About

I am currently a Research Fellow at Harvard University in the Computational Robotics Lab, advised by Prof. Heng Yang and Prof. Yilun Du. Prior to that, I completed my MS at ETH Zürich, advised by Prof. Thomas Hofmann in the Data Analytics Lab. I also spent some time on the RACER challenge at the NASA Jet Propulsion Laboratory, working on perception for autonomous navigation in extreme off-road terrains.

I am broadly interested in data-centric methods for generative modelling and representation learning, particularly through the lens of visual intelligence. My research is aimed at exploring the following axes:

1. How can intelligence emerge from data with little to no labels?
2. How can we leverage high-quality synthetic data in data-scarce domains?
3. How should an intelligent agent represent its environment across different modalities?

Please feel free to reach out for research, collaborations, a casual chat, or mentorship, especially if you are a junior, disadvantaged, or underrepresented student.

News

• [2024/09] Our team won the MIT x AI21Labs Hackathon on agentic LLMs
• [2024/04] I joined Harvard as a Research Fellow focusing on Diffusion Models
• [2023/10] My first-author paper will appear in NeurIPS 2023 Workshop on SyntheticData4ML
• [2023/09] I joined NASA JPL as a Visiting Researcher, working on RACER

Image-Editing Specialists: An RLAIF Approach for Diffusion Models

E. Benarous, Y. Du, H. Yang

Preprint, 2024

[Paper] [Code]

Harnessing Synthetic Datasets: The Role of Shape Bias in Deep Neural Network Generalization

E. Benarous, S. Anagnostidis, L. Biggio, T. Hofmann

Advances in Neural Information Processing Systems Workshop (NeurIPS), 2023

[Paper] [Code]

Enforcing Style Invariance in Patch Localization

E. Benarous*, D. Brunner*, J. Manz*, F. Yang*, T. Hofmann

Preprint, 2022

[Paper] [Code]