Federated Class-Incremental Learning on CIFAR-100 Heterogeneous (test)
1Communication Overhead FactorFedAvg
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FedAvgGroup=FL only, Budget=None2026.06 | 1 | 7.86 | 48.64 | 82.74 | |
| FedProxGroup=FL only, Budget=None2026.06 | 1 | 8.92 | 49.86 | 80.96 | |
| FedProtoGroup=FL only, Budget=Proto2026.06 | 1.08 | 18.74 | 52.38 | 66.82 | |
| PROGroup=Ours, Budget=Projected memory2026.06 | 1.12 | 43.28 | 63.72 | 31.54 | |
| FedSSIGroup=Heterogeneous FCIL, Budget=Memory2026.06 | 1.15 | 34.62 | 55.74 | 46.38 | |
| PRO-MAXGroup=Ours, Budget=Projected memory2026.06 | 1.16 | 46.12 | 65.06 | 26.18 | |
| FedSpaceGroup=Heterogeneous FCIL, Budget=Proto2026.06 | 1.25 | 33.84 | 55.18 | 47.92 | |
| TARGETGroup=Generator / replay FCIL, Budget=Synthetic2026.06 | 2.8 | 27.86 | 53.42 | 57.84 | |
| FedCBDRGroup=Heterogeneous FCIL, Budget=Replay2026.06 | 2.8 | 38.96 | 59.36 | 39.72 | |
| MFCLGroup=Generator / replay FCIL, Budget=Generative replay2026.06 | 3 | 29.42 | 54.76 | 54.63 | |
| HRGroup=Generator / replay FCIL, Budget=Hybrid replay2026.06 | 3.2 | 32.18 | 56.92 | 49.76 |