Unverified50% confidenceOpinionExact time
If training data contains historical biases against specific ethnicities, income classes, or regions, the model will entrench and scale those biases.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified数据增强从偏差-方差权衡视角看，本质是通过增加训练样本的有效多样性来降低模型方差76% similarUnverifiedMultiple studies in 2024 found that the strong performance of certain models on standard benchmarks resulted from benchmark test questions being mixed into the training data (data contamination).66% similarUnverifiedThe root causes of cross-model citation behavior differences include differences in pre-training corpora, differences in SFT datasets, and differences in the human annotator populations at the RLHF stage.65% similarUnverifiedTraining/serving skew can be further subdivided into three types: data skew, feature transformation skew, and algorithm skew.65% similarUnverifiedPerforming StandardScaler normalization on the entire dataset before the train/test split causes the mean and variance information of the test set to leak into the training set.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/900642API
curl https://kongchang.com/api/v1/knowledge/claims/900642MCP
get_claim(id=900642)