Presentation
MALICE: Manipulation Attacks on Learned Image ComprEssion
DescriptionDeep learning has shown promising results in image compression, with competitive bitrate and image reconstruction quality from compressed latent. However, its robustness to adversarial images has never received deliberation. In this work, we, for the first time, investigate whether imperceptible perturbation of input images can stealthily precipitate a significant increase in the compressed bitrate without compromising reconstruction quality. We mount white-box and black-box attacks, which achieve up to 55.018x and 1.906x bpp change, respectively, revealing devastating fragility. To improve robustness, we propose a novel compression architecture factoratt that presents a promising trade-off between the rate-distortion performance and adversarial robustness.
Event Type
Work-in-Progress Poster
TimeWednesday, July 12th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security