Unverified50% confidenceFactExact time
经过充分训练的神经网络权重矩阵奇异值呈现重尾特征，前10%的奇异值可能捕获超过90%的信息能量
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified当不动动作占据80%以上帧时，模型通过最小化交叉熵损失可在不学习有意义特征的情况下达到80%准确率，这称为多数类捷径67% similarUnverifiedExperimental data shows that when key information is located at the beginning and end of the input sequence, the model's retrieval accuracy can be as high as over 90%, but when located in the middle, it may drop sharply to below 60%.66% similarUnverifiedA new MIT paper shows that pruning techniques can shrink a neural network by 90%, directly reducing inference costs by 10x65% similarUnverified超高思考强度一条消息消耗的token可能是低强度的5-10倍64% similarUnverifiedKnowledge distillation has the student network learn the soft probability distribution output by the teacher network, enabling the student model to retain core reasoning capabilities while reducing its parameter count by more than 10 times.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561357API
curl https://kongchang.com/api/v1/knowledge/claims/561357MCP
get_claim(id=561357)