Abstract
Deep neural networks deployed in safety-critical, resource-constrainedenvironments must balance efficiency and robustness. Existing methods treatcompression and certified robustness as separate goals, compromising eitherefficiency or safety. We propose CACTUS (Compression Aware Certified TrainingUsing network Sets), a general framework for unifying these objectives duringtraining. CACTUS models maintain high certified accuracy even when compressed.We apply CACTUS for both pruning and quantization and show that it effectivelytrains models which can be efficiently compressed while maintaining highaccuracy and certifiable robustness. CACTUS achieves state-of-the-art accuracyand certified performance for both pruning and quantization on a variety ofdatasets and input specifications.