The bifurcating neuron networks.

Format:

Book

Thesis/Dissertation

Author/Creator:

Lee, Geehyuk.

Contributor:

Farhat, Nabil H., advisor.

University of Pennsylvania.

Language:

English

Subjects (All):

Computer science.

Electrical engineering.

0544.

0984.

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

Local Subjects:

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

0544.

0984.

Physical Description:

298 pages

Contained In:

Dissertation Abstracts International 61-10B.

System Details:

Mode of access: World Wide Web.

text file

Summary:

Among many newly raised issues in neuroscience, we have been particularly interested in three issues, time coding, the role of coherent activities, and the role of chaotic activities. The Bifurcating Neuron (BN) is our model neuron designed with these three issues in mind: it is a chaotic model that can deal with time coding and has a built-in mechanism to incorporate the influence of coherent activity in its environment. The Bifurcating Neuron Network 1 (BNN-1) is a binary associative memory based on chaotic attractors. The BNN-1, utilizing the bistability of the BN controlled by attractor-merging crisis, was shown to have a better recall ability than the continuous-time Hopfield network. The BNN-1 is particularly suited for circuit realization because the only required circuit components are relaxation oscillators and harmonic oscillators. Another feature of the BNN-1 is that its chaotic activity is self-organizing: it starts in a maximally chaotic state and settles down to a less chaotic state as a recall process proceeds. The self-organizing behavior turned out to be useful when the BNN-1 was used to solve an optimization problem: the BNN-1 could reach a solution without any external control of a network parameter. The BN Network 2 (BNN-2) is another BN network that is designed to store analog patterns. It is based on the amplitude-to-phase transformation characteristics of the BN and the constructive interference, in the sense of wave optics, among neuronal spikes. A Hebbian learning scheme results in the formation of attractors with large basins of attraction. Also, the firing-time pattern of BNs induced by the same input pattern becomes different when the frequency of the relaxation level oscillation changes, and this led us to consider the possibility of volume-holographic memory. In a numerical simulation, we could configure the BNN-2 to maintain memories of two sets of patterns, one of which becomes accessible when the frequency of the relaxation level oscillation is tuned to that of the recording phase.

Notes:

Thesis (Ph.D. in Electrical Engineering) -- University of Pennsylvania, 2000.

Source: Dissertation Abstracts International, Volume: 61-10, Section: B, page: 5475.

Supervisor: Nabil H. Farhat.

Local Notes:

School code: 0175.

ISBN:

9780599970106

Access Restriction:

Restricted for use by site license.