Unverified50% confidenceFactExact time
The class-mean direction outperformed the null hypotheses of random subspace, label permutation, and covariance matching in balanced accuracy in all 12 model-label tests.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe evaluation adopted a paired-sample design in which each type of flaw had one flawed version and one structurally identical but problem-free twin version, used to measure the model's false-positive rate.72% similarUnverifiedThe standard for checking whether a fine-tuning comparison study contains bias is to examine the hyperparameter search section and confirm that each control group has an independent learning rate tuning record and a sufficient number of random seed repetitions.69% similarUnverifiedThe evaluation constructed test items targeting 7 common validity flaws in machine learning research, including treating synthetic data as real evidence, test-set leakage, over-generalization from small samples, reporting accuracy under class imbalance, degenerate confidence intervals, uncorrected multiple comparisons, and best-of-N seed selection.68% similarUnverifiedThe experiments were tested across six heterogeneous benchmarks.68% similarUnverifiedThe experiment used balanced accuracy as the evaluation metric, covering 8 classification datasets.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153326API
curl https://kongchang.com/api/v1/knowledge/claims/1153326MCP
get_claim(id=1153326)