Sentiment Classification on SST-5
70.67AccuracyLS
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LSCalibration category=Unlearnable, Backbone=T52022.10 | 70.67 | 73.5 | 5.55 | 22.53 | 63.95 | |
| I-IterCalibration category=Learnable, Backbone=T52022.10 | 70.67 | 33.17 | 38.49 | 60.59 | 18.13 | |
| VanillaCalibration category=Unlearnable, Backbone=T52022.10 | 69.73 | 82.78 | 13.52 | 12.3 | 71.72 | |
| TSCalibration category=Unlearnable, Backbone=T52022.10 | 69.73 | 71.98 | 4.94 | 23.01 | 60.69 | |
| E-MLPCalibration category=Learnable, Backbone=T52022.10 | 69.73 | 84.06 | 14.73 | 16.04 | 84.28 | |
| E-T5Calibration category=Learnable, Backbone=T52022.10 | 69.73 | 35.23 | 38.72 | 57.7 | 18.95 | |
| I-SimulCalibration category=Learnable, Backbone=T52022.10 | 69.54 | 31.45 | 38.26 | 61.73 | 15.88 | |
| EnsembleCalibration category=Unlearnable, Backbone=T52022.10 | 69.35 | 83 | 13.66 | 12.13 | 72 | |
| I-VanillaCalibration category=Learnable, Backbone=T52022.10 | 68.23 | 32.7 | 38.85 | 58.35 | 13.49 | |
| EDACalibration category=Unlearnable, Backbone=T52022.10 | 67.67 | 87.58 | 20.2 | 7.97 | 78.27 | |
| BOOSTAUGModel=DeBERTa, Base Augmentation=EDA2022.10 | 57.78 | — | — | — | — | |
| VanillaModel=FastLCF2022.10 | 56.59 | — | — | — | — | |
| VanillaModel=DeBERTa2022.10 | 56.47 | — | — | — | — | |
| GRZOBackbone=RoBERTa-large (350M), k-shot setting=512, Optimization Protocol=Zeroth-Order, Training Steps=20k, Batch Size=16, Precision=FP162026.06 | 54.8 | — | — | — | — | |
| FZOOBackbone=RoBERTa-large (350M), k-shot setting=512, Optimization Protocol=Zeroth-Order, Training Steps=20k, Batch Size=16, Precision=FP162026.06 | 54.2 | — | — | — | — | |
| VanillaModel=BERT2022.10 | 53.53 | — | — | — | — | |
| MeZOBackbone=RoBERTa-large (350M), k-shot setting=512, Optimization Protocol=Zeroth-Order, Training Steps=20k, Batch Size=16, Precision=FP162026.06 | 53.2 | — | — | — | — | |
| LP(FO)Backbone=RoBERTa-large (350M), k-shot setting=512, Optimization Protocol=Linear Probing (First-Order), Training Steps=20k, Batch Size=16, Precision=FP162026.06 | 51.7 | — | — | — | — | |
| FastGASAnnotation budget (|L|)=1002024.06 | 50.26 | — | — | — | — | |
| KiteBackbone=Qwen 2.5–1.5B2025.09 | 49.59 | — | — | — | — | |
| Llama3.2-1BForgetting Task=none2025.05 | 49.55 | — | — | — | — | |
| CEILBackbone=Qwen 2.5–1.5B2025.09 | 47.88 | — | — | — | — | |
| BM25Backbone=Qwen 2.5–1.5B2025.09 | 47.86 | — | — | — | — | |
| BM25Backbone=Llama 3.2–3B2025.09 | 47.85 | — | — | — | — | |
| KiteBackbone=Llama 3.2–3B2025.09 | 47.59 | — | — | — | — | |
| Vanilla IPABackbone=RoBERTa-large2026.03 | 47.5 | — | — | — | — | |
| FT(FO)Backbone=RoBERTa-large (350M), k-shot setting=512, Optimization Protocol=Fine-Tuning (First-Order), Training Steps=20k, Batch Size=16, Precision=FP162026.06 | 47.5 | — | — | — | — | |
| EPR + IDAICLPLM=GPT-Neo, m=122024.06 | 47.2 | — | — | — | — | |
| DenseBackbone=Qwen 2.5–1.5B2025.09 | 46.64 | — | — | — | — | |
| FastGASAnnotation budget (|L|)=182024.06 | 46.61 | — | — | — | — | |
| CEILBackbone=Llama 3.2–3B2025.09 | 46.45 | — | — | — | — | |
| Original TextCNNType=ANN, Time Steps=−2026.05 | 45.48 | — | — | — | — | |
| VanillaModel=LSTM2022.10 | 45.29 | — | — | — | — | |
| SEETOKText Source=Visual-Text, 5-shot sampling=true2025.10 | 44.4 | — | — | — | — | |
| DenseBackbone=Llama 3.2–3B2025.09 | 44.14 | — | — | — | — | |
| EPR + IDAICLPLM=GPT-2 0.8B, m=122024.06 | 43.9 | — | — | — | — | |
| Tailored TextCNNType=ANN, Time Steps=−2026.05 | 43.48 | — | — | — | — | |
| Stiefel LowRank-LRBackbone=RoBERTa-large2026.03 | 43.2 | — | — | — | — | |
| MetaICL + IDAICLPLM=GPT-2 0.8B, m=122024.06 | 42.6 | — | — | — | — | |
| Coordinate LowRank-LRBackbone=RoBERTa-large2026.03 | 42.6 | — | — | — | — | |
| RandomBackbone=Llama 3.2–3B2025.09 | 42.41 | — | — | — | — | |
| OursType=Direct Training, Time Steps=502026.05 | 41.9 | — | — | — | — | |
| Channel ICL + IDAICLPLM=GPT-2 0.8B, m=122024.06 | 41.8 | — | — | — | — | |
| Conv SNN + FTType=Hybrid, Time Steps=502026.05 | 41.63 | — | — | — | — | |
| Conv SNNType=ANN2SNN, Time Steps=502026.05 | 41.4 | — | — | — | — | |
| IDAICLPLM=GPT-2 1.5B, m=122024.06 | 41.1 | — | — | — | — | |
| Gaussian LowRank-LRBackbone=RoBERTa-large2026.03 | 41.1 | — | — | — | — | |
| Vanilla LRBackbone=RoBERTa-large2026.03 | 40.8 | — | — | — | — | |
| KiteBackbone=GPT-Neo 2.7B2025.09 | 40.6 | — | — | — | — | |
| RandomBackbone=Qwen 2.5–1.5B2025.09 | 40.53 | — | — | — | — | |
| IDAICLPLM=GPT-2 0.8B, m=122024.06 | 40.1 | — | — | — | — | |
| IDAICLPLM=GPT-2 0.8B, m=82024.06 | 39.6 | — | — | — | — | |
| IDAICLPLM=GPT-2 0.8B, m=42024.06 | 38.3 | — | — | — | — | |
| PICLShot=4-shot, Model=GPT2-XL (1.5B)2023.05 | 38 | — | — | — | — | |
| CEILBackbone=GPT-Neo 2.7B2025.09 | 36.84 | — | — | — | — | |
| BM25Backbone=GPT-Neo 2.7B2025.09 | 36.05 | — | — | — | — | |
| DenseBackbone=GPT-Neo 2.7B2025.09 | 35.96 | — | — | — | — | |
| PICLShot=4-shot, Model=GPT-Neo (2.7B)2023.05 | 35.7 | — | — | — | — | |
| Zero-shotBackbone=RoBERTa-large2026.03 | 35.5 | — | — | — | — | |
| Zero-shotBackbone=RoBERTa-large (350M), k-shot setting=5122026.06 | 35.5 | — | — | — | — | |
| MetaICLShot=4-shot, Model=GPT2-XL (1.5B)2023.05 | 34.6 | — | — | — | — | |
| MetaICLShot=4-shot, Model=GPT-Neo (2.7B)2023.05 | 32.8 | — | — | — | — | |
| VanillaICLShot=4-shot, Model=GPT2-XL (1.5B)2023.05 | 32.4 | — | — | — | — | |
| RandomBackbone=GPT-Neo 2.7B2025.09 | 32.31 | — | — | — | — | |
| VanillaICLShot=4-shot, Model=GPT-Neo (2.7B)2023.05 | 32.1 | — | — | — | — | |
| Qwen2.5-VL 3BText Source=Pure-Text, 5-shot sampling=true2025.10 | 28.8 | — | — | — | — | |
| Qwen2.5-VL 3BText Source=Visual-Text, 5-shot sampling=true2025.10 | 25.21 | — | — | — | — | |
| Directly-trained SNNType=Direct Training, Time Steps=502026.05 | 23.08 | — | — | — | — | |
| LWFForgetting Task=gsm8k2025.05 | 2.83 | — | — | — | — | |
| LWFForgetting Task=qasc2025.05 | 2.54 | — | — | — | — | |
| LWFForgetting Task=dental2025.05 | 2.1 | — | — | — | — | |
| LWFForgetting Task=mixed2025.05 | 2.1 | — | — | — | — | |
| LWFForgetting Task=psychol2025.05 | 1.27 | — | — | — | — |