Theory of Mind vs Self-Attribution in Large Language Models
Explore how Theory of Mind and self-attributions differ in LLMs and the impact of safety fine-tuning on their socio-cognitive abilities.
Beta-Scheduling: Boost Neural Network Training Efficiency
Discover beta-scheduling, a momentum technique using critical damping to speed up neural network training and improve diagnostic accuracy.
Robust Multi-Agent RL for sUAS GPS Spoofing Defense
Enhance small UAS safety with robust multi-agent reinforcement learning against GPS degradation and spoofing threats in high-density operations.
OccSim: Scalable Multi-Kilometer 3D Driving Simulation
OccSim enables scalable multi-kilometer 3D driving simulation using long-horizon occupancy world models without HD maps or driving logs.
OptiMer: Efficient Vector Merging for Continual Pre-Training
Discover OptiMer, a novel method that optimizes distribution vector merging to enhance continual pre-training efficiency and model adaptation.