[KongchangAI]
Unverified50% confidenceTradeoffExact time

DIAYN内在奖励尺度若与SAC熵项量级不匹配会导致一方主导优化,应适当缩放判别器奖励使两者量级相当

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/982260
API
curl https://kongchang.com/api/v1/knowledge/claims/982260
MCP
get_claim(id=982260)