Tal Amir’s Homepage
I am a research associate at the Faculty of Mathematics, Technion, working with Dr. Nadav Dym.
My main research interests are machine learning and optimization, with a focus on deep learning on structured data, such as sets, point clouds, and graphs. More info
I completed my Ph.D. in 2020 under the supervision of Prof. Boaz Nadler at the Department of Computer Science and Applied Mathematics in the Weizmann Institute of Science.
Research Highlights
Currently working on a low-distortion Euclidean embedding for 3D point clouds and molecular data.
Showed that oversquashing in graph neural networks is not limited to long-range tasks. [2025, LoG best paper award]
Developed the first bounded-distortion Euclidean embedding for multisets, and proved a fundamental impossibility result for such embeddings on distributions. [2025, ICLR]
Developed the first injective Euclidean embedding for multisets and measures based on neural functions. [2023, NeurIPS spotlight]
Developed a state-of-the-art method for sparse signal recovery. [2021, SIMODS]
News 💥
- Mar 2026 — The Trimmed Lasso / GSM sparse optimization solver is now available as a PyTorch package on PyPI:
sparse-approx-gsm. Thanks to Shachar Cohen (Weizmann) for this contribution! - Dec 2025 — Our paper on short-range oversquashing in GNNs received the best paper award at LoG 2025.
- Jan 2025 — Fourier Sliced-Wasserstein Embedding accepted to ICLR 2025.
- Dec 2023 — Neural Injective Functions for Multisets, Measures and Graphs accepted to NeurIPS 2023 as a spotlight.