Craftax
PulseAugur coverage of Craftax — every cluster mentioning Craftax across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
New RL scaling laws discovered, improving performance prediction
Researchers have identified empirical scaling laws for reinforcement learning (RL) in two environments, Starpilot and Craftax. These laws accurately predict performance, resource allocation, and hyperparameter configura…
-
New framework uses LLMs to generate code for AI agent training
Researchers have developed a new framework called Dreaming in Code (DiCode) to improve open-ended learning for AI agents. This unsupervised environment design framework uses large language models to generate executable …
-
New research targets ethical RL agents by focusing on per-episode violations
A new research paper proposes training reinforcement learning agents to exhibit ethical behavior by focusing on per-episode distributions rather than average performance. The study compares four training methods within …
-
New Error Diffusion method enables biologically plausible AI learning
Researchers have developed a novel method called modulo error routing to extend Error Diffusion (ED) for use in biologically plausible dual-stream neural networks that adhere to Dale's principle. This approach allows fo…
-
New Method Discovers Hierarchical Skills in Reinforcement Learning
Researchers have developed a new method for unsupervised skill discovery and hierarchical structure learning in reinforcement learning. This approach segments unlabelled trajectories into skills and organizes them into …
-
AI agents learn to preemptively solve future problems using multitask preplay
Researchers have introduced a novel algorithm called Multitask Preplay, which models how humans use experience from one task to preemptively learn solutions for other, related tasks. This method involves simulating acce…
-
PACE method improves reinforcement learning generalization via parameter change evaluation
Researchers have introduced PACE, a novel approach to Unsupervised Environment Design (UED) for enhancing reinforcement learning generalization. PACE directly measures an environment's value by assessing the policy para…