Login
Talks & Seminars
Title: Research Directions in Theoretical Machine Learning
Prof. Sanjeev Arora, Princeton University and Institute for Advanced Study
Date & Time: January 8, 2018 17:05
Venue: Room # SIC 201, 02nd Floor, C Block, Department of Computer Science and Engineering, Kanwal Rekhi (KReSIT) Building
Abstract:
This talk will start with a 15-min introduction to machine learning. Then it will cover two interesting research directions in unsupervised learning where I think theoretical insight is needed. The first is the problem of capturing the “meaning” of text, where the goal is to represent the meaning geometrically as a vector. Two key notions here are word embeddings and text embeddings. I'll describe a theoretical model of text generation that is simple and intuitive, and can be shown to give new insight into methods for creating text embeddings, as well as the properties of the embeddings themselves. (Joint work with Li, Liang, Ma, Risteski; TACL 2016; and with Liang and Ma ICLR’17). The second research direction is training deep generative models for complicated data (such as real-life images). A hot idea in this area is Generative Adversarial Nets (GANs) and our results suggest that this framework is not as powerful as one had hoped, and furthermore even in practice the framework is quite far from learning the target distribution. (Joint work with Ge, Liang, Ma, Zhang ICML'17, and some unpublised work with Risteski and Zhang (arxiv)
Speaker Profile:
Details available at https://www.cs.princeton.edu/~arora/.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback