Unverified85% confidenceTradeoffExact time
On class-imbalanced datasets, Macro F1 is extremely sensitive to rare class performance, where small threshold changes cause dramatic fluctuations in precision and recall for rare classes
1
Sources
85%
Confidence
Long-term
Relevance
8/2/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在类别不平衡场景下单纯的准确率可能具有误导性，需要结合混淆矩阵进行更细粒度评估62% similarUnverified权重初始化的随机性对模型性能影响最深远，数据批次打乱顺序的随机性影响最小61% similarUnverified温度超参数τ越小分布越集中，模型对难负样本越敏感，但τ过小会导致梯度对异常值敏感、训练不稳定61% similarUnverified长尾分布是指少数类别占据大量样本，而大量类别仅有极少样本的数据分布特征，会导致模型对头部类别过拟合、对尾部类别预测准确率极低60% similarUnverified量化的核心挑战在于处理权重分布中的异常值，LLM权重中少量极大值会显著放大量化误差59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/680250API
curl https://kongchang.com/api/v1/knowledge/claims/680250MCP
get_claim(id=680250)