Unverified50% confidenceFactExact time
2017年Guo等人在ICML发表的《On Calibration of Modern Neural Networks》首次大规模实证校准失当现象，发现ResNet等现代深度网络比早期浅层网络校准更差
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
元认知反馈：用强化学习让大模型准确表达不确定性
hackernewshackernews7/7/2026
Related Claims
VerifiedResNet (Residual Network) was proposed by Microsoft Research's He Kaiming team in 2015 and solved the vanishing gradient problem in deep network training by introducing residual connections.73% similarUnverifiedIn deep networks, traditional single residual connections still suffer from the vanishing gradient problem, resulting in limited effective information learned by deeper layers.70% similarUnverifiedResNet（残差网络）通过跳跃连接解决了深层网络的梯度消失问题，能提取层次化视觉特征68% similarUnverifiedExperiments by Zhang et al. (2017) demonstrated that standard deep neural networks can perfectly memorize training data with random labels, meaning that traditional generalization bounds based on model capacity completely break down.68% similarUnverifiedRecent theoretical research has shown that in over-parameterized deep networks, the quality of most local minima is close to the global optimum, and saddle points are the true obstacle to optimization.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/279890API
curl https://kongchang.com/api/v1/knowledge/claims/279890MCP
get_claim(id=279890)