Class-incremental learning on CIFAR-100 10-split
97.87AccuracySoft-TF-L[10,11,12]
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Soft-TF-L[10,11,12]Backbone=ViT-B/16, #Tr.Params.=6.93M, Task ID=Gradient ID2024.11 | 97.87 | 0.21 | 13.87 | 130 | |
| Upper-Bound of Soft-TFBackbone=ViT-B/16, #Tr.Params.=6.93M2024.11 | 95.9 | — | — | — | |
| Soft-TF-L[3,4,5]Backbone=ViT-B/16, #Tr.Params.=6.93M, Task ID=Gradient ID2024.11 | 93.76 | 1.83 | 13.87 | 130 | |
| Soft-TF-L[10,11,12]Backbone=ViT-B/16, #Tr.Params.=6.93M, Task ID=Prompt ID2024.11 | 92.35 | 2.98 | 13.87 | 80 | |
| Soft-TF-L[3,4,5]Backbone=ViT-B/16, #Tr.Params.=6.93M, Task ID=Prompt ID2024.11 | 91.77 | 3.37 | 13.87 | 80 | |
| DualLoRA+variant=enhanced2024.11 | 90.94 | 3.2 | — | — | |
| ARCL-ViTVenue=This work2026.02 | 90.87 | 3.92 | — | — | |
| Upper-Bound of DualPromptBackbone=ViT-B/16, #Tr.Params.=0.03M2024.11 | 90.85 | — | — | — | |
| VPT-CPGVenue=AAAI’252026.02 | 90.63 | 3.98 | — | — | |
| APGVenue=ICCV’232026.02 | 89.35 | 6.01 | — | — | |
| LoRA-DRSVenue=CVPR’252026.02 | 89.14 | — | — | — | |
| DualLoRA2024.11 | 89.13 | 4.08 | — | — | |
| ConvPromptVenue=CVPR’242026.02 | 88.87 | 4.75 | — | — | |
| SD-LoRAVenue=ICLR’252026.02 | 88.01 | — | — | — | |
| EvoPromptVenue=AAAI’242026.02 | 87.97 | 2.6 | — | — | |
| FedProTIPData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 87.94 | 1.3 | — | — | |
| CPromptVenue=CVPR’242026.02 | 87.82 | 5.06 | — | — | |
| EASEVenue=CVPR’242026.02 | 87.76 | 5.94 | — | — | |
| BiLoRAVenue=CVPR’252026.02 | 87.46 | — | — | — | |
| CAPromptVenue=AAAI’252026.02 | 87.27 | 4.9 | — | — | |
| CPrompt-KACVenue=CVPR’252026.02 | 87.19 | — | — | — | |
| InfLoRAVenue=CVPR’242026.02 | 87.06 | 6.22 | — | — | |
| SEMAVenue=CVPR’252026.02 | 86.98 | — | — | — | |
| PGPVenue=ICLR’242026.02 | 86.92 | 5.35 | — | — | |
| + PGP*Backbone=ViT-B/16, #Tr.Params.=0.03M, Task ID=Prompt ID2024.11 | 86.92 | 5.35 | 12.21 | 76 | |
| OS-Prompt++Venue=ECCV’242026.02 | 86.68 | 1.18 | — | — | |
| DualPromptVenue=ECCV’222026.02 | 86.51 | 5.16 | — | — | |
| DualPrompt*Backbone=ViT-B/16, #Tr.Params.=0.03M, Task ID=Prompt ID2024.11 | 86.5 | 5.77 | 12.12 | 76 | |
| ESNVenue=AAAI’232026.02 | 86.34 | 4.76 | — | — | |
| CODA-PromptVenue=CVPR’232026.02 | 86.25 | 1.67 | — | — | |
| FedProTIPData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 86 | 0.83 | — | — | |
| OVOR-DeepVenue=ICLR’242026.02 | 85.99 | 6.42 | — | — | |
| DualPrompt2024.11 | 85.85 | 5.41 | — | — | |
| CodaPrompt2024.11 | 85.77 | 4.07 | — | — | |
| InfLoRA2024.11 | 85.62 | 4.34 | — | — | |
| iGSP2026.05 | 85.34 | — | — | — | |
| PGP2024.11 | 85.28 | 5.6 | — | — | |
| MoE-Adapter(++)2026.05 | 85.21 | — | — | — | |
| L2P2024.11 | 83.97 | 6.41 | — | — | |
| DER++Backbone=ViT-B/16, #Tr.Params.=5,000, Task ID=-2024.11 | 83.94 | — | 14.55 | — | |
| L2PVenue=CVPR’222026.02 | 83.83 | 7.63 | — | — | |
| ZSCL2026.05 | 82.15 | — | — | — | |
| FedProTIPData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 81.94 | 1.35 | — | — | |
| BiCBackbone=ViT-B/16, #Tr.Params.=5,000, Task ID=-2024.11 | 81.42 | — | 17.31 | — | |
| iCaRL2026.05 | 79.35 | — | — | — | |
| LwF-VR2026.05 | 78.81 | — | — | — | |
| DNE2026.05 | 74.86 | — | — | — | |
| DER2026.05 | 74.64 | — | — | — | |
| CLIP zero-shotmode=zero-shot2026.05 | 74.47 | — | — | — | |
| DyTox2026.05 | 74.1 | — | — | — | |
| LoRA2024.11 | 73.32 | 20.4 | — | — | |
| BiC2026.05 | 68.8 | — | — | — | |
| DER+MCGBackbone=ViT-B/16, #Tr.Params.=2,000, Task ID=-2024.11 | 67.62 | — | 14.64 | — | |
| S-Prompt2024.11 | 67.03 | 24.8 | — | — | |
| BiCBackbone=ViT-B/16, #Tr.Params.=1,000, Task ID=-2024.11 | 66.11 | — | 35.24 | — | |
| iCaRLBackbone=ViT-B/16, #Tr.Params.=5,000, Task ID=-2024.11 | 66 | — | 5.33 | — | |
| LwF2026.05 | 65.86 | — | — | — | |
| Fine-tune2026.05 | 65.46 | — | — | — | |
| iCaRLBackbone=ViT-B/16, #Tr.Params.=1,000, Task ID=-2024.11 | 61.25 | — | 14.19 | — | |
| DER++Backbone=ViT-B/16, #Tr.Params.=1,000, Task ID=-2024.11 | 61.06 | — | 39.87 | — | |
| LWFBackbone=ViT-B/16, #Tr.Params.=-, Task ID=-2024.11 | 60.69 | — | 27.77 | — | |
| UCIR2026.05 | 58.66 | — | — | — | |
| PODNet2026.05 | 58.03 | — | — | — | |
| Seq-FTVenue=Baseline2026.02 | 52.85 | 49.24 | — | — | |
| FedProTIP (-t)Data Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 52.3 | 15.66 | — | — | |
| FedProTIP (-t)Data Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 48.41 | 15.59 | — | — | |
| EWCBackbone=ViT-B/16, #Tr.Params.=-, Task ID=-2024.11 | 47.01 | — | 33.27 | — | |
| FOTData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 46.86 | 21.11 | — | — | |
| FedProTIP (-t)Data Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 42.19 | 14.91 | — | — | |
| FOTData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 41.8 | 20.86 | — | — | |
| LANDERData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 39.09 | 9.27 | — | — | |
| LANDERData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 37.59 | 10.21 | — | — | |
| FOTData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 34.65 | 18.09 | — | — | |
| FTBackbone=ViT-B/16, #Tr.Params.=-, Task ID=-2024.11 | 33.61 | — | 86.87 | — | |
| TARGETData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 29.56 | 42.73 | — | — | |
| TARGETData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 27.37 | 37.6 | — | — | |
| LANDERData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 23.56 | 13.28 | — | — | |
| TARGETData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 23.05 | 34.63 | — | — | |
| FedAvgData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 18.92 | 63.2 | — | — | |
| FedAvgData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 15.76 | 52.8 | — | — | |
| FedAvgData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 15.35 | 62.9 | — | — | |
| LGAData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 14.93 | 72.06 | — | — | |
| LGAData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 14.35 | 71.09 | — | — | |
| GLFCData Heterogeneity (Dirichlet α)=IID, Inference Protocol=Task-agnostic inference2025.09 | 14.07 | 69.17 | — | — | |
| GLFCData Heterogeneity (Dirichlet α)=α = 0.5, Inference Protocol=Task-agnostic inference2025.09 | 11.86 | 68.2 | — | — | |
| LGAData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 11.67 | 65.82 | — | — | |
| GLFCData Heterogeneity (Dirichlet α)=α = 0.2, Inference Protocol=Task-agnostic inference2025.09 | 10.33 | 63.98 | — | — |