Concentration of target information on GPT-2 Small (test 1)
0.33Gini CoefficientIdentity
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Identitymatrix=identity matrix2025.08 | 0.33 | 0.05 | 0.23 | |
| Randommatrix=random orthogonal matrix2025.08 | 0.36 | 0.1 | 0.11 | |
| PCA 1eigenvector_sorting=smallest subspace receives biggest variance2025.08 | 0.43 | 0.42 | 0.37 | |
| Featureclustering_source=SAE features2025.08 | 0.6 | 0.44 | 0.6 | |
| PCA 2eigenvector_sorting=smallest subspace receives smallest variance2025.08 | 0.66 | 0.56 | 0.39 | |
| NDM2025.08 | 0.89 | 0.89 | 0.9 |