Taming a Normalizing Flow Model

Join us to a talk by Shimon Malnick about "Taming a Normalizing Flow Model".

Talk will be held in Hebrew.

14:10 > Talk starts
15:00 > Estimated finish

ABSTRACT: Generative models are becoming ever more powerful,
being able to synthesize highly realistic images. We propose an algorithm for taming these models — changing the probability that the model will produce a specific image or image category. We focus on generative models that are powered by normalizing flows, which allows us to reason about the exact generation probability likelihood for a given image. Our method is general purpose, and we exemplify it using models that generate human faces, a subdomain with many interesting privacy and bias considerations. Our method can be used in the context of privacy, e.g., removing a specific person from the output of a model, and also in the context of de-biasing by forcing a model to output specific image categories according to a given target distribution. Our method uses a fast fine-tuning process without retraining the model from scratch, achieving the goal in less than1% of the time taken to initially train the generative model. We evaluate qualitatively and quantitatively, to examine the success of the taming process and output quality.

ABOUT SHIMON: Shimon Malnick is a Ph.D. candidate in Electrical Engineering at Tel Aviv University, under the supervision of Prof. Shai Avidan and Dr. Ohad Fried. His research focuses on privacy issues in computer vision and deep learning models. Before that, he completed his Master's in Computer Science under the same supervision, focusing on the control of pre trained normalizing flow models.
LinkedIn profile: https://www.linkedin.com/in/shimon-malnick-1b8404125/

HOW DID IT ALL START?
The Data Science Bond started as an internal community of data scientists at Microsoft R&D Center Israel. For few years the community members, all Microsoft employees, have met on a bi-weekly basis to learn new things together and to share internally cross teams knowledge. From time to time, external speakers are invited to give a talk. These talks are all professional talks, it can be an academic level seminar or a data scientist from another company telling how data science is done in his/hers organization.
One day we had a thought... “Hey, these talks are great! Why not inviting more people to join and learn together with us???” So when possible – we will be glad hosting you at our campus at Herzliya or virtually for a short and fruitful data science session.