The training source code of "Efficient Search of Comprehensively Robust Neural Architectures via Multi-fidelity Evaluation"
"train.py" is the code of retraining the searched architecture
- you can run train.py to retrain the model
After training the model, we evaluate the model robustness using the Composite adversarial attacks(CAA) code.
Our code is based on following two papers:
-
Mao X, Chen Y, Wang S, et al. Composite adversarial attacks[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2021, 35(10): 8884-8892.
-
Li L, Talwalkar A. Random search and reproducibility for neural architecture search[C]//Uncertainty in artificial intelligence. PMLR, 2020: 367-377.
The search code will be released when it is further organized.