Distributed Optimization under Byzantine and DP Adversaries
2UtilityByz-SGDM
Evaluation Results
| Method | Links | |
|---|---|---|
| Byz-SGDMDP=✗, BR=✓, Assumptions=L-smoothness Bounded variance, Additional Comments=nCVX, Type=E2026.03 | 2 | |
| Robust D-SHBDP=✗, BR=✓, Assumptions=L-smoothness Full-batch gradients, Additional Comments=nCVX CVX, Type=E2026.03 | 2 | |
| Byz-VR-MARINADP=✗, BR=✓, Assumptions=L-smoothness Local functions are finite-sums, Additional Comments=nCVX, Type=E2026.03 | 2 | |
| Safe-DSHBDP=✓, BR=✓, Assumptions=M-bounded gradients, Additional Comments=sCVX, local-DP, Type=E2026.03 | 2 | |
| DP-BREMDP=✓, BR=✓, Assumptions=L-smoothness Momentum buffers are i.i.d, Additional Comments=nCVX, central DP, Type=E2026.03 | 2 | |
| Byz-Clip21-SGD2MDP=✓, BR=✓, Assumptions=L-smoothness σ-sub-Gaussian noise, Additional Comments=nCVX, local-DP, Type=P2026.03 | 2 |