Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[theory-seminar] Hongyang's thesis defense on Monday June 3

Hongyang Zhang hongyang at cs.stanford.edu
Fri May 31 15:29:02 PDT 2019


Hi friends,

I'm going to do my thesis defense on *Monday, June 3, 2019 at 10:30am*
*at Gates 463A*. You are all welcome to come! More information of the talk
can be found below.

-----------------------------------------------------------------------------------------
University Oral Examination

*Title: Algorithms and Generalization for Large-Scale Matrices and Tensors*

Hongyang Zhang
Computer Science Department
Stanford University

Advised by Ashish Goel and Gregory Valiant

*Monday, June 3, 2019 at 10:30am* (refreshments served at 10:15am)
*Gates Building, Room 463A*

*Abstract:* Over the past decade, machine learning methods such as deep
neural networks have made a huge impact on a variety of complex tasks. On
the other hand, very little is understood about when and why these ML
methods work in practice. Bridging this gap requires better understanding
of the non-convex optimization paradigm commonly used in training deep
neural networks, as well as better modeling of real world data. My thesis
aims at providing principled algorithms and insights by examining
analytically tractable objects such as matrices and tensors, that are
intimately connected to neural networks.

This talk will show a few results:
i) we study gradient based optimization methods and their generalization
performance (or sample efficiency) in over-parameterized matrix models. Our
result highlights the role of the optimization algorithm in explaining
generalization when there are more trainable parameters than the size of
the dataset.
ii) we consider the problem of predicting the missing entries of
high-dimensional tensor data. We show an interesting representation-sample
trade-off in the choice of tensor models for fitting the data.
iii) we present new methods for the classic distance query problem that
creates state of the art data structures on a variety of large-scale graph
data.
-----------------------------------------------------------------------------------------

Best,
Hongyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.stanford.edu/pipermail/theory-seminar/attachments/20190531/ab70d2fc/attachment.html>


More information about the theory-seminar mailing list