Photo of Elad Feldman

Elad Feldman

PhD Student, Tel Aviv University
Advised by Dr. Ben Nassi AI security · LLM safety · adversarial ML

About

I am a PhD student at Tel Aviv University, supervised by Dr. Ben Nassi . My research focuses on the security and robustness of modern machine learning systems, with a particular emphasis on large language models and AI systems deployed in real-world environments.

I am interested in understanding failure modes of learning-based systems under adversarial, unexpected, or malicious inputs, and in developing practical, model-agnostic defenses.

Research

My research interests include:

  • Security and safety of large language models
  • Adversarial machine learning and robustness evaluation
  • LLM-based systems and agent security
  • Model-agnostic defenses and empirical benchmarks

Publications

  • PaniCar: Securing the Perception of Advanced Driving Assistance Systems Against Emergency Vehicle Lighting
    E. Feldman, et al.
    arXiv preprint, 2025.
    [paper] [media]
  • Hate Speech Targets Detection in Parler using BERT
    E. Feldman, et al.
    arXiv preprint, 2023.
    [paper]
  • Demo: Identifying Drones Based on Visual Tokens
    E. Feldman, et al.
    Network and Distributed System Security Symposium (NDSS), AutoSec Workshop, 2022.
    [paper]

Background

Prior to my PhD, I completed both my B.Sc. in Software Engineering and my M.Sc. in Information Systems Engineering at Ben-Gurion University of the Negev, as part of the Meitar outstanding students program.

My M.Sc. thesis studied the robustness of machine learning–based perception systems in computer vision, with a focus on autonomous and advanced driver-assistance systems (ADAS). In particular, I analyzed how visual perturbations and environmental artifacts affect object detection and perception reliability.

This work shaped my interest in evaluating and securing learning-based systems under realistic deployment conditions, which directly motivates my current PhD research.

Contact

Feel free to reach me via email: eladfld@gmail.com.