The AI So-Go-Chi Center is pleased to announce that the following paper, co-authored by Assistant Professor Yongsong Huang, was accepted to and presented at the 19th European Conference on Computer Vision (ECCV 2026), held in Malmö, Sweden, from September 8 to 12, 2026. ECCV is one of the premier international conferences in computer vision and machine learning, bringing together researchers and industry professionals to share advances in the field.

Accepted Paper

Taïga Gonçalves, Yongsong Huang, Tomo Miyazaki, Shinichiro Omachi,

Improving Adversarial Robustness via Activation Amplification and Attenuation

Proceedings of the 19th European Conference on Computer Vision (ECCV 2026), Malmö, Sweden, September 8–12, 2026.

Paper: https://arxiv.org/abs/2606.27784

Code: https://github.com/tgoncalv/A3

The paper introduces Activation Amplification and Attenuation (A3), a lightweight plug-in module designed to improve the robustness of neural networks against adversarial perturbations. A3 uses a shared activation-scaling mechanism with two complementary modes: attenuation reduces the influence of adversarial perturbations, while amplification deliberately strengthens their effects to provide negative references during training. By jointly learning from both modes, the model develops stronger adversarial defenses. At inference time, only the attenuation mode is used.

Experiments across multiple network architectures, datasets, and training methods demonstrate consistent improvements in adversarial robustness with negligible additional computational and memory overhead. These results highlight A3 as a practical approach to strengthening the reliability of deep learning systems without substantial architectural changes.

On-site Presentation

The work was presented in an on-site poster session at ECCV 2026, sharing the proposed method and experimental findings with the international computer vision research community.

Congratulations to the authors on this achievement!