Project:

Research into the use deep neural networks (DNNs) to model complex probability distributions.

Project:

This project, which is being co-supervised by myself and Dr. Brent Harrison (also in CS) is working on coalition formation games, and wants to try to learn strategies for agents in such games using deep learning.

Research Techniques:

We will use the technique of neural networks to attempt to model the probability distribution over the space of natural images. Efficient GPUs implementations of neural networks will be used. The goal is to randomly generate a wide range of realistic natural images.

Software to be used:

Python, Theano, Pylearn2, Caffe, GCC, Intel MKL Math Kernel Library, Matlab

Personnel:

Richard “Drew” Duncan (PhD student)
Kshitija Taywade, ITS

Publications:

2016 - 2019

Kshitija Taywade, Judy Goldsmith, and Brent Harrison. "Decentralized Multiagent Approach for Hedonic Games." In European Conference on Multi-Agent Systems, pp. 220-232. Springer, Cham, 2018.

2014

MPCA: EM-based PCA for mixed-size image datasets F Shi, M Zhai, D Duncan, N Jacobs Image Processing (ICIP), 2014 IEEE International Conference on, 1807-1811 2014
Covariance-Based PCA for Multi-size Data M Zhai, F Shi, D Duncan, N Jacobs Pattern Recognition (ICPR), 2014 22nd International Conference on, 1603-1608 2014
CP-nets with indifference. TE Allen Allerton, 1488-1495 1 2013 Making CP-Nets (More) Useful TE Allen Twenty-Eighth AAAI Conference on Artificial Intelligence 2014
Counting, Ranking, and Randomly Generating CP-nets TE Allen, J Goldsmith, N Mattei Workshops at the Twenty-Eighth AAAI Conference on Artificial Intelligence 2014
A Model for Intransitive Preferences S Saarinen, C Tovey, Jt Goldsmith Workshops at the Twenty-Eighth AAAI Conference on Artificial Intelligence 2014

2013

Learning CP-net preferences online from user queries JT Guerin, TE Allen, J Goldsmith Algorithmic Decision Theory, 208-220 2 2013

2014

MPCA: EM-based PCA for mixed-size image datasets F Shi, M Zhai, D Duncan, N Jacobs Image Processing (ICIP), 2014 IEEE International Conference on, 1807-1811 2014
Covariance-Based PCA for Multi-size Data M Zhai, F Shi, D Duncan, N Jacobs Pattern Recognition (ICPR), 2014 22nd International Conference on, 1603-1608 2014

Project:

Investigating/testing models of preferences and reasoning with those models.

Abstract: We are studying the properties of CP-nets, a combinatorial structure used to represent preferences over factored outcomes. Our research involves learning CP-nets from (possibly inconsistent) example data and also using CP-net models to reason about the relationship of arbitrary pairs of outcomes. In general these problems are known to be NP-hard, and certain related problems require exponential time. However, by exploring the problem space, we have found parameters that can be leveraged for heuristic search. We believe such heuristics will make CP-nets a practical alternative for modeling preferences in complex engineering applications, such as e-commerce and smart homes.

In addition, we will be exploring stochastic models of dynamic preferences, leveraging work in Bayesian networks and hidden Markov models to test our hypotheses about human preferences for variety, novelty, and surprise.

Software to be used:

Python, Pylearn2, Caffe, GCC, Matlab

Personnel:

Thomas E. Allen (PhD student)
Cory Siler (undergraduate student)
Samuel Saarinen (undergraduate student)

Publications:

2014

CP-nets with indifference. TE Allen Allerton, 1488-1495 1 2013 Making CP-Nets (More) Useful TE Allen Twenty-Eighth AAAI Conference on Artificial Intelligence 2014
Counting, Ranking, and Randomly Generating CP-nets TE Allen, J Goldsmith, N Mattei Workshops at the Twenty-Eighth AAAI Conference on Artificial Intelligence 2014
A Model for Intransitive Preferences S Saarinen, C Tovey, Jt Goldsmith Workshops at the Twenty-Eighth AAAI Conference on Artificial Intelligence 2014

2013

Learning CP-net preferences online from user queries JT Guerin, TE Allen, J Goldsmith Algorithmic Decision Theory, 208-220 2 2013

Browser not supported