GPT-3 for Vision

On Wednesday, October 7, 2020 at 2:00 p.m., Dr. Ehsan Kamalinejad will present “GPT-3 for Vision”.

Day & Time: Wednesday, October 7, 2020
2:00 p.m. – 3:00 p.m.

Speaker: Ehsan Kamalinejad, PhD
Co-Founder & CTO at Visual One
Associate Professor at Cal State East Bay University
Former Senior Machine Learning Scientist at Apple
San Francisco, USA

Organizer: IEEE Toronto Signal Processing Chapter

Location: Virtual – Click here for the Google Meets link.

Contact: Mehrnaz Shokrollahi

Abstract: Deep learning in computer vision (CV) has proved to be very effective in solving many problems in real world. However, while the raw number of researches done in standard CV problems (such as ImageNet
object classification/detection) has exploded, the measurable progress in these fields has slowed down. Additionally, there are many real-world problems in vision that are simply not compatible with the current approaches. This demands a new wave of problem statements in CV (and a new set of benchmarks). This talk focuses on one important set of such problem statements. We propose that many real-world problems in vision are “event recognition” problems. We introduce a concrete definition for the event recognition problem. We will see that this definition of event detection prohibits large sample sets. Hence, these events need to be recognize based on very few samples. We start by reviewing the current literature and we propose some promising directions for approaching this problem. At the end we show some demos from our recent effort on wrestling with this very challenging problem. Our solution can be best described by the “vision counterpart of GPT-3 few shot learner”.

Register: Please check back soon for the registration link.

Biography: Ehsan Kamalinejad (EK) is a senior machine learning engineer. He is currently working on Visual One which is a YCombinator backed startup he co-founded. Before that he was working for several years at
Apple and Amazon as a staff machine learning engineer. Ehsan holds a faculty position as an associate professor at Cal State East Bay University. He got his PhD from University of Toronto. He has more than 7 years of experience delivering machine learning products in computer vision and natural language processing. His current project, Visual One, is about bringing next level intelligence to surveillance cameras.