Unverified50% confidenceFactExact time
表示坍缩指模型将所有输入映射到同一点或低维流形，使损失趋近于零但毫无意义，是自监督学习的核心挑战
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedModel collapse occurs in two progressive stages: in the early collapse stage the output mean drifts, and in the late collapse stage the output variance approaches zero and generated content converges into a homogeneous, singular pattern.73% similarUnverifiedA reasonable prediction loss combined with 0% downstream control accuracy is highly consistent with representation collapse, meaning the encoder has fooled the self-supervised objective without learning any useful dynamic information.72% similarUnverified在自学习隐空间中预测时，预测质量取决于表示本身质量，由此带来循环依赖问题，容易产生表示坍缩68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/533771API
curl https://kongchang.com/api/v1/knowledge/claims/533771MCP
get_claim(id=533771)