Dr. John Langford, Microsoft Research NYC

Talk Date and Time: September 8, 2022 at 03:00 pm - 03:45 pm EST followed by 10 minutes of Q&A on Zoom and IRB-5105

Topic: Latent State Learning

Abstract:

Hunting spiders have a strong spatial sense, providing them with the ability to plan and then execute a plan in an environment. This ability is not evident in reinforcement learning systems leading to the question: How can we learn a _latent_ state representation and transition structure from complex high dimensional representations? A good latent state structure ignores nearly all information, keeping only the positional/pose information of the agent and environment elements the agent can affect.

Bio:

John Langford (born January 2, 1975) is a computer scientist working in machine learning and learning theory, a field that he says "is shifting from an academic discipline to an industrial tool".^[1]

He is well known for work on the Isomap embedding algorithm, CAPTCHA challenges, Cover Trees for nearest neighbor search, Contextual Bandits (which he coined^[2]) for reinforcement learning applications,^[3] and learning reductions.^[4]

John is the author of the blog hunch.net and the principal developer of Vowpal Wabbit. He works at Microsoft Research New York, of which he was one of the founding members, and was previously affiliated with Yahoo! Research,^[1] Toyota Technological Institute at Chicago, and IBM's Watson Research Center. He studied Physics and Computer Science at the California Institute of Technology, earning a double bachelor's degree in 1997, and he received his Ph.D. in Computer Science from Carnegie Mellon University in the year of 2002.

John was the program co-chair for the 2012 International Conference on Machine Learning (ICML),^[5] general chair for the 2016 ICML,^[6] and is the President of ICML^[7] from 2019–2021.

His Wikipedia Page can be found here.