Title

Image Classification with Approximately Biologically Realistic Elements

Presenter Information

Sami Abdul-Wahid

Document Type

Oral Presentation

Location

SURC 137B

Start Date

21-5-2015

End Date

21-5-2015

Keywords

Machine Learning, Computational Intelligence, Artificial Spiking, Neural Network

Abstract

Image classification is a well studied problem, with applications such as face recognition and natural image classification. Here, image classification is done using a neural network of spiking neurons in a feedforward heirarchy that resembles certain structures of the visual cortex. Image encoding is done first through edge detection on the image, after which an HMAX model is used to bring about degrees of scale, position, rotation, and contrast-reverse invariance. Then, a single spiking neural network layer is trained to classify the encoded image using supervised learning. Results are shown for classification of single digit handwritten numbers.

Faculty Mentor(s)

Razvan Andonie

Department/Program

Computational Science

Additional Mentoring Department

Computer Science

This document is currently not available here.

Share

COinS
 
May 21st, 12:40 PM May 21st, 1:00 PM

Image Classification with Approximately Biologically Realistic Elements

SURC 137B

Image classification is a well studied problem, with applications such as face recognition and natural image classification. Here, image classification is done using a neural network of spiking neurons in a feedforward heirarchy that resembles certain structures of the visual cortex. Image encoding is done first through edge detection on the image, after which an HMAX model is used to bring about degrees of scale, position, rotation, and contrast-reverse invariance. Then, a single spiking neural network layer is trained to classify the encoded image using supervised learning. Results are shown for classification of single digit handwritten numbers.