Classification on Synthetic (test)
98AccuracyDALS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DALSGeneration=SOTA2026.04 | 98 | — | |
| Fixed SGDGeneration=Gen 12026.04 | 97.9 | — | |
| SGDRGeneration=Gen 22026.04 | 97.8 | — | |
| DALS-FastGeneration=SOTA2026.04 | 97.8 | — | |
| DALS-AccGeneration=SOTA2026.04 | 97.8 | — | |
| LARSGeneration=Gen 42026.04 | 97.7 | — | |
| Cosine Decay SGDGeneration=Gen 22026.04 | 97.6 | — | |
| SAMGeneration=Gen 52026.04 | 97.5 | — | |
| SAM+DiscriminativeGeneration=SOTA2026.04 | 97.4 | — | |
| GrokfastGeneration=Gen 52026.04 | 97.3 | — | |
| Discriminative LRGeneration=Gen 42026.04 | 97.1 | — | |
| LionGeneration=Gen 52026.04 | 97 | — | |
| AdamGeneration=Gen 32026.04 | 96.8 | — | |
| RAdamGeneration=Gen 52026.04 | 96.5 | — | |
| AdamWGeneration=Gen 32026.04 | 96.2 | — | |
| Lookahead+AdamWGeneration=Gen 52026.04 | 96 | — | |
| STLR+DiscriminativeGeneration=Gen 52026.04 | 95.9 | — | |
| AdaBoundGeneration=Gen 32026.04 | 95.5 | — | |
| LDAType=Euclidean, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 86.1 | — | |
| Saddle-SVCnnz=0.52017.05 | 84.4 | 369 | |
| LinearSVCnnz=0.52017.05 | 84.3 | 214 | |
| LinearSVCnnz=0.12017.05 | 83 | 28.2 | |
| LinearSVCnnz=0.92017.05 | 82.8 | 537 | |
| Saddle-SVCnnz=0.92017.05 | 82.5 | 363 | |
| R-LDAType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 81.1 | — | |
| Saddle-SVCnnz=0.12017.05 | 80.4 | 393 | |
| FedEM2021.08 | 74.7 | — | |
| FedEMclient sampling=20%, communication rounds=12002021.08 | 74.7 | — | |
| D-FedEMtopology=Binomial Erdős-Rényi graph (p=0.5), communication=fully decentralized2021.08 | 73.8 | — | |
| MOCHAlambda=10^02021.08 | 73.4 | — | |
| pFedMe2021.08 | 69.2 | — | |
| Clustered FL2021.08 | 69.1 | — | |
| pFedMeclient sampling=20%, communication rounds=12002021.08 | 69.1 | — | |
| APFLclient sampling=20%, communication rounds=12002021.08 | 69.1 | — | |
| FedAvg+client sampling=20%, communication rounds=12002021.08 | 69 | — | |
| FedAvg+2021.08 | 68.9 | — | |
| FedAvg2021.08 | 68.2 | — | |
| FedAvgclient sampling=20%, communication rounds=12002021.08 | 68.2 | — | |
| Local2021.08 | 65.7 | — | |
| R-LEType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 63.3 | — | |
| IsomapType=Euclidean, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 61.1 | — | |
| R-RPCAType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 57.8 | — | |
| R-IsomapType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 55 | — | |
| R-PGAType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 54.4 | — | |
| PCAType=Euclidean, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 50.6 | — | |
| R-ONPPType=Riemannian, Number of components (nc)=3, Subsampling strategy=stratified, max 2000 samples2026.02 | 40 | — |