List your company

Backdoor Defense on Average of four datasets (test)

93.24AccuracyNo Defense (Before)
73.719278.787183.85588.9229May 8, 2023Nov 2, 2023Apr 28, 2024Oct 23, 2024Apr 19, 2025Oct 14, 2025Apr 10, 2026

Evaluation results

MethodLinks
No Defense (Before)Backbone=RoBERTa, Attack Type=BadSent2023.0593.2498.98
No Defense (Before)Backbone=RoBERTa, Attack Type=BadWord2023.0592.4498.92
Before (No Defense)Backbone=ROBERTa, Attack=BadWord2023.0592.4498.92
No Defense (Before)Backbone=RoBERTa, Attack Type=Average2023.0592.3498.95
NADBackbone=RoBERTa, Attack Type=BadWord2023.0592.2598.96
Before (No Defense)Backbone=ROBERTa, Attack=BadSent2023.0592.2498.98
NADBackbone=RoBERTa, Attack Type=Average2023.0592.1198.95
KDBackbone=RoBERTa, Attack Type=BadWord2023.0592.0497.92
NADBackbone=RoBERTa, Attack Type=BadSent2023.0591.9798.94
KDBackbone=RoBERTa, Attack Type=Average2023.0591.8898.43
KDBackbone=RoBERTa, Attack Type=BadSent2023.0591.7298.94
No Defense (Before)Backbone=BERT, Attack Type=BadSent2023.0591.6398.6
Before (No Defense)Backbone=BERT, Attack=BadSent2023.0591.6298.6
NADBackbone=BERT, Attack Type=BadWord2023.0591.5998.65
No Defense (Before)Backbone=BERT, Attack Type=Average2023.0591.4998.63
NADBackbone=BERT, Attack Type=Average2023.0591.4798.66
No Defense (Before)Backbone=BERT, Attack Type=BadWord2023.0591.4698.65
Before (No Defense)Backbone=BERT, Attack=BadWord2023.0591.3698.65
Fine-tuningBackbone=ROBERTa, Attack=BadSent2023.0591.3698.92
NADBackbone=BERT, Attack Type=BadSent2023.0591.3598.67
KDBackbone=BERT, Attack Type=BadWord2023.0591.2298.75
Fine-tuningBackbone=ROBERTa, Attack=BadWord2023.0591.1297.46
KDBackbone=BERT, Attack Type=Average2023.0591.198.72
KDBackbone=BERT, Attack Type=BadSent2023.0590.9898.69
Fine-tuningBackbone=BERT, Attack=BadWord2023.0590.6598.6
Fine-tuningBackbone=BERT, Attack=BadSent2023.0590.4198.66
RAPBackbone=RoBERTa, Attack Type=BadWord2023.0589.9521.34
RAPBackbone=RoBERTa, Attack Type=Average2023.0589.8259.56
RAPBackbone=RoBERTa, Attack Type=BadSent2023.0589.6997.78
RAPBackbone=BERT, Attack Type=BadSent2023.0589.279.98
RAPBackbone=BERT, Attack Type=Average2023.0589.1151.48
RAPBackbone=BERT, Attack Type=BadWord2023.0589.0222.98
Fine-purifyingBackdoor Attack=Anchoring2023.0588.114.67
Fine-purifyingBackdoor Attack=Syntactic2023.0587.5421.21
Fine-pruningBackbone=ROBERTa, Attack=BadWord2023.0587.591.17
ONIONBackbone=BERT, Attack Type=BadSent2023.0587.4282.51
ONIONBackbone=BERT, Attack Type=Average2023.0587.3947.65
ONIONBackbone=BERT, Attack Type=BadWord2023.0587.3512.78
Fine-mixingBackdoor Attack=Syntactic2023.0587.125.42
Fine-purifyingBackbone=RoBERTa, Attack Type=BadSent2023.0586.8519.2
Fine-purifyingBackbone=ROBERTa, Attack=BadSent2023.0586.8519.2
Fine-purifyingBackbone=RoBERTa, Attack Type=Average2023.0586.7518.4
ONIONBackbone=RoBERTa, Attack Type=BadSent2023.0586.7284.85
Fine-purifyingBackbone=RoBERTa, Attack Type=BadWord2023.0586.6417.59
Fine-purifyingBackbone=ROBERTa, Attack=BadWord2023.0586.6417.56
ONIONBackbone=RoBERTa, Attack Type=Average2023.0586.5848.67
ONIONBackbone=RoBERTa, Attack Type=BadWord2023.0586.4412.48
Fine-pruningBackbone=ROBERTa, Attack=BadSent2023.0586.4162.53
Fine-pruningBackbone=BERT, Attack=BadWord2023.0586.3990.48
Fine-mixingBackbone=ROBERTa, Attack=BadWord2023.0586.3918.12
Fine-pruningBackbone=BERT, Attack=BadSent2023.0586.3674.21
Fine-mixingBackdoor Attack=Anchoring2023.0586.2719.96
Fine-purifyingBackdoor Attack=EP2023.0586.1411.49
Fine-purifyingBackdoor Attack=BadWord2023.0586.1324.71
Fine-mixingBackbone=ROBERTa, Attack=BadSent2023.0586.1135.97
Fine-purifyingBackdoor Attack=Layerwise2023.0585.8113.55
Fine-purifyingBackbone=BERT, Attack Type=BadSent2023.0585.6425.78
Fine-purifyingBackbone=BERT, Attack=BadSent2023.0585.6425.78
Fine-purifyingBackbone=BERT, Attack Type=BadWord2023.0585.6231.83
Fine-purifyingBackbone=BERT, Attack=BadWord2023.0585.6231.82
Fine-purifyingBackbone=BERT, Attack Type=Average2023.0585.5328.8
Fine-mixingBackdoor Attack=BadWord2023.0585.5328.94
Fine-mixingBackdoor Attack=EP2023.0585.1417.67
Fine-mixingBackbone=BERT, Attack=BadSent2023.0585.0352.07
Fine-mixingBackbone=BERT, Attack=BadWord2023.0584.6639.75
Fine-mixingBackdoor Attack=Layerwise2023.0584.6221.11
Fine-purifyingBackdoor Attack=EWC2023.0583.4219.25
Fine-mixingBackdoor Attack=EWC2023.0582.2127.42
CI-triggerAttack=WaNet2026.0482.192.7
CI-triggerAttack=BadCLIP2026.0481.899
CI-triggerAttack=Blended2026.0481.799.7
CI-triggerAttack=BadCLIP_adaptive2026.0481.664.1
CI-triggerAttack=Average2026.0480.684.4
Fine-mixingBackdoor Attack=Surgery2023.0576.4432.75
CI-triggerAttack=SIBA2026.0475.966.3
Fine-purifyingBackdoor Attack=Surgery2023.0574.4726.96