Computer Science -- Applications of Machine Learning. This course is done online. This course is an advanced class,
targeted at graduate students with
experience in either machine learning or one of the focus application
areas or both.
You do not have to be an ICS student to take the course, but you do
have to be proficient in the following:
Installing and running software.
(Applied) Mathematics (at least calculus and linear algebra).
Reading technical papers.
Literature searches, and using the library.
English (reading and writing).
The focus application areas this
The course is organized into
projects. You will work either in small groups or alone with my
supervision. You will receive your grade for your final project report.
During the first week, you should either pic one of the projects, or
propose a project of your own. Please email me with your choice. The
course will be shaped by your choices. Possible Projects:
Robotics: Interactive Learning (IL). Play with the interactive
learning algorithm from the following publication: S. Still.
Information theoretic approach to interactive learning. EPL 85 (2009)
28005. You will receive existing IL software written by L. Miller
(graduate student in the Still Lab) and S. Still, and you will work in
close collaboration with both of us.
Robotics: Exploration. Can we think about some simple
principles such that, built into the robot, they enable the robot to
explore its environment effectively and efficiently? Can these
principles also lead to other self-organized behavior, such as movement
coordination, cooperation inrobot swarms, and even self-assembly? You
will read the literature and come up with some ideas of your own.
Quantitative Finance: Regularized Portfolio Optimization. Based
on the work in:
S.Still and I. Kondor: Regularizing Portfolio Optimization. New
Journal of Physics 12 (2010) 075034 (15pp). Special Issue on
Statistical Physics Modeling in Economics and Finance. You will receive
existing software written by S. Still and F. Caccioli and L. Bottou.
You will test this software on both synthetic data and actual market
data and compare the results to other commonly used methods. You will
work in close collaboration with me and with Dr. Kondor.
Geoinformatics: Cluster analysis of multivariate data. You will
use methods descried in: S. Still and W. Bialek: How many clusters? An
information theoretic perspective. Neural Computation, 16(12):2483-2506
(2004) and S. Still, W. Bialek and L. Bottou: Geometric Clustering
using the Information Bottleneck method. In Advances in Neural
Information Processing Systems 16. MIT Press (2004). You use a
geochemical database of different volcanic materials from Iceland,
provided by T. Thordarson, U Edinburgh, and
C. Hamilton, NASA. You will work in close collaboration with me and Dr.
Geoinformatics: Visual pattern recognition of geological
landforms in satellite imagery. You will use a training data set
provided by C. Hamilton, NASA. You will train and test classifiers for
image recognition on this problem. You will work in close collaboration
with me and with Dr. Hamilton.
For the robotics projects, we
will use a robotic simulator.
You can choose which one you want to use. Examples include: