Development of Stable, Robust and Secure Intelligent Systems for Autonomous Vehicles
This project will harness the collective strengths of machine learning, stable control, and generative AI to develop a robust and intelligent system that can adjust its neural network’s behaviour and characteristics in real-time to counter adversarial attacks in complex environments.
Lead PI: Shuzhi Sam GE, NUS
1. Mike Zheng SHOU, NUS
2. Lin ZHAO, NUS
3. Yong LIU, A*STAR
4. Liangli ZHEN, A*STAR
5. Huazhu FU, A*STAR
6. Rick Siow Mong GOH, A*STAR
7. Yew Soon ONG, A*STAR
8. Rong SU, NTU
1. Jiashi FENG (Tiktok Pte Ltd)
To overcome the vulnerabilities of artificial intelligence (AI) computer vision systems and protect against adversarial attacks in real-world scenarios, we will combine the strengths of machine learning, control theory and generative AI to develop a robust and intelligent defence system. The system will be able to adjust its neural network’s behaviour and characteristics in real-time to counter adversarial attacks in complex environments. Our framework is designed for safety-critical applications including autonomous vehicles (AVs), medical analysis, and stable cognitive in brain science.
Our project focuses on achieving three specific goals: (1) Intelligent Control-Enabled Adversarial Robustness, (2) Co-Evolutionary Multimodal Adversarial attacks, and (3) Efficient and Trustworthy Evaluation of Adversarial Robustness.
We envision the following outcomes and benefits from our project:
(1) Understanding vulnerabilities: By understanding the vulnerabilities of machine learning models for AVs, we can gain valuable insights for developing effective countermeasures.
(2) Improving AI robustness: Enhancing the robustness of machine learning systems makes them more reliable and secure. This will help ensure the integrity and safety of AI-driven applications, such as AVs.
(3) Enhancing trust in AI: By addressing vulnerabilities and improving robustness, we can foster greater trust in AI systems, thereby accelerating the widespread adoption of AI technologies and bringing their benefits to more industries and applications.
(4) Beyond autonomous vehicles: the developed solution can be adapted for other safety-critical AI applications, including unmanned vessels and medical diagnosis systems.