UP: Unbounded Positive Asymmetric Optimization for Breaking the Exploration-Stability Dilemma
Reinforcement learning frameworks for large language models face exploration-stability trade-offs, which are addressed through a novel universal objective called Unbounded Positive…
Hugging Face · Daily Papers
·Chongyu Fan, Pengfei Liu
·
·▲ 1 upvotes
Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.
Autores: Chongyu Fan, Pengfei Liu, Jingjia Huang, Sijia Liu, Yi Lin
- 1 upvotes da comunidade
- Temas: reinforcement learning, importance sampling, exploration-stability dilemma, policy update budget, probability capacity, conservative clipping
Resumo
Resumo original (em inglês), extraído do paper:
Reinforcement learning frameworks for large language models face exploration-stability trade-offs, which are addressed through a novel universal objective called Unbounded Positive Asymmetric Optimization that enables stable training with enhanced exploration capabilities.Onde ler
// relacionados
Leia também
Editorial
GPT-Live-1: a OpenAI aposenta o walkie-talkie e faz a voz ouvir e falar ao mesmo tempo
Editorial
LingBot-VLA 2.0: a Ant abre um "cérebro" de 6B treinado em 60 mil horas de robôs reais
Editorial
UniClawBench: no mundo real, nenhum agente de IA passa da metade das tarefas
Blog