Text Classification on AG News (test)
94.7AccuracyBERT-base w/ NS-LF
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| BERT-base w/ NS-LFBackbone=BERT-base, NS-LF strategy=true2026.04 | 94.7 | — | — | — | — | — | — | — | |
| PLMsupervision=Supervised, training_data=100%2022.03 | 94.4 | — | — | — | — | — | — | — | |
| BERT-baseBackbone=BERT-base, NS-LF strategy=false2026.04 | 94.4 | — | — | — | — | — | — | — | |
| CADStudent Encoder=BERT-base, Augmentation Model=T5-base2026.03 | 94.3 | — | — | — | — | — | — | — | |
| CHARFORMER SBaseTokenization=Character, Parameters=134M2021.06 | 94.1 | — | — | — | — | — | — | — | |
| CHARFORMER BaseTokenization=Character, Parameters=203M2021.06 | 94 | — | — | — | — | — | — | — | |
| ABLEStudent Encoder=BERT-base, Augmentation Model=T5-base2026.03 | 93.92 | — | — | — | — | — | — | — | |
| MASlambda_a=0.3, lambda_s=0.72020.11 | 93.86 | — | — | — | — | — | — | — | |
| MASlambda_a=0.4, lambda_s=0.62020.11 | 93.82 | — | — | — | — | — | — | — | |
| MASlambda_a=0.5, lambda_s=0.52020.11 | 93.72 | — | — | — | — | — | — | — | |
| Byte-level T5 BaseTokenization=Byte, Parameters=200M2021.06 | 93.6 | — | — | — | — | — | — | — | |
| MASlambda_a=0.6, lambda_s=0.42020.11 | 93.55 | — | — | — | — | — | — | — | |
| T5 Base, SubwordTokenization=Subword, Parameters=220M2021.06 | 93.5 | — | — | — | — | — | — | — | |
| Byte-level T5+LASC BaseTokenization=Byte, Architecture=LASC, Parameters=205M2021.06 | 93.5 | — | — | — | — | — | — | — | |
| RETVecModel=BERT2023.02 | 93.5 | — | — | — | — | — | — | — | |
| fastTextModel=BERT2023.02 | 93.4 | — | — | — | — | — | — | — | |
| MASlambda_a=0.7, lambda_s=0.32020.11 | 93.19 | — | — | — | — | — | — | — | |
| RETVec-rawModel=BERT2023.02 | 93.1 | — | — | — | — | — | — | — | |
| FedAvg + Post-trainModel=FLAVA2025.09 | 93.05 | — | — | — | — | — | — | — | |
| Per-FedAvg + Post-trainModel=FLAVA2025.09 | 93.03 | — | — | — | — | — | — | — | |
| FedAvgModel=FLAVA2025.09 | 93.02 | — | — | — | — | — | — | — | |
| EXAMModel Type=Word-based2018.11 | 93 | — | — | — | — | — | — | — | |
| DisentAFLModel=FLAVA2025.09 | 92.96 | — | — | — | — | — | — | — | |
| DisentAFL + Post-trainModel=FLAVA2025.09 | 92.86 | — | — | — | — | — | — | — | |
| TAPModel=FLAVA2025.09 | 92.83 | — | — | — | — | — | — | — | |
| W.C RegionEmbModel Type=Word-based2018.11 | 92.8 | — | — | — | — | — | — | — | |
| FedDAT + Post-trainModel=FLAVA2025.09 | 92.8 | — | — | — | — | — | — | — | |
| fastTextModel=RNN2023.02 | 92.7 | — | — | — | — | — | — | — | |
| SWEM-concat2018.05 | 92.66 | — | — | — | — | — | — | — | |
| AdamW2020.11 | 92.62 | — | — | — | — | — | — | — | |
| Per-FedAvgModel=FLAVA2025.09 | 92.62 | — | — | — | — | — | — | — | |
| RETVecModel=RNN2023.02 | 92.6 | — | — | — | — | — | — | — | |
| FedDATModel=FLAVA2025.09 | 92.6 | — | — | — | — | — | — | — | |
| Adamlambda_a=1, lambda_s=02020.11 | 92.55 | — | — | — | — | — | — | — | |
| fastText (bigram)features=bigram2018.05 | 92.5 | — | — | — | — | — | — | — | |
| Bigram-FastTextModel Type=Word-based2018.11 | 92.5 | — | — | — | — | — | — | — | |
| WhitespaceModel=CNN2023.02 | 92.5 | — | — | — | — | — | — | — | |
| SWEM-hier2018.05 | 92.48 | — | — | — | — | — | — | — | |
| N-grams TFIDFModel Type=Feature Engineering2018.11 | 92.4 | — | — | — | — | — | — | — | |
| SentencePieceModel=CNN2023.02 | 92.4 | — | — | — | — | — | — | — | |
| Char-CNNDepth=6, #Conv Params [M]=1.37, #FC Params [M]=11.34, #Total Params [M]=12.71, Storage Size [MB]=43.252019.01 | 92.36 | — | — | — | — | — | — | — | |
| fastTextModel=CNN2023.02 | 92.3 | — | — | — | — | — | — | — | |
| SCPLArchitecture=LSTM2026.01 | 92.12 | — | — | — | — | — | — | — | |
| BPEModel=BERT2023.02 | 92.1 | — | — | — | — | — | — | — | |
| N-gramsModel Type=Feature Engineering2018.11 | 92 | — | — | — | — | — | — | — | |
| BPArchitecture=LSTM2026.01 | 91.97 | — | — | — | — | — | — | — | |
| Transformer-2L (Direct)Model=Transformer-2L, #Params=378,372, Training Protocol=Direct2026.04 | 91.86 | — | — | — | — | — | — | — | |
| Transformer-2LEvaluation Protocol=Direct training2026.04 | 91.86 | — | — | — | — | — | — | — | |
| OoMMixTrain size=10K2021.05 | 91.83 | — | — | — | — | — | — | — | |
| SWEM-max2018.05 | 91.79 | — | — | — | — | — | — | — | |
| FRAGEMethod variant=with FRAGE2018.09 | 91.73 | — | — | — | — | — | — | — | |
| SWEM-aver2018.05 | 91.71 | — | — | — | — | — | — | — | |
| SCPLArchitecture=Transformer2026.01 | 91.64 | — | — | — | — | — | — | — | |
| Non-Privateprivacy_setting=Non-Private2026.01 | 91.61 | — | — | — | — | — | — | — | |
| WhitespaceModel=RNN2023.02 | 91.6 | — | — | — | — | — | — | — | |
| ALArchitecture=LSTM2026.01 | 91.53 | — | — | — | — | — | — | — | |
| Best T4width (d)=768, Encoder variant=lin, Encoder Parameter Saving=28.4 M → 7.1 M2026.03 | 91.52 | — | — | — | — | — | — | 0.05 | |
| TMixTrain size=10K2021.05 | 91.51 | — | — | — | — | — | — | — | |
| fastText2018.05 | 91.5 | — | — | — | — | — | — | — | |
| Large word CNNModel Type=Word-based2018.11 | 91.5 | — | — | — | — | — | — | — | |
| OriginalTrain size=10K2021.05 | 91.5 | — | — | — | — | — | — | — | |
| MixText†Train size=10K, unlabeled data=20K2021.05 | 91.5 | — | — | — | — | — | — | — | |
| WhitespaceModel=BERT2023.02 | 91.5 | — | — | — | — | — | — | — | |
| SentencePieceModel=BERT2023.02 | 91.5 | — | — | — | — | — | — | — | |
| Standard Transformerwidth (d)=768, Encoder variant=Std2026.03 | 91.47 | — | — | — | — | — | — | — | |
| Large word CNN2018.05 | 91.45 | — | — | — | — | — | — | — | |
| Char-CRNNModel Type=Character-based2018.11 | 91.4 | — | — | — | — | — | — | — | |
| Standard Transformerwidth (d)=256, Encoder variant=Std2026.03 | 91.4 | — | — | — | — | — | — | — | |
| mixup-transformerTrain size=10K2021.05 | 91.37 | — | — | — | — | — | — | — | |
| VDCNNModel Type=Character-based2018.11 | 91.3 | — | — | — | — | — | — | — | |
| SentencePieceModel=RNN2023.02 | 91.3 | — | — | — | — | — | — | — | |
| BPEModel=RNN2023.02 | 91.3 | — | — | — | — | — | — | — | |
| RETVec-rawModel=RNN2023.02 | 91.3 | — | — | — | — | — | — | — | |
| SGDlambda_a=0, lambda_s=12020.11 | 91.28 | — | — | — | — | — | — | — | |
| Deep CNNlayers=292018.05 | 91.27 | — | — | — | — | — | — | — | |
| VDCNNDepth=29, #Conv Params [M]=4.6, #FC Params [M]=12.59, #Total Params [M]=17.19, Storage Size [MB]=64.162019.01 | 91.27 | — | — | — | — | — | — | — | |
| BPArchitecture=Transformer2026.01 | 91.27 | — | — | — | — | — | — | — | |
| RETVecModel=CNN2023.02 | 91.2 | — | — | — | — | — | — | — | |
| ALArchitecture=Transformer2026.01 | 91.17 | — | — | — | — | — | — | — | |
| VDCNNDepth=17, #Conv Params [M]=4.4, #FC Params [M]=12.59, #Total Params [M]=16.99, Storage Size [MB]=62.742019.01 | 91.12 | — | — | — | — | — | — | — | |
| TMix†Train size=10K2021.05 | 91 | — | — | — | — | — | — | — | |
| TextCNN w/ NS-LFBackbone=TextCNN, NS-LF strategy=true2026.04 | 91 | — | — | — | — | — | — | — | |
| Transformer-2L (UHN)Model=Transformer-2L, #Params=612,117, Training Protocol=UHN2026.04 | 90.99 | — | — | — | — | — | — | — | |
| UHNModel=Transformer-2L, Evaluation Protocol=Single-task2026.04 | 90.99 | — | — | — | — | — | — | — | |
| VDCNNDepth=9, #Conv Params [M]=2.2, #FC Params [M]=12.59, #Total Params [M]=14.79, Storage Size [MB]=54.752019.01 | 90.83 | — | — | — | — | — | — | — | |
| Best T4width (d)=256, Encoder variant=lin, Encoder Parameter Saving=3.95 M → 1.00 M2026.03 | 90.76 | — | — | — | — | — | — | 0.64 | |
| BPEModel=CNN2023.02 | 90.7 | — | — | — | — | — | — | — | |
| UHNModel=Transformer-2L, Evaluation Protocol=Multi-task2026.04 | 90.62 | — | — | — | — | — | — | — | |
| SVDCNNDepth=29, #Conv Params [M]=1.56, #FC Params [M]=0.02, #Total Params [M]=1.58, Storage Size [MB]=6.032019.01 | 90.55 | — | — | — | — | — | — | — | |
| MIPT-hierParams=248K, Memory=168 MB, N=512, seeds=32026.04 | 90.5 | — | — | — | — | — | — | — | |
| BaselineMethod variant=Original2018.09 | 90.47 | — | — | — | — | — | — | — | |
| PAC-private responsesper-step MI budget (b)=infinity2026.01 | 90.44 | — | — | — | — | — | — | — | |
| SVDCNNDepth=17, #Conv Params [M]=1.43, #FC Params [M]=0.02, #Total Params [M]=1.45, Storage Size [MB]=5.522019.01 | 90.43 | — | — | — | — | — | — | — | |
| PAC-private responsesper-step MI budget (b)=2^-42026.01 | 90.35 | — | — | — | — | — | — | — | |
| TextCNNBackbone=TextCNN, NS-LF strategy=false2026.04 | 90.3 | — | — | — | — | — | — | — | |
| OoMMixTrain size=2.5K2021.05 | 90.25 | — | — | — | — | — | — | — | |
| LocalModel=FLAVA2025.09 | 90.25 | — | — | — | — | — | — | — | |
| SVDCNNDepth=9, #Conv Params [M]=0.71, #FC Params [M]=0.02, #Total Params [M]=0.73, Storage Size [MB]=2.82019.01 | 90.13 | — | — | — | — | — | — | — | |
| TMixTrain size=2.5K2021.05 | 90.07 | — | — | — | — | — | — | — | |
| OriginalTrain size=2.5K2021.05 | 89.92 | — | — | — | — | — | — | — |