Ido Nitzan Hidekel: Biography

Ido Nitzan Hidekel is a Ph.D. student (direct track) in Electrical Engineering at Tel Aviv University, advised by Dan Raviv in the Geometric Deep Learning Lab. Ido's research studies the emergence of structural low rank and spectral bias in deep learning, and how they shape generalization, forgetting, and optimization. This work uses these insights to develop principled methods for applications spanning classification, language and vision models, implicit neural representations (INRs) and graphics, forgetting and model merging, and post-training.

In parallel, Ido is an AI Research Scientist Intern in the Multimodal AI group at IBM Research, led by Eli Schwartz, working on post-training for vision-language models and diffusion language models. Previously, Ido led the deep learning group at Dtect Vision, building production-grade detectors for AI-generated audio and images. Before moving to the direct Ph.D. track, Ido received a B.Sc. in Applied Mathematics and completed M.Sc. thesis work in Electrical Engineering (signal processing track), both at Tel Aviv University. Recent work includes SONAR, a spectral-contrastive approach to generalizable audio deepfake detection (ICML 2026), and a function-space theory showing that catastrophic forgetting is low-rank (ICML 2026 CATS Workshop).