Unverified50% confidenceFactExact time
Anti-collapse strategies include EMA target encoders (e.g., BYOL, DINO), VICReg variance-covariance regularization, and Barlow Twins feature whitening.
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
复现TS-JEPA踩坑：从0%到74%控制精度的差距在哪
Related Claims
UnverifiedThe choice of momentum coefficient for the EMA target encoder is extremely sensitive: too high and the target encoder barely updates, too low and it degenerates into an autoencoder, both of which can lead to collapse.73% similarUnverifiedMACD captures trend reversals through the difference between fast and slow EMAs.60% similarUnverifiedDAgger算法将累积误差从O(T²)降至O(T)，效果显著优于纯行为克隆，专门针对分布漂移问题设计60% similarUnverifiedEmbedding fine-tuning improves recall coverage, while reranking improves precision ranking accuracy; the combination of the two is the standard solution for production-grade RAG systems.58% similarUnverifiedOOSM 除回溯重放外还有 Algorithm A（用当前增益近似处理）和 Algorithm B（基于储存协方差进行精确后验修正）等方案，各有计算复杂度与精度权衡56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/756922API
curl https://kongchang.com/api/v1/knowledge/claims/756922MCP
get_claim(id=756922)