Dennis Jacob

I'm a research specialist at Prof. David Wagner's group at UC Berkeley, where I work on building trustworthy large language models (LLMs).

During my undergraduate studies at Princeton University, I worked with Prof. Prateek Mittal's group, where I focused on protecting computer vision systems against adversarial attacks.

Email  |  CV  |  Scholar  |  LinkedIn  |  Github

RESEARCH  |  ENGINEERING  |  COMMUNITY   

Research

My research intersts include analyzing security vulnerabilities in deep learning systems (i.e., large language models, computer vision classifiers, etc). I am also interested in the impact of ML-driven recommendation algorithms on social networks.

2024

A certifiably robust framework for multi-label classifiers against adversarial patches, which is generalizable and can provably extend any existing certifiable defense for single-label classification.

Used a combination of agent-based modeling (ABM) techniques and analytical methods to understand the dynamics of platform selection; we find that polarization is possible despite individual preferences for diversity.

Directly extended the double-masking methodology from PatchCleanser to construct the first certifiable defense for multi-label classifiers against the adversarial patch threat model.
Distinguished Paper Award

2023

Proposed a novel “virtual worlds” concept to represent the structure of social media platforms in agent-based modeling (ABM) frameworks.

Engineering

I also enjoy coming up fun engineering solutions to problems I encounter in day-to-day life. Some of my projects and IPs are listed below.

A sole system cloud based application remotely connects to and adjusts the air chambers according to user preferences

An apparatus for a shoe sole with an air pump for adjusting pressure levels in air chambers distributed throughout the sole, and a battery

Vice President and founding member of the Hoagie Club at Princeton, a student developer group. Co-led development of HoagieStuff, the exchange platform (everything on campus, from bulletin to sales, in one place) for Princeton students.

Community

Througout my undergraduate studies at Princeton University, I TAed for various courses. One of my main contributions was developing a set of lecture notes for COS 324, the introductory machine learning course at Princeton.