Matthew Joseph Kusner

I am pursing my doctoral degree in Machine Learning under the direction of Kilian Weinberger at Washington University in St. Louis. My work is in budgeted learning, submodular optimization, dataset compression, and Bayesian optimization

Resume

Site Last updated 07/26/14

Publications
[PDF] Zhixiang (Eddie) Xu, Matt J. Kusner, Kilian Q. Weinberger, Minmin Chen, Olivier Chapelle
Classifier Cascades and Trees for Minimizing Feature Evaluation Cost
Journal of Machine Learning Research (JMLR), 2014

[PDF]Matt J. Kusner, Wenlin Chen, Quan Zhou, Zhixiang (Eddie) Xu, Kilian Q. Weinberger, Yixin Chen
Feature-Cost Sensitive Learning with Submodular Trees of Classifiers
AAAI Conference on Artificial Intelligence (AAAI), 2014

[PDF] [Poster]Matt J. Kusner, Stephen Tyree, Kilian Q. Weinberger, Kunal Agrawal
Stochastic Neighbor Compression
International Conference on Machine Learning (ICML), 2014

[PDF]Jake Gardner, Matt J. Kusner, Zhixiang (Eddie) Xu, Kilian Q. Weinberger, John P. Cunningham
Bayesian Optimization with Inequality Constraints
International Conference on Machine Learning (ICML), 2014

[PDF] Zhixiang (Eddie) Xu, Matt J. Kusner, Gao Huang, Kilian Q. Weinberger
Anytime Feature Learning
International Conference on Machine Learning (ICML), 2013

[PDF] Zhixiang (Eddie) Xu, Matt J. Kusner, Kilian Q. Weinberger, Minmin Chen
Cost-Sensitive Tree of Classifiers
International Conference on Machine Learning (ICML), 2013

Time-Sensitive Datasets

This section is currently under construction...

Carnegie Mellon University
Emotional Phrase Generation
Poster
I worked with Professor Reid Simmons to develop a natural language generation system for
physical therapy instruction.
Autonomous Mowing with Range
Poster
Working with Professor Sanjiv Singh, I converted a system for controlling a robotic lawn
mower from USB to ethernet communication, allowing the robot to effectively follow paths
without oscillation.


University of Iowa
Genotype analysis with block heuristic
Presentation
With Professor Thomas Casavant I worked on a heuristic for finding homozygous regions
of interest in canine SNP genotypes.
Audiogram Interpolation
Project Link
Additionally, I worked on techniques for audiogram interpolation for the AudioGene project
with in Dr. Casavant's lab.


Washington University, St. Louis
Drosophila Sequence Finishing and Annotation
Program Link
I attended a teaching assistant workshop at Washington University to learn about
genome sequencing and gene annotation. Specifically, we honed our skills on the
drosophila erectus genome.


Macalester College
Pteridium Aquilinum chloroplast sequencing and annotation
I worked with researcher Paul Wolf at Utah State University and Professor Paul
Overvoorde at Macalester College to sequence and gene annotate the chloroplast
genome of Pteridium Aquilinum using various web and wet lab tools.


The Blossom Algorithm for Maximum Matching
In CS369, Discrete Applied Mathematics, I collaborated with professor Stan Wagon
to develop a Mathematica implementation of Edmonds's Blossom Algorithm for finding
a maximum matching in any unweighted, undirected graph. We developed a visualization
for the algorithm that is a Wolfram Demonstration and can be found at:
http://demonstrations.wolfram.com/TheBlossomAlgorithmForMaximumMatching/
Below is the Mathematica project report with code:
Report (Mathematica file, .nb)
Report (.pdf)

Here's a shot of the Wolfram Demonstration:



Physical Therapy Feedback
In CS484, Introduction to Artificial Intelligence, I developed an exercise ontology
for reasoning about physical therapy exercise feedback. The system takes as input an
ontology representation of a patient's exercise and outputs a feedback sentence. The
reasoning system hooks into the emotional natural language generation software that I
developed while I was at Carnegie Mellon's Robotics Institute. This project is done in
collaboration with Professor Susan Fox and Professor Reid Simmons. Below is the project
report:
Report

Here is a picture of the Exercise class within the physical therapy ontology:



Ant Colony Modeling of Web Searching
In CS494, Collective Intelligence, I used ant colony simulation to model how web users
forage for information using search engines. This project is done in collaboration with
Professor Shilad Sen. Below is the report, presentation, and source code for the project:
Report
Presentation
Code

Here is a picture of the ant colony simulation I wrote in Breve; a simulation environment:



The Lady in the Lake
In MATH432, Mathematical Modeling, I detailed and furthered Paul J. Nahin's solution to
the Lady-in-the-Lake problem proposed by Martin Gardner in Scientific American
(November and December 1965). Here is a link to the Mathematica project write-up:
Report (Mathematica file, .nb)
Report (.pdf)

Go-Kart
Over the summer my brothers and I constructed a go-kart essentially
from scratch. What I mean is that, we acquired parts meant for
go-karts such as a centrifugal clutch, a disc brake, a go-kart axle,
etc... But the assembly was custom. Here's a front shot of
the final product:












Kinect + Processing + Arduino
Currently, I'm working on a project in which particular gestures,
recognized by an Xbox Kinect, control an Arduino-based robot. The
main idea is to experiment with a simple, intuitive method for robot
control. With this in mind, my younger brothers and I are working on
making small robots which turn lightswitches on or off if a particular
gesture is performed.

Below is a point cloud, gotten from initial testing with Processing:












Environment classification
Fall semester of my sophomore year I was engaged in an independent
study with Professor Susan Fox of the Computer Science department
at Macalester College. For this, I implemented a multi-layered
feed-forward neural network to classify four different environments:
hallways, common areas, stairwells, and classrooms.
For greater detail and the python code, here is the project write-up:
Write-up

Radio-controlled lawn mower
The summer after my first year of undergraduate study I designed
and created a radio-controlled device to mow grass. The device is
equipped with an electric weed-wacker motor as well as two electric
drive motors and two casters.
While the first prototype drives effectively on smooth surfaces, the
drive motors do not have sufficient torque to move the device on
grass.

Below are pictures of the first prototype:











Contact Information
Office: 422A Jolley Hall
Address: Washington University. 1 Brookings Drive. St. Louis, MO 63130
Feel free to email me at: matt dot kusner at gmail dot com