College - Author 1

College of Engineering

Department - Author 1

Mechanical Engineering Department

College - Author 2

College of Science and Mathematics

Department - Author 2

Mathematics Department

Advisor

Dr. Brendon Anderson, College of Engineering, Mechanical Engineering

Funding Source

Cal Poly’s Mechanical Engineering Department

Date

10-2025

Abstract/Summary

Modern artificial intelligence (AI) systems exhibit highly sensitive and unsafe behavior when subjected to undetectable cyberattacks. For instance, human-imperceptible manipulations of the pixels in image data can cause traffic sign classifiers to mispredict stop signs as yield signs. In this project, we will design and analyze new methods to robustify machine learning (ML) models against these adversarial threats. Specifically, we will explore randomization techniques that "smooth out" the ML model's decision making process by intentionally corrupting input data with small amounts of noise. Optimizing this noise to enhance resilience against attacks while maintaining the system's accuracy poses a major open problem that we will pursue. The end goal of this project is not only to design and implement such a randomized smoothing method that increases the state-of-the-art robustness of ML systems, but also to develop safety certificates of the proposed method using mathematical analysis. This project is inherently cross-disciplinary, drawing on tools from mechatronics and autonomous systems, computer science and programming, and mathematics. As such, all students with backgrounds and interests in any of these areas are encouraged to apply.

Share

COinS
 

URL: https://digitalcommons.calpoly.edu/ceng_surp/171

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.