Perceptual Science Series

PLEASE NOTE: Click here for respective titles and abstracts

IGERT Student Talks - Edinah Gnang and Paul Ringstad

Monday, April 20, 2009, 12:00pm - 07:00pm

Rutgers University

Copy to My Calendar (iCal) Download as iCal file


Edinah Gnang

Spectral theory and framework for higher order matrices with applications to vision

IGERT & Department of Computer Science 

Many problems in computer vision quite naturally reduced to processing n-dimensional arrays of numbers. Whether we are trying to perform a principal component analysis, Bayesian Inference, model problem through Markov models, analyze multidimensional statistics through the theory of  cummulants, use Kernels methods, or even study volumetric images through the theory of hypergraphs, one way or another we are led to n-dimensional arrays of numbers also known as tensors.

Our work has been to establish a coherent framework and tool set for extracting Knowledge from higher order matrices for problems in computer vision.  Tough multilinear algebra offers indeed tools and ways to study such objects, current methods in computer vision and some partial results suggest that spectral approaches to multilinear algebraic formulation of computer vision problems generalizes quite naturally to methods already being used in the field. From the perceptual science perspective these methods may offer discriminative tool and hypothesis that can be compared with human discriminative abilities in low level  image processing.




Paul Ringstad

"Reinforcement Learning for Adapting Performances in Conversational Agents"

Reinforcement learning is a branch of machine learning that is well suited for interaction with the real world. The goal of this research is to develop a framework for using reinforcement learning to endow conversational agents with the ability to adapt performances based on user feedback. The talk will survey recent work in this area and present a proposal for future work. 




IGERT Student Talks - Edinah Gnang and Paul Ringstad