FedPara: Low-Rank Hadamard Product for Communication-Efficient Federated Learning
About
In this work, we propose a communication-efficient parameterization, FedPara, for federated learning (FL) to overcome the burdens on frequent model uploads and downloads. Our method re-parameterizes weight parameters of layers using low-rank weights followed by the Hadamard product. Compared to the conventional low-rank parameterization, our FedPara method is not restricted to low-rank constraints, and thereby it has a far larger capacity. This property enables to achieve comparable performance while requiring 3 to 10 times lower communication costs than the model with the original layers, which is not achievable by the traditional low-rank methods. The efficiency of our method can be further improved by combining with other efficient FL optimizers. In addition, we extend our method to a personalized FL application, pFedPara, which separates parameters into global and local ones. We show that pFedPara outperforms competing personalized FL methods with more than three times fewer parameters.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | GSM8K (test) | Accuracy19.79 | 751 | |
| Question Answering | SQuAD 2.0 | F185.01 | 190 | |
| Summarization | Xsum | ROUGE-218.08 | 108 | |
| Question Answering | SQuAD v1.1 | F188.02 | 79 | |
| Summarization | CNN Daily Mail | ROUGE-139.98 | 67 | |
| Natural Language Understanding | GLUE (test val) | MRPC Accuracy88.85 | 59 | |
| Mathematical Reasoning | MathQA (test) | Accuracy19.46 | 33 | |
| Question Answering | ARC (25-shot), MMLU (5-shot), HellaSwag (10-shot), TruthfulQA (0-shot), and WinoGrande (0-shot) (test) | ARC Accuracy49.06 | 32 | |
| Mathematical Reasoning | MetaMathQA (test) | Accuracy30.17 | 26 | |
| MRI Image Generation | ADNI (evaluation) | FID12.502 | 12 |