Classification on MNLI Dir alpha=0.1
84.98Generalized AccuracyFEDYOGI
Evaluation results
| Method | Links | |
|---|---|---|
| FEDYOGIBackbone=RoBERTa Large, Optimization Approach=Backpropagation-based2024.05 | 84.98 | |
| FEDAVGBackbone=RoBERTa Large, Optimization Approach=Backpropagation-based2024.05 | 84.29 | |
| SPRYBackbone=RoBERTa Large, Optimization Approach=First-order Forward Mode AD2024.05 | 80.32 | |
| FWDLLM+Backbone=RoBERTa Large, Optimization Approach=Zero-order-based2024.05 | 72.66 | |
| FEDMEZOBackbone=RoBERTa Large, Optimization Approach=Zero-order-based2024.05 | 66.66 | |
| BAFFLE+Backbone=RoBERTa Large, Optimization Approach=Zero-order-based2024.05 | 62.85 |