Technion
academicThe Technion — Israel Institute of Technology (founded 1912) is Israel's oldest university and one of its leading centers for AI research, with a deep bench spanning low-precision training, generative vision, and NLP interpretability.
Flagship lines: Daniel Soudry's low-precision training work — Scaling FP8 Training to Trillion-Token LLMs (ICLR 2025 Spotlight) and FP4 All the Way (NeurIPS 2025 Spotlight), both joint with Intel but university-led; Tomer Michaeli's diffusion-editing line (Edit-Friendly DDPM inversion, FlowEdit, and TokenVerse — SIGGRAPH 2025 Best Paper, a Google DeepMind / Weizmann collaboration); and Yonatan Belinkov's interpretability line (CoT-faithfulness via unlearning — EMNLP 2025 Outstanding Paper; hallucination representations; arithmetic-as-heuristics circuits).
Industry ties are strong (Intel, Google DeepMind), and Joseph Keshet's MLSpeech lab leads speech forced-alignment. Shie Mannor also researches at NVIDIA, so his NVIDIA-branded output is attributed to NVIDIA — this entry keeps university-shipped work only.
Your notes
People
- Daniel Soudry Google Scholar — Associate Professor, EE (optimization / low-precision training)
- Yonatan Belinkov Google Scholar — Associate Professor, Taub CS (NLP interpretability)
- Tomer Michaeli Google Scholar — Professor, ECE (computer vision / diffusion)
- Shie Mannor Google Scholar — Professor, EE (reinforcement learning) (formerly Researcher, NVIDIA)
- Joseph Keshet Google Scholar — Professor, ECE; Director, Speech, Language & Deep Learning Lab (MLSpeech) (formerly Bar-Ilan University (until 2022))