesian persuasion pricing mechanism for online federated learning (OFL) that addresses the challenge of two-sided incomplete information (TII) regarding resources. It formulates the interaction between the server and clients as a dynamic signaling and pricing allocation problem within a Bayesian persuasion game, demonstrating the existence of a unique Bayesian persuasion Nash equilibrium. Evaluations on real and synthetic datasets demonstrate that DaringFed optimizes accuracy and convergence speed and improves the server's utility.
MBZUAI PhD graduate William de Vazelhes is researching hard-thresholding algorithms to enable AI to work from smaller datasets. His work focuses on optimization algorithms that simplify data, making it easier to analyze and work with, useful for energy-saving and deploying AI models on low-memory devices. He demonstrated that his approach can obtain results similar to those of convex algorithms in many usual settings. Why it matters: This research could broaden AI accessibility by reducing computational costs, and has potential applications in sectors like finance, particularly for portfolio management under budgetary constraints.
A new method is proposed to reduce the verbosity of LLMs in step-by-step reasoning by retaining moderately easy problems during Reinforcement Learning with Verifiable Rewards (RLVR) training. This approach acts as an implicit length regularizer, preventing the model from excessively increasing output length on harder problems. Experiments using Qwen3-4B-Thinking-2507 show the model achieves baseline accuracy with nearly twice shorter solutions.