Unverified50% confidenceFactExact time
双重下降现象为OpenAI等机构提出的神经网络缩放定律（Scaling Laws）提供了理论基础
1
Sources
50%
Confidence
Long-term
Relevance
8/6/2026
First Seen
Sources
Related Claims
UnverifiedDouble Descent refers to the phenomenon where over-parameterized neural networks should theoretically suffer from severe overfitting but in practice exhibit good generalization ability, and it remains an active area of research.70% similarUnverifiedOpenAI在2019年的研究中证实双重下降现象同时存在于模型复杂度、训练时间和样本数量三个维度65% similarUnverifiedOpenAI、DeepMind等机构的研究表明模型性能随参数量、数据量和计算量的增加呈幂律提升（Scaling Law）63% similarUnverifiedThe choice of bottleneck dimensionality is essentially a trade-off between information compression rate and reconstruction quality, corresponding to rate-distortion theory; Shannon proved in 1959 that there exists a theoretical minimum coding rate lower bound for a given distortion tolerance.62% similarUnverified网关的自动重试和降级机制可能在主模型超时后降级到价格更高的备用模型，导致账单在无预警情况下翻数倍60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/693826API
curl https://kongchang.com/api/v1/knowledge/claims/693826MCP
get_claim(id=693826)