Unverified50% confidenceSolutionExact time
Negative samples for contrastive learning fine-tuning should use hard negatives, such as positions in the same field but at different levels, rather than random negative samples.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对比学习中的推远操作只要求负样本相似度低于正样本，并不要求达到-1的极端值77% similarUnverified困惑度是测试集上每个token平均负对数似然的指数，值越低代表模型对语料的预测越准确74% similarUnverified评估基准测试质量的关键指标是假阳性率和假阴性率，理想基准应将两者控制在较低水平73% similarUnverifiedInfoNCE 损失通过在一批候选中最大化正样本对相似度、最小化负样本对相似度进行训练，温度超参数τ越小分布越集中69% similarUnverifiedIn scenarios with extreme class imbalance (such as fraud detection), looking solely at accuracy as an evaluation metric is completely meaningless.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/829669API
curl https://kongchang.com/api/v1/knowledge/claims/829669MCP
get_claim(id=829669)