Tony Martinez's Group
Group Leader: Tony R. Martinez
E-mail:martinez@cs.byu.edu
- Michael S. Gashler
E-mail:gashlerm@yahoo.com
- Kristine Monteith (Kristine Perry)
E-mail:kristine@axon.cs.byu.edu
- Adam H. Peterson
E-mail:ahp@byu.edu
- Current research:
-
I'm working on a way to use distance measures between learning algorithms to visualize the space of interesting problems. Ideally, by knowing which learners or algorithms are "close" to each other, the construction of effective ensembles that combine different strengths can be easier, and trying a variety of solutions on a new problem can be more straightforward. Understanding these relationships can provide a foundation for further understanding of the set of interesting problems.
- Mike Smith
E-mail:msmith@axon.cs.byu.edu
Dan Ventura's Group
Group Leader: Dan Ventura
E-mail:ventura@cs.byu.edu
- Heather Chan
E-mail:psychochan@juno.com
- Aaron Dennis
E-mail:2adennis@gmail.com
- Kyle Dickerson
E-mail:kyle.dickerson@gmail.com
- Adam Drake
E-mail:adam_drake1@yahoo.com
- Current research:
-
I am currently studying the practical application of Fourier-based learning. The Fourier transform has been used extensively in the computational learning theory community;
however, Fourier-based approaches have received little attention as a means of solving real-world problems. My research in this area is primarily focused on two tasks:
improving methods for rapidly determining which of the exponential number of Fourier basis functions are most highly correlated with an arbitrary set of data and
determining the best ways to combine basis functions to create accurate classifiers. I have developed a branch-and-bound search algorithm that has been shown on a variety
of real-world problems to quickly find a large number of the most highly correlated basis functions while exploring only a small portion of the search space. In addition,
high accuracies have been obtained on these real-world problems by using the approximated Fourier representation as a classifier and by using the highly correlated basis
functions as features in other learning algorithms.
- David Norton
E-mail:ghotikun@hotmail.com
- Ilya Raykhel
E-mail:iraykhel@gmail.com
- Neil Toronto
E-mail:ntoronto@cs.byu.edu
- Current research:
-
My area of study is the application of machine learning in computer graphics.
There are many areas in computer graphics that machine learning is especially
well-suited to, but has not been applied. For example, the problem of 2D image
interpolation (for scaling, rotating, and other operations) is not often
viewed as a neural network regression problem. "Locally weighted learning"
techniques, however, can be very good at learning functions
that approximate the major geometries in a 2D image. I have gotten very good
results in image scaling by sampling the learned functions at subpixel
coordinates.
Machine learning may also be able to replace various heuristics commonly used
in computer graphics with more well-founded approaches.
- Rob Van Dam
E-mail:rvandam00@gmail.com
View Research Group Alumni.
|
