Competition on "Evolutionary Computation in Black-box MultiLabel Adversarial Examples" at the 2027 IEEE Congress on Evolutionary Computation (IEEE CEC 2027)
July 25 - July 28, 2027, Edinburgh, Scotland, UK
Overview
Artificial intelligence (AI), represented by deep learning, has deeply penetrated into multiple fields including production, life, and scientific research,
and has been one of the core technological engines that drives the progress of human society.
AI replaces repetitive labor through automation, and breaks through
the boundaries of human cognition.
However, much work has demonstrated that AI models are faced with some serious security and privacy threats, such as adversarial examples.
Adversarial examples—carefully crafted inputs that mislead AI systems—pose significant threats to the reliability and safety of these models.
By investigating the adversarial examples and developing efficient
attack testing methodologies, we can not only assess the security of models, but also enhance their intrinsic trustworthiness.
The generation of adversarial examples is a typical large-scale optimization problems, where evolutionary algorithms have shown remarkable effectiveness.
In this competition, we focus on a challenging category of adversarial examples—multi-label adversarial examples targeting deep neural networks.
Download
- Technical Report: Link
The document presents the contents about the competition on "Evolutionary Computation in MultiLabel Adversarial Examples"
- Source Code: Link
This project presents the source code of the competition on "Evolutionary Computation in MultiLabel Adversarial Examples"
Submission
- The data results and the brief analysis report (1-2 pages) should be provided and sent to diaoyiyacug@gmail.com.
- Participants may enter either or both tracks. For each selected track, report the best, median, worst, mean, and standard deviation of attack success rate (ASR) over 10 independent runs for each of the 12 model–dataset combinations (four models × three datasets). Each track will be ranked independently by overall mean ASR across the 12 combinations; ties will be resolved by lower variance of the overall ASR across runs. One winner will be selected for each track.
- The winner of each track must submit their source code after the competition, which should be publicly accessible.
Authors may upload it to GitHub or other platforms.
- This competition does not require papers, but we encourage participants to submit papers to IEEE CEC 2027.
Important Dates
- Deadline of submitting results: June 1, 2027
- IEEE CEC 2027 Conference: July 25 - July 28, 2027
Results
- The final competition report and data will be uploaded to GitHub. The CEC 2027 results link will be announced after the competition.
Organizers
- Yiya Diao, Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China, Email: diaoyiyacug@gmail.com
- Wenjian Luo, Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China, Email: luowenjian@hit.edu.cn
- Zhijian Chen, Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China, Email: 21B951010@stu.hit.edu.cn
- Yuhui Shi, Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen, China, Email: shiyh@sustech.edu.cn
© Machine Intelligence Laboratory --- HITSZ