Verified65% confidenceFactExact time
In 2019, Belkin et al. published a paper in PNAS formally describing the double descent phenomenon, whereby test error begins to decrease again once the number of model parameters exceeds the interpolation threshold.
3
Sources
65%
Confidence
Long-term
Relevance
8/13/2026
First Seen
Sources
泛化才是机器学习的核心：训练前决定成败的关键因素
Related Claims
VerifiedThe double descent phenomenon indicates that when the number of model parameters far exceeds the amount of data, the test error may actually decrease again.75% similarUnverifiedThe double descent phenomenon occurs not only along the model size dimension, but can also appear with changes in training epochs and data volume.66% 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% similarUnverifiedInference frameworks such as vLLM and TensorRT-LLM improve throughput through batching and KV Cache techniques, reducing the per-inference cost of models with tens of billions of parameters by an order of magnitude over the past three years65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/742534API
curl https://kongchang.com/api/v1/knowledge/claims/742534MCP
get_claim(id=742534)