Learning Diverse-Structured Networks for Adversarial Robustness
This work aims to improve adversarial robustness for deep learning models by optimizing network architectures specifically for adversarial training, which is an incremental improvement for the field of robust AI.
This paper addresses the problem of finding optimal network architectures for adversarial training (AT), arguing that architectures optimal for standard training (ST) are not optimal for AT. They propose Diverse-Structured Networks (DS-Net) which significantly reduce the search space by using predefined atomic blocks and weighting them, demonstrating empirical advantages.
In adversarial training (AT), the main focus has been the objective and optimizer while the model has been less studied, so that the models being used are still those classic ones in standard training (ST). Classic network architectures (NAs) are generally worse than searched NAs in ST, which should be the same in AT. In this paper, we argue that NA and AT cannot be handled independently, since given a dataset, the optimal NA in ST would be no longer optimal in AT. That being said, AT is time-consuming itself; if we directly search NAs in AT over large search spaces, the computation will be practically infeasible. Thus, we propose a diverse-structured network (DS-Net), to significantly reduce the size of the search space: instead of low-level operations, we only consider predefined atomic blocks, where an atomic block is a time-tested building block like the residual block. There are only a few atomic blocks and thus we can weight all atomic blocks rather than find the best one in a searched block of DS-Net, which is an essential trade-off between exploring diverse structures and exploiting the best structures. Empirical results demonstrate the advantages of DS-Net, i.e., weighting the atomic blocks.