I'm a fifth year PhD
student playing with machine learning algorithms.
My new favorites are no-regret algorithms, and I'm dabbling in game
theory. I'm also interested in other flavors of online algorithms as well as
stochastic / online optimization and algorithms for bandit/etc settings.
I can't say I can leave other aspects of learning behind, though, such as
graphical models and semi-supervised learning.
Ultimately, it'd be nice to save the world with machine learning / game
theory (e.g. solving environmental problems, designing public policy,
eliminating inefficiency and lack of rationality wherever possible).
My advisor is Geoff Gordon.
Previously I worked with Tom Mitchell
on
reading the web and co-training-style semi-supervised learning algorithms.
I've been also working on a couple art projects that use machine learning.
Resume [pdf]
Research
- Geoff Gordon, Sue Ann Hong, Miroslav Dudík.
First-Order Mixed Integer Linear
Programming. In Proc. 25th Conf. on
Uncertainty in Artificial Intelligence (UAI), 2009.
- Read the Web. Our ultimate goal is to build an information extraction
system that learns to extract and accumulate facts from text, and uses
those facts to learn to extract better. The basis of the system builds
upon ideas from bootstrap information extraction and co-training to
classify entities and relations for predefined classes, starting with a
handful of examples each. Preliminary results and a starting system
description can be found in our position paper:
Justin Betteridge, Andrew Carlson, Sue Ann Hong, Estevam R. Hruschka Jr.,
Edith L. M. Law, Tom M. Mitchell, Sophie H. Wang.
Toward Never Ending Language Learning.
AAAI Spring Symposium on Learning by Reading and Learning to Read, 2009.
Some Interesting Class Projects
- An Experimental analysis of no-Phi-regret algorithms for online convex
games [pdf], Optimization
class project.
- Linear-Time Inverse Covariance Matrix Estimation in Gaussian Processes
[pdf],
Probabilistic Graphical Models class project.
- An Empirical Investigation of Active Learning for Bootstrap
Information Extraction [pdf],
Read the Web class project.
- Hot? Or Not: learning to evaluate facial attractiveness based on images
and ratings. We employed eigenfaces and fisherfaces with SVMs to classify
the attractiveness of rectified facial images given the average of many
ratings from a popular website. Unfortunately, the method did not yield
great results, most likely due to the weak feature representation.
Computer Vision class project, Caltech.
Courses
Teaching:
ART
Curator
top
