Unverified50% confidenceFactExact time
Entropy 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.
1
Sources
50%
Confidence
Long-term
Relevance
9/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedScaling Laws表明模型的交叉熵损失会随着计算量、数据量和参数量的增加而平滑下降，且这种下降关系可以用简单的幂律函数精确拟合76% similarUnverified标准交叉熵损失在目标占图像面积较小时会导致严重的类别不平衡问题，模型倾向于全部预测为背景75% similarUnverified如果交叉熵稳步下降而Dice波动剧烈，问题几乎可以确定出在Dice损失的配置上74% similarUnverifiedThe model uses class-weighted cross-entropy loss, assigning each class a weight inversely proportional to its frequency to mitigate data imbalance.71% similarUnverifiedDuring model collapse, the model successively experiences a reduction in output diversity, loss of minority categories and edge knowledge, and ultimately produces meaningless or highly repetitive outputs.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937677API
curl https://kongchang.com/api/v1/knowledge/claims/937677MCP
get_claim(id=937677)