Presentation
Fast Adversarial Training with Dynamic Batch-level Attack Control
DescriptionDespite the fact that adversarial training provides an effective protection against adversarial attacks, it suffers from a huge computational overhead.
To mitigate the overhead, we propose DBAC, a fast adversarial training with dynamic batch-level attack control.
Based on the prior study where attack strength should gradually grow throughout the training, we control the number of samples attacked per batch for higher throughput.
Additionally, we collect samples from multiple batches to form a pseudo-batch and attack them together for better GPU utilization.
We implement DBAC on PyTorch to show its superior throughput with similar robust accuracy compared to the prior art.
To mitigate the overhead, we propose DBAC, a fast adversarial training with dynamic batch-level attack control.
Based on the prior study where attack strength should gradually grow throughout the training, we control the number of samples attacked per batch for higher throughput.
Additionally, we collect samples from multiple batches to form a pseudo-batch and attack them together for better GPU utilization.
We implement DBAC on PyTorch to show its superior throughput with similar robust accuracy compared to the prior art.
Event Type
Research Manuscript
TimeWednesday, July 12th1:55pm - 2:10pm PDT
Location3003, 3rd Floor
Design
AI/ML System and Platform Design