Robust AI

Exploring vulnerability & Advancing robustness

To develop efficient AI, rigorous evaluation is essential. If the evaluation is incorrect, the efficiency of the AI model may be inaccurately assessed, which can impede its real-world deployment. If trained models are well equipped with visual intelligence, they should be robust against small deviations of input images from the trained domain that can occur naturally (e.g., successive image compression) or intentionally (e.g., adversarial attacks). However, when we focus exclusively on enhancing model efficiency, we may overlook critical aspects like robustness. Therefore, we have investigated the vulnerabilites of trained models and developed more robust models (Choi et al., 2019; Choi et al., 2020; Choi et al., 2022; Hwang et al., 2021; Kim et al., 2020; Kim et al., 2022). We have also explored fair evaluation methods for generated images (Choi et al., 2020; Lee et al., 2023).

References

2023

  1. AAAI
    Demystifying randomly initialized networks for evaluating generative models
    Junghyuk Lee, Jun-Hyuk Kim, and Jong-Seok Lee
    AAAI Conference on Artificial Intelligence, 2023
    arXiv HTML

2022

  1. Deep image destruction: Vulnerability of deep image-to-image models against adversarial attacks
    Jun-Ho Choi, Huan Zhang, Jun-Hyuk Kim, Cho-Jui Hsieh, and Jong-Seok Lee
    International Conference on Pattern Recognition, 2022
    arXiv HTML
  2. Successive learned image compression: Comprehensive analysis of instability
    Jun-Hyuk Kim, Soobeom Jang, Jun-Ho Choi, and Jong-Seok Lee
    Neurocomputing, 2022
    HTML

2021

  1. ICCV
    Just one moment: Structural vulnerability of deep action recognition against one frame attack
    Jaehui Hwang, Jun-Hyuk Kim, Jun-Ho Choi, and Jong-Seok Lee
    IEEE/CVF International Conference on Computer Vision, 2021
    arXiv HTML

2020

  1. Adversarially robust deep image super-resolution using entropy regularization
    Jun-Ho Choi, Huan Zhang, Jun-Hyuk Kim, Cho-Jui Hsieh, and Jong-Seok Lee
    Asian Conference on Computer Vision, 2020
    HTML
  2. MM
    Instability of successive deep image compression
    Jun-Hyuk Kim, Soobeom Jang, Jun-Ho Choi, and Jong-Seok Lee
    ACM International Conference on Multimedia, 2020
    HTML
  3. SRZoo: An integrated repository for super-resolution using deep learning
    Jun-Ho Choi, Jun-Hyuk Kim, and Jong-Seok Lee
    IEEE International Conference on Acoustics, Speech and Signal Processing, 2020
    arXiv HTML Code

2019

  1. ICCV
    Evaluating robustness of deep image super-resolution against adversarial attacks
    Jun-Ho Choi, Huan Zhang, Jun-Hyuk Kim, Cho-Jui Hsieh, and Jong-Seok Lee
    IEEE/CVF International Conference on Computer Vision, 2019
    arXiv HTML