Unverified50% confidenceDecision RuleExact time
The recommended starting alpha value is 0.10; a stronger alpha value does not mean better results, the degree of variation differs across seeds, and this method is not a universal improvement.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified模型的基准分数更高并不等于在所有场景下都更好用68% similarUnverified对比学习中的推远操作只要求负样本相似度低于正样本，并不要求达到-1的极端值66% similarUnverifiedIn actual alpha factor research, a single factor whose IC can consistently reach 0.03-0.05 is already considered very excellent.64% similarVerifiedAccording to common industry standards, AUC > 0.9 is excellent, 0.8–0.9 is good, 0.7–0.8 is fair, 0.5–0.7 is poor, and 0.5 is equivalent to random guessing.63% similarUnverifiedThe power-law exponent between test loss and compute is approximately 0.05, meaning that every 10x increase in compute yields approximately a 12% improvement in model performance.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/911897API
curl https://kongchang.com/api/v1/knowledge/claims/911897MCP
get_claim(id=911897)