TwinViT-DeepJSCC: Adversarially Robust Semantic Image Communication
TwinViT-DeepJSCC defends semantic image links, gaining up to 10.8 dB PSNR under adversarial channel attacks.
The paper proposes TwinViT-DeepJSCC, a semantic image transceiver with two Vision Transformer deep joint source-channel coding branches, sensitivity-aware masking, and receiver-side confidence fusion plus diffusion purification that needs no attack metadata. On CIFAR-100 it is evaluated against FGSM, PGD, natural evolution strategies, and Carlini–Wagner attacks, plus jamming and channel-aware waveforms over AWGN and Rayleigh fading. Under matched budgets it reports maximum PSNR gains of about 9.5 dB under 20-step PGD and 10.8 dB under channel-aware attacks, and up to about 38 percentage points higher Top-1 accuracy than an undefended baseline under PGD.
- Two ViT-DeepJSCC branches use sensitivity-aware masking
- Receiver fuses signals and purifies with SNR-conditioned diffusion
- About 9.5 dB PSNR gain under 20-step PGD on CIFAR-100
- Up to 10.8 dB PSNR gain under channel-aware waveform attacks
- Products
- TwinViT-DeepJSCCCIFAR-100
From the original article
1 min read at arxiv.orgLearning-based semantic communication is vulnerable to adversarial perturbations introduced before semantic encoding or over wireless channels. This paper proposes TwinViT-DeepJSCC, a preventive-corrective semantic image transceiver operating under a fixed channel-use budget.
By Maedeh Fallahreyhani · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.