[KongchangAI]
Unverified50% confidenceFactExact time

经验回放池中旧数据与新策略的分布偏移会使Bellman目标估计方差上升,可能触发critic梯度爆炸或Q值发散

1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen

Sources

SAC训练崩溃之谜:策略熵坍塌与DIAYN调试指南

Related Entities

Cite This Claim

Stable URI
https://kongchang.com/claim/982255
API
curl https://kongchang.com/api/v1/knowledge/claims/982255
MCP
get_claim(id=982255)