Domain Generalization via Nuclear Norm Regularization
About
The ability to generalize to unseen domains is crucial for machine learning systems deployed in the real world, especially when we only have data from limited training domains. In this paper, we propose a simple and effective regularization method based on the nuclear norm of the learned features for domain generalization. Intuitively, the proposed regularizer mitigates the impacts of environmental features and encourages learning domain-invariant features. Theoretically, we provide insights into why nuclear norm regularization is more effective compared to ERM and alternative regularization methods. Empirically, we conduct extensive experiments on both synthetic and real datasets. We show nuclear norm regularization achieves strong performance compared to baselines in a wide range of domain generalization tasks. Moreover, our regularizer is broadly applicable with various methods such as ERM and SWAD with consistently improved performance, e.g., 1.7% and 0.9% test accuracy improvements respectively on the DomainBed benchmark.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Air Quality Regression | Air Quality R-212 → R-69, NO2 | Normalized MAE0.073 | 16 | |
| Domain Adaptation | EEG Signals S33 → S00 | RMSE (Normalized)0.18 | 8 | |
| Pollutant concentration estimation | Air Quality R-69 → R-212 (NO2) (test) | Normalized RMSE10.2 | 8 | |
| Air Quality Regression | Air Quality R-212 → R-69, O3 | Normalized MAE15.9 | 8 | |
| Air Quality Regression | Air Quality Average across Scenarios | Normalized MAE11.1 | 8 | |
| O3 estimation | Real-Time Deployment R-212 -> R-69 (O3) (final 20% of target domain data (streaming simulation)) | Normalized RMSE0.132 | 8 | |
| Pollutant concentration estimation | Air Quality R-212 → R-69 (O3) (test) | Normalized RMSE0.184 | 8 | |
| Pollutant concentration estimation | Air Quality R-212 → R-69 (NO2) (test) | Normalized RMSE0.146 | 8 | |
| Air Quality Regression | Air Quality R-69 → R-212, O3 | MAE (Normalized)9.9 | 8 | |
| Pollutant concentration estimation | Air Quality R-69 → R-212 (O3) (test) | Normalized RMSE0.124 | 8 |