Unverified50% confidenceFactExact time
Li等人2018年发现神经网络的有效参数空间维度远低于名义参数量，训练轨迹被限制在低维子空间中
1
Sources
50%
Confidence
Long-term
Relevance
8/6/2026
First Seen
Sources
Related Claims
Unverified深度神经网络天然倾向于过拟合训练分布，对分布外样本的处理能力往往大幅下降80% similarUnverifiedNeural networks have an overconfidence defect: even when inputs fall outside the training distribution, the Softmax layer may still output inflated confidence scores.74% similarUnverifiedLeCun批评自回归LLM在系统2慢思考上存在根本性局限、对物理世界缺乏基础、token预测训练信号是平面的72% similarUnverified梯度学习的核心是连续优化，神经网络参数空间是高维实数空间，损失函数近似可微72% similarUnverifiedDuring cross-domain fine-tuning, an excessively large learning rate will destroy pre-trained weights, while an excessively small one will result in almost no learning, requiring a careful trade-off.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/694287API
curl https://kongchang.com/api/v1/knowledge/claims/694287MCP
get_claim(id=694287)