Roy Assaf

Overview

Roy Assaf

Title

Research Staff Member - Computer Vision

Location

IBM Research Europe - Zurich Zurich, Switzerland

Bio

Dr. Roy Assaf joined IBM Research Zurich in March 2018 where he currently works with the AI Automation group. His research is focused on deep learning for Computer Vision, particularly in the context of Foundation Models.

He leads the Foundation Models for Computer Vision project which aims to i) Develop Self-Supervised learning methods for training Foundation Models and ii) Develop tuning methods (Fine tuning, prompt tuning and in-context learning) for obtaining a portfolio of technical domain Foundation Models. He is also leading the development of defect detection models on civil infrastructure using dense recognition such as Object Detection and Instance Segmentation.

His previous work at IBM includes:

  • Developing a deep learning pipeline for explainable anomaly detection of IBM storage devices. He developed a parallel pipeline which can process data from over 10k storage systems efficiently. An essential aspect for applying deep learning for large-scale systems. His research aimed at developing an explainability methods that extracts impactful KPIs and ultimately enables support engineers to achieve quick root cause analysis.
  • Leading the machine learning model development effort in WP3 of the EU Horizon 2020 ROMEO project https://www.romeoproject.eu/). He lead the effort where IBM and other industrial partners such as EDF, Siemens and Iberdrola coordinate to develop diagnostic and prognostic models for offshore wind turbines.

Roy holds a Ph.D. in Robotics. He has been awarded a Distinction Master of Science in Robotics, The European Marie Curie Research Fellowship, and received the Best Paper Award at the 2017 IEEE Prognostics and Health Management conference.

Publications

Projects

Visual Prompting

Using AI to build computer vision models within minutes, through quick intuitive prompts, and with just a few images

Inspecto – Large Vision Model Inspection Service

Increasing speed and quality of enterprise visual inspection, leveraging AI and domain-specific Large Vision Models

Top collaborators