Construction of a Framework for the Realization of Neuro-Fuzzy Computing Structures Based on the Physical Behavior of Processing Elements

Please enable javascript in your browser.

Merrikh-Bayat, Farnood | 2012

678 Viewed

Type of Document: Ph.D. Dissertation
Language: Farsi
Document No: 43084 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Bagheri Shouraki, Saeed
Abstract:
In this thesis we concentrated on the application of memristor-rossbars for the hardware implementation of neuro-fuzzy systems. At first we showed that using memristive devices for this purpose has several problems and demonstrated the availability of the positive feedback in the behavior of this passive element. Then, by proposing new learning structure we tried to overcome some of these drawbacks. In the rest of this thesis we focused on the hardware implementation of neuro-fuzzy systems based on memristor crossbar structures. At first, hardware implementation of the Active Learning Method (ALM) is presented.
Next, new learning and computing method is proposed by creating and using fuzzy relations. Actually here we showed that it is possible to do computation without using exact mathematical formulas. The proposed method had this advantage that its hardware implementation was very simple. Finally, we demonstrated the similarities between digital logic, fuzzy inference systems and neural networkd and based on these similarities we proposed a new learning method based on the creation of fuzzy minterms
Keywords:
Artificial Neural Network ; Hardware Implementation ; Active Learning ; Neurofuzzy Computing System ; Memristor ; Memristive Systems ; Fuzzy Minterms

No TOC