[2021-Mar-03] Generating Text and Images with GPT

Speaker:

Mark Chen

Senior Research Scientist, OpenAI.

Topic:

Generating Text and Images with GPT

Date:

13:20-15:00 Wednesday 03-Mar-2021

Locate

Delta R105

Hosted by:

Prof. Po-Chih, Kuo

Abstract

The success of neural networks in machine learning has fueled the development of powerful new generative models. In particular, the Transformer has emerged as the state-of-the-art architecture for generating text and its representations. In this talk, I will introduce generative modeling and its applications, as well as recent developments in modeling text and images, with a focus on work done at OpenAI. Specifically, I will review the rise of generative pre-trained Transformers (GPT) in text, culminating in GPT-3. I will also describe work done to extend GPT to images with iGPT and end on recent work unifying these two modalities with DALL-E.

Bio.

Mark Chen is a senior research scientist at OpenAI working on unsupervised learning and generative modeling with Ilya Sutskever. His work on GPT-3 won Best Paper at NeurIPS 2020 and his work on Image GPT won Best Paper Honorable Mention at ICML 2020. He also serves as a coach for the USA International Olympiad in Informatics (IOI) team. Prior to OpenAI, Mark worked as a partner at a high frequency trading firm recently acquired by Citadel. He graduated with a degree in math with computer science from MIT.

All faculty and students are welcome to join.