Deep learning theory and practice: activation function design and numerical behavior under low-precision arithmetic; self-supervised representation learning for medical imaging; learned control and reinforcement learning for robotic systems.
Teaching Areas
Introduction to computer science
Data structures
Introduction to Data science
Machine learning
Deep learning
Reinforcement learning
the Workshop for Machine learning algorithms for multimodal applications
Algorithms 1
the Lab for human & humanoid robots interaction through AI
Experiences
Co-Founder and CTO at Aiode Ltd., Israel: Generative music: LLM-based creative agents, and diffusion models under style conditioning. Work on audio embeddings and interactive agents. Inventor on a pending PCT application, “System and Method for Generating Media”.
Senior Computer Vision Algorithm Researcher at Planet Watchers Ltd., Israel: Deep learning for satellite imagery. The recurring problem was that optical data disappears under cloud while SAR remains available but is far harder to interpret, which led to CNN-LSTM change detection and segmentation over seasonal sequences and to multi-sensor fusion with explainable components. Co-inventor on a patent for crop monitoring using Bayesian crop models over seasonal satellite imagery.
CTO at DermaDetect, Israel: Chief Technology Officer - DermaDetect, Israel2019-2020 Smartphone-image-based dermatological assessment deployed as a clinical service. Led the technology through to acquisition by Maccabi Healthcare Services, which put the system in front of a national HMO patient population rather than a research cohort - image quality, distribution shift and clinician trust as operational problems rather than benchmark ones.
R&D roles at SwissVortex (Zurich), Easy-Septal (New Jersey), 4Tech (Ireland) and Apexum (Israel); three acquisitions. The technical thread throughout was navigation and measurement inside a beating heart from poor images - image processing and computer vision on CT, fluoroscopy and ultrasound, under the verification and validation discipline of ISO 13485. This is where my interest in what medical images actually encode began, and it is the direct source of the echocardiography work above.
