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