Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Hierarchical Associative Memory Based on Oscillatory Neural Network

Fang, Yan (2013) Hierarchical Associative Memory Based on Oscillatory Neural Network. Master's Thesis, University of Pittsburgh. (Unpublished)

PDF (1st edit)
Primary Text

Download (3MB) | Preview


In this thesis we explore algorithms and develop architectures based on emerging nano-device technologies for cognitive computing tasks such as recognition, classification, and vision. In particular we focus on pattern matching in high dimensional vector spaces to address the nearest neighbor search problem. Recent progress in nanotechnology provides us novel nano-devices with special nonlinear response characteristics that fit cognitive tasks better than general purpose computing. We build an associative memory (AM) by weakly coupling nano-oscillators as an oscillatory neural network and design a hierarchical tree structure to organize groups of AM units. For hierarchical recognition, we first examine an architecture where image patterns are partitioned into different receptive fields and processed by individual AM units in lower levels, and then abstracted using sparse coding techniques for recognition at higher levels. A second tree structure model is developed as a more scalable AM architecture for large data sets. In this model, patterns are classified by hierarchical k-means clustering and organized in hierarchical clusters. Then the recognition process is done by comparison between the input patterns and centroids identified in the clustering process. The tree is explored in a "depth-only" manner until the closest image pattern is output. We also extend this search technique to incorporate a branch-and-bound algorithm. The models and corresponding algorithms are tested on two standard face recognition data-sets. We show that the depth-only hierarchical model is very data-set dependent and performs with 97% or 67% recognition when compared to a single large associative memory, while the branch and bound search increases time by only a factor of two compared to the depth-only search.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Fang, Yanyaf13@pitt.eduYAF13
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairLevitan, Stevenlevitan@pitt.eduLEVITAN
Committee CoChairChiarulli, Donalddon@cs.pitt.eduDON
Committee MemberChen, Yiranyic52@pitt.eduYIC52
Committee MemberMohanram, Kartikkmram@pitt.eduKMRAM
Date: 27 June 2013
Date Type: Publication
Defense Date: 2 April 2013
Approval Date: 27 June 2013
Submission Date: 5 April 2013
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 88
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Electrical Engineering
Degree: MS - Master of Science
Thesis Type: Master's Thesis
Refereed: Yes
Uncontrolled Keywords: Associative Memory, Neural Network, Pattern Recognition, non-Boolean System
Date Deposited: 27 Jun 2013 14:47
Last Modified: 15 Nov 2016 14:11


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item