Unverified50% confidenceFactExact time
当目标数据域与源数据域差异悬殊时，预训练权重可能产生负迁移（Negative Transfer），反而干扰目标域的学习
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedDuring 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.73% similarUnverified在稠密检索中，批内负样本训练要求批次包含有信息量的负样本，随机采样会导致训练损失快速收敛但泛化性能差73% similarUnverifiedData Drift refers to the continuous degradation of model performance without any errors when the statistical properties of real online data deviate from the training set.73% similarUnverified仅用成功轨迹训练的模型往往存在分布外泛化的严重缺陷，遇到工具返回错误时便丧失纠偏能力72% similarUnverifiedTraining on a dataset with an overly homogeneous object type causes the model's performance to drop sharply when faced with objects of other materials or forms, which is the domain shift problem.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/527162API
curl https://kongchang.com/api/v1/knowledge/claims/527162MCP
get_claim(id=527162)