Unverified50% confidenceSolutionExact time
Increasing the entropy coefficient after curriculum learning reaches its upper limit is a "exploit first, then explore" strategy: low entropy in the early stage quickly masters the basic gait, while high entropy in the later stage helps the policy find more robust solutions within the disturbance range it has adapted to.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
Unverified排查SAC训练崩溃应监控策略熵和α变化曲线，若熵持续下降至接近零说明探索消亡，可固定α为较大常数或重设目标熵为-action_dim77% similarUnverifiedEntropy collapse refers to the premature convergence of a model policy's output distribution, resulting in a loss of exploration capability and the model becoming trapped in a local optimum.73% similarUnverified跨分词器蒸馏前应先测量熵保留率或概率质量对齐度以避免在错误目标上浪费算力72% similarUnverifiedEntropy collapse is especially prominent in long Chain-of-Thought training, because reward signals are only given at the end of the reasoning chain and gradient signals at intermediate steps are extremely sparse.70% similarUnverifiedComputing Graph Entropy typically involves constructing a k-nearest-neighbor graph and computing eigenvalues or random-walk entropy; its complexity is higher than simple distance averaging, but it is feasible on medium-sized datasets.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1079729API
curl https://kongchang.com/api/v1/knowledge/claims/1079729MCP
get_claim(id=1079729)