The Instability in Reinforcement Learning
A fresh perspective on flow-matching policies reveals the real culprit behind training instability in reinforcement learning. VINE offers a promising solution.
Read the full articleYou might also wanna read
Understanding Reinforcement Learning for Model Training, and future directions with GRAPE
This paper provides a self-contained, from-scratch, exposition of key algorithms for instruction tuning of models: SFT, Rejection Sampling,
arxiv.org·9mo ago
GFlowRL: A Scalable Distribution-Matching RL Algorithm for Large Language Models
Generative Flow Networks (GFlowNets) offer a promising alternative to reward-maximizing reinforcement learning (RL) for large reasoning mode
arxiv.org·3h ago
River: "‼️I wrote a new blog post‼️ "An Exploration into…" - DEF CON Social
‼️I wrote a new blog post‼️ "An Exploration into Reinforcement Learning" I talk about how RL is different from modern generative "AI" system
defcon.social·1mo agoReinforcement Learning to Train Large Language Models to Explain Human Decisions
Article URL: Comments URL: Points: 6 # Comments: 0
RL Systems Mind the Gap: Matching Trainer and Generator Throughput
RL Training Infrastructure, GRPO, PipelineRL, Async RL, Policy Staleness, RL Sandbox Infra, CPU Requirements, TCO Analysis, Thinking Machine
Optimistic Thompson Sampling for No-Regret Learning in Unknown Games
We study the problem of learning to play a repeated multi-player game with an unknown reward function and bandit feedback. The central chall
Comments
Sign in to join the conversation.
No comments yet. Be the first.