[Colloquium] Kolchinski/MS Presentation/May 23, 2014

[Margaret Jaffey]

This is an announcement of Alex Kolchinski's MS Presentation.

----------------------------------------------------------
Department of Computer Science
The University of Chicago

Date:  Friday, May 23, 2014

Time:  3:30 pm

Place:  Ryerson 277

Bx/MS Candidate:  Alex Kolchinski

MS Paper Title:  Convergence Patterns of Artificial Neural Net Synapses and Support Vector Machine Weights with Respect to Training Data

Abstract:
In supervised learning, both support vector machines (SVMs) and artificial neural networks (ANNs) learn how to classify training points by adapting vectors of weights or synapses to reflect the underlying structure of the training data. While the intuition behind how SVMs draw hyperplanes separating disparate classes' clusters of points  is clear, there is no easy intuitive explanation for how an ANN's synapse weights converge to respond to the structure of a class of inputs. As ANNs do not typically accommodate easy analytic solutions, significant research has been done into the stochastic methods that are typically used to train ANNs, but the way that the synaptic weights in the neural net actually adapt to training data has been much less thoroughly investigated. In this paper, I present an investigation of how ANN synapse weights converge to reflect training data, and how their convergence compares to that of an SVM's hyperplane weights on the same data. Using both artificially generated vectors of Bernoulli-distributed random variables and handwritten digit data from the MNIST database, I present an analysis of how ANN synapses and SVM weights respond to training data, both when it is artificially generated to be easy-to-interpret and when it corresponds to real-world problems. As ANNs can be interpreted in some ways to model the function of biological neurons, this analysis of their convergence patterns may be useful to further the understanding how the brain adapts to reflect sensory data. In addition, a better understanding of the underlying functioning of ANNs is likely to be useful with a view to designing better learning algorithms in the future.

Alex's MS advisor:  Professor Yali Amit

A draft copy of Alex's MS paper will be available in Margaret's office, Ry 156, on Monday, May 19th.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret Jaffey
margaret at cs.uchicago.edu
Department of Computer Science
Student Affairs Administrator
Ryerson 156
773-702-6011
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20140516/579c00aa/attachment.htm