Marina  Zhang

Marina Zhang

Scholar Title

MIT Quest for Intelligence | Undergraduate Research and Innovation Scholar

Research Title

Building Robust Neural Architectures to Defend against Adversarial Attacks

Cohort

2020–2021

Department

EECS

Research Areas
  • Artificial Intelligence & Machine Learning
Supervisor

Luca Daniel

Abstract

Despite the successes of deep learning, recent work in adversarial attacks have demonstrated the vulnerability of deep neural networks and the limited robustness guarantees of such systems. Neural networks often play a central role in critical applications including autonomous vehicles, healthcare, and fraud detection, where networks susceptible to adversarial attacks could make decisions that ultimately result in fatalities or discrimination. This project is focused on 1) designing novel neural network architectures that will offer better robustness guarantees and 2) to develop a methodology for finding robust architectures, with the goal of improving the security and trustworthiness of state-of-the-art models.

Quote

I am participating in SuperUROP because I wish to gain research experience and be able to apply my previous experience in CS & mathematics to a longer-term project. I’ve taken ML courses at MIT and interned on the security & anti-abuse research team at Google, so I’m extremely excited to expand on that knowledge and be a part of the lab.

Back to Scholars