Discovering Hidden Gems in Model Repositories
EMNLP 2026
Hey, I'm Eliahu Horwitz
Pronounced eh • lee • yah • hoo, but you can call me Eli (EE • lye)
I am an incoming postdoctoral fellow at Carnegie Mellon University (CMU), working with Prof. Jun-Yan Zhu, and a Carnegie Bosch Institute (CBI) Postdoctoral Fellow. I completed my PhD in Computer Science at the Hebrew University of Jerusalem, advised by Prof. Yedid Hoshen, where I was a Google PhD Fellow in Machine Learning and ML Foundations and researched Weight Space Learning—treating neural networks as data and developing models that operate on neural networks.
My research focuses on recovering neural network training trajectories and learning effective weight-space representations, ultimately enabling new ways to analyze, retrieve, and possibly generate models.
Before transitioning into research, I was a self-taught software developer, working across the tech stack at companies ranging from small startups to large corporations. This hands-on experience continues to shape my approach, bridging theoretical insights with real-world applications.
Recent News
- Starting a postdoc at Carnegie Mellon University this October, working with Prof. Jun-Yan Zhu.
- Honored to be awarded the Carnegie Bosch Institute (CBI) Postdoctoral Fellowship.
- Our paper Discovering Hidden Gems in Model Repositories was accepted to EMNLP 2026.
- Received my PhD in Computer Science from the Hebrew University of Jerusalem.
- Our paper Zero-Shot Model Search via Text-to-Logit Matching was accepted to TMLR.
- Gave a talk at the Israel Computer Vision Day.
- Presented my research at Colin Raffel's group meeting.
- Visiting EPFL to give a talk in Amir Zamir's lab.
- Visiting Berkeley to give a talk in Alexei Efros' lab.
- Visiting CMU to give a talk at the VASC Seminar.
- Gave a talk at IBM Research.
- Honored to be selected for the Google PhD Fellowship in Machine Learning and ML Foundations.
- Our position paper Charting and Navigating Hugging Face's Model Atlas was accepted to NeurIPS 2025 (Position Track; <8% acceptance rate).
- Visiting and giving a talk about my work at MIT CSAIL and David Bau's lab.
- Our paper Learning on Model Weights using Tree Experts was accepted to CVPR 2025.
- Our paper Distilling Datasets Into Less Than One Image was accepted to TMLR.
- Two papers accepted to ICLR 2025: Unsupervised Model Tree Heritage Recovery and Deep Linear Probe Generators for Weight Space Learning.
- Our workshop Neural Network Weights as a New Data Modality will take place at ICLR 2025.
- Honored to receive the Israel Council for Higher Education (VATAT) Scholarship for Outstanding PhD Students in AI and Data Science.
- Our paper Recovering the Pre-Fine-Tuning Weights of Generative Models was accepted to ICML 2024.
Selected Publications
We Should Chart an Atlas of All the World's Models
NeurIPS 2025 - Position Paper
Learning on Model Weights using Tree Experts
CVPR 2025
Unsupervised Model Tree Heritage Recovery
ICLR 2025
Recovering the Pre-Fine-Tuning Weights of Generative Models
ICML 2024