Skip to content Skip to navigation

The RAIN seminar is held on Wednesdays from 12:00-1:00pm in Y2E2 101 . And yes, lunch is provided!

RAIN schedule for Winter Quarter 2016-17

Date Speaker Topic
January 25
Devavrat Shah Blind Regression, Recommendation System and Collaborative Filtering
February 8
Udi Weinsberg Data-Science-Driven Products at Facebook
February 22
Anima Anandkumar Large-scale Machine Learning: Theory and Practice
March 8
Éva Tardos TBA

RAIN schedule for Spring Quarter 2016-17

Date Speaker Topic
April 12
Alireza Tahbaz-Salehi TBA
April 26
Aaron Roth TBA
May 10
Mohsen Bayati TBA
May 24
Herve Moulin TBA
May 31
John Dickerson TBA

Google Calendar for RAIN

Previous year's talks

Archived talks can be accessed here.

Talk Abstracts

Blind Regression, Recommendation System and Collaborative Filtering
Devavrat Shah, MIT

We discuss the framework of Blind Regression (also known as Latent Variable Model) motivated by the problem of Matrix Completion for recommendation systems: given a collection of users and movies, the goal is to predict the unknown rating of a user for a movie using the known observations, i.e. complete the partially observed matrix. We posit that each user and movie is associated with their latent feature, and the rating of a user for a movie equals the noisy version of latent function applied to the associated latent features. Therefore, completing the matrix boils down to predicting the latent function value for user-movie pairs for which ratings are unknown, just like the classical regression setting. However, unlike the setting of regression, features are not observed here -- hence "Blind" Regression. Such a model arises as a canonical characterization due to multi-dimensional exchangeability property a la Aldous and Hoover (early 1980s).

In this talk, using inspiration from the classical Taylor's expansion for differentiable functions, we shall propose a prediction algorithm that is consistent for all Lipschitz continuous functions. We provide finite sample analysis that suggests that even when observing a vanishing fraction of the matrix, the algorithm produces accurate predictions. We discuss relationship with spectral algorithm for matrix completion, and the collaborative filtering.

The talk is based on joint works with Christina Lee, Yihua Li and Dogyoon Song (MIT).

Bio: Devavrat Shah is a Professor with the department of Electrical Engineering and Computer Science at Massachusetts Institute of Technology. His current research interests are at the interface of Statistical Inference and Social Data Processing. His work has been recognized through prize paper awards in Machine Learning, Operations Research and Computer Science, as well as career prizes including 2010 Erlang prize from the INFORMS Applied Probability Society and 2008 ACM Sigmetrics Rising Star Award. He is a distinguished young alumni of his alma mater IIT Bombay.

Data-Science-Driven Products at Facebook
Udi Weinsberg, Facebook

This talk will give an overview of a range of data-driven products that the Core Data Science group helped building, mostly by applying machine learning and statistical methods on large-scale data. We'll talk about analysis of cascades and product adoption, identifying trends in real-time, fighting scams, understanding the true meanings behind emoji, and figuring out how people laugh online.

Bio: Udi Weinsberg leads the Algorithms group in Core Data Science in Facebook. The group helps product teams across Facebook to tackle difficult product problems and deliver new features by leveraging vast amounts of data together with a range of machine learning techniques. Before Facebook, Udi was a senior researcher at Technicolor, working on privacy in machine learning using cryptographic methods.

Large-scale Machine Learning: Theory and Practice
Anima Anandkumar, Amazon and Caltech

Large-scale machine learning requires blending computational thinking with statistical frameworks. Designing fast, efficient and distributed learning algorithms with statistical guarantees is an outstanding grand challenge. I will present perspectives from theory and practice. I will demonstrate how spectral optimization can reach the globally optimal solution for many learning problems despite being non-convex. This includes unsupervised learning of latent variable models, training neural networks and reinforcement learning of partially observable Markov decision processes. In practice, tensor methods yield enormous gains both in running times and learning accuracy over traditional methods such as variational inference. I will then talk about the recent advances in large-scale deep learning methods. Our lab at AWS is actively innovating on the MXNet package. It is a highly flexible and developer-friendly open-source deep learning framework designed for both efficiency and flexibility. It is based on the distributed parameter-server framework. I will demonstrate how to use preconfigured Deep Learning AMIs and CloudFormation Templates on AWS to help speed up deep learning research and development. I will conclude on outstanding challenges on how we can bridge the gaps between theory and practice, and how we can design and analyze large-scale learning algorithms.

Bio: Anima Anandkumar is currently a principal scientist at Amazon Web Services. She will be joining Caltech CMS department in summer 2017 as a Bren endowed chair. Her research interests are in the areas of large-scale machine learning, non-convex optimization and high-dimensional statistics. In particular, she has been spearheading the development and analysis of tensor algorithms. She is the recipient of several awards such as the Alfred. P. Sloan Fellowship, Microsoft Faculty Fellowship, Google research award, ARO and AFOSR Young Investigator Awards, NSF Career Award, Best Thesis Award from the ACM Sigmetrics society, IBM Fran Allen PhD fellowship, and several best paper awards. She has been featured in a number of forums such as the Quora ML session, Huffington post, Forbes, O’Reilly media, and so on. She received her B.Tech in Electrical Engineering from IIT Madras in 2004 and her PhD from Cornell University in 2009. She was a postdoctoral researcher at MIT from 2009 to 2010, an assistant professor at U.C. Irvine between 2010 and 2016, and a visiting researcher at Microsoft Research New England in 2012 and 2014.