Classification on IMDB
100AccuracyRole-Aware AI
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Role-Aware AIGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 100 | — | 100 | — | — | — | — | |
| LogLikelihoodGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 100 | — | 100 | — | — | — | — | |
| LogRankGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 99 | — | 100 | — | — | — | — | |
| EntropyGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 98 | — | 99 | — | — | — | — | |
| LogRankGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 97 | — | 100 | — | — | — | — | |
| RoBERTaGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 95 | — | 98 | — | — | — | — | |
| Role-Aware AIGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 95 | — | 99 | — | — | — | — | |
| RoBERTaGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 92 | — | 96 | — | — | — | — | |
| LogLikelihoodGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 88 | — | 95 | — | — | — | — | |
| DIRECTORBase architecture=KEP-2/5, Ensemble=Deep Ensembles2026.02 | 87.88 | 3.43 | 82.96 | 72.29 | 1.24 | 2.91 | 17.7 | |
| DIRECTORBase architecture=KEP-1/5, Ensemble=Deep Ensembles2026.02 | 87.85 | 3.44 | 83.2 | 71.1 | 1.25 | 2.89 | 17.62 | |
| KEP-2/5Backbone=KEP, GP-reparameterized blocks=2/5, Ensemble=Deep Ensembles2026.02 | 87.77 | 3.51 | 82.44 | 73.14 | 2.85 | 3.04 | 18.18 | |
| DIRECTORBase architecture=KEP-5/5, Ensemble=Deep Ensembles2026.02 | 87.66 | 3.43 | 82.95 | 72.13 | 2.07 | 2.95 | 18.05 | |
| DIRECTORBase architecture=Transformer, Ensemble=Deep Ensembles2026.02 | 87.45 | 3.56 | 82.83 | 70.68 | 1.97 | 3 | 18.25 | |
| KEP-1/5Backbone=KEP, GP-reparameterized blocks=1/5, Ensemble=Deep Ensembles2026.02 | 87.44 | 3.63 | 82.34 | 72.21 | 1.56 | 3.01 | 18.39 | |
| TransformerBackbone=Transformer, Ensemble=Deep Ensembles2026.02 | 87.31 | 3.55 | 82.92 | 71.45 | 2.36 | 3.04 | 18.44 | |
| CurvatureGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 87 | — | 94 | — | — | — | — | |
| KEP-5/5Backbone=KEP, GP-reparameterized blocks=5/5, Ensemble=Deep Ensembles2026.02 | 86.49 | 4.16 | 81.54 | 74.07 | 2.63 | 3.28 | 19.72 | |
| CurvatureGenerator Model=LLaMA-3-Instruct, Generator Parameters=3 billion2026.05 | 86 | — | 91 | — | — | — | — | |
| Transformer (Full Attn)Source=This work, Tokenization=GPT-2 BPE2026.06 | 84.96 | — | — | — | — | — | — | |
| CHIAR DCT+AttnSource=This work, Tokenization=GPT-2 BPE2026.06 | 83.72 | — | — | — | — | — | — | |
| EntropyGenerator Model=GPT-2, Generator Parameters=124 million2026.05 | 83 | — | 91 | — | — | — | — | |
| CDVM10kTraining Data Percentage=30% Data2026.05 | 81.3 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=25% Data2026.05 | 80.8 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=30% Data2026.05 | 80.4 | — | — | — | — | — | — | |
| CDVM10kTraining Data Percentage=25% Data2026.05 | 80 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=25% Data2026.05 | 79.5 | — | — | — | — | — | — | |
| CDVM10kTraining Data Percentage=20% Data2026.05 | 79.4 | — | — | — | — | — | — | |
| CDVM10kTraining Data Percentage=15% Data2026.05 | 79.4 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=30% Data2026.05 | 79.4 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=30% Data2026.05 | 79.4 | — | — | — | — | — | — | |
| CDVM10kTraining Data Percentage=10% Data2026.05 | 79.3 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=20% Data2026.05 | 78.9 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=10% Data2026.05 | 78.9 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=25% Data2026.05 | 78.8 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=15% Data2026.05 | 78.7 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=10% Data2026.05 | 78.7 | — | — | — | — | — | — | |
| CDVM10kTraining Data Percentage=5% Data2026.05 | 78.3 | — | — | — | — | — | — | |
| CDVM5kTraining Data Percentage=5% Data2026.05 | 78.2 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=20% Data2026.05 | 78.2 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=15% Data2026.05 | 78.2 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=30% Data2026.05 | 78.2 | — | — | — | — | — | — | |
| DataOOBTraining Data Percentage=5% Data2026.05 | 78 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=20% Data2026.05 | 77.4 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=15% Data2026.05 | 77.4 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=30% Data2026.05 | 77.2 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=25% Data2026.05 | 77.2 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=20% Data2026.05 | 76.5 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=25% Data2026.05 | 75.8 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=15% Data2026.05 | 75.1 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=10% Data2026.05 | 74.8 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=20% Data2026.05 | 74.8 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=15% Data2026.05 | 73.9 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=10% Data2026.05 | 72.9 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=10% Data2026.05 | 72.1 | — | — | — | — | — | — | |
| InfOptTraining Data Percentage=5% Data2026.05 | 68 | — | — | — | — | — | — | |
| BanzhafTraining Data Percentage=5% Data2026.05 | 67.8 | — | — | — | — | — | — | |
| RandomTraining Data Percentage=5% Data2026.05 | 67.5 | — | — | — | — | — | — | |
| PerformerSource=Tay et al. 2021, Tokenization=Character-level2026.06 | 65.4 | — | — | — | — | — | — | |
| FNetSource=Tay et al. 2021, Tokenization=Character-level2026.06 | 65.11 | — | — | — | — | — | — | |
| TransformerSource=Tay et al. 2021, Tokenization=Character-level2026.06 | 64.27 | — | — | — | — | — | — | |
| LinformerSource=Tay et al. 2021, Tokenization=Character-level2026.06 | 53.94 | — | — | — | — | — | — |