Unverified50% confidenceFactExact time
InfoNCE损失最早由van den Oord等人于2018年在时序数据表示学习中提出
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedInfoNCE损失的理论根源可追溯至DeepMind的van den Oord等人于2018年提出的CPC（Contrastive Predictive Coding）80% similarUnverifiedThe core loss functions of contrastive learning (such as InfoNCE and NT-Xent) essentially pull the representations of similar samples closer together while pushing the representations of dissimilar samples further apart.65% similarVerifiedThe Knowledge Cutoff of training data causes the model's knowledge of actively iterated open-source projects to often lag behind the actual version by months or even years63% similarUnverifiedLoss Scaling最初由NVIDIA在2018年提出，用于解决FP16训练中梯度下溢问题61% similarUnverifiedThe pre-training stage uses Cross-Entropy Loss to measure the gap between the predicted distribution and the actual next token.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/520165API
curl https://kongchang.com/api/v1/knowledge/claims/520165MCP
get_claim(id=520165)