DESIGN OF A SELF- ORGANIZING LEARNING ARRAY SYSTEM Dr. Janusz Starzyk Tsun-Ho Liu Ohio University School of Electrical Engineering and Computer Science May th, 2003 IEEE International Symposium on Circuits and Systems
May th, 2003 School of Electrical Engineering and Computer Science2 Outline Introduction Self-Organizing Learning Array Structure Neuron Structure and Self-Organizing Principles Data Preprocessing Software Simulation Result Conclusion and Future Work
May th, 2003 School of Electrical Engineering and Computer Science3 Introduction Digital computers are good at: Fast arithmetic calculation Precise software execution Artificial Neural Networks are good at: Software free Robust classification and pattern recognition Recommendation of an action Massive parallelism
May th, 2003 School of Electrical Engineering and Computer Science4 Introduction (Cont’d) Research Objective: Less interconnection Self-organizing Local Learning Nonspecific classification
May th, 2003 School of Electrical Engineering and Computer Science5 Self-Organizing Learning Array Structure (Cont’d) Feed forward organization and structure
May th, 2003 School of Electrical Engineering and Computer Science6 Self-Organizing Learning Array Structure (Cont’d) Initial Wiring
May th, 2003 School of Electrical Engineering and Computer Science7 Neuron Structure and Self- Organizing Principles Neuron Input - System clock
May th, 2003 School of Electrical Engineering and Computer Science8 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron Input - Data input
May th, 2003 School of Electrical Engineering and Computer Science9 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron Input - Threshold control input (TCI)
May th, 2003 School of Electrical Engineering and Computer Science10 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron Input - Input information deficiency Indication of how much the input space (corresponding to this selected TCI) has been learned [0, 1] 1 is set initially at the first input layer 0 indicates this neuron has solved the problem 100%
May th, 2003 School of Electrical Engineering and Computer Science11 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron inside Transformation functions Linear and nonlinear Single input or multiple inputs Information index calculation
May th, 2003 School of Electrical Engineering and Computer Science12 Neuron Structure and Self- Organizing Principles (Cont’d)
May th, 2003 School of Electrical Engineering and Computer Science13 Neuron Structure and Self- Organizing Principles (Cont’d)
May th, 2003 School of Electrical Engineering and Computer Science14 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron output - System output
May th, 2003 School of Electrical Engineering and Computer Science15 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron output - Output Clock
May th, 2003 School of Electrical Engineering and Computer Science16 Neuron Structure and Self- Organizing Principles (Cont’d) Neuron output - Output information deficiency of TCO = Input information deficiency of TCOT = Input information deficiency * local information deficiency (pass threshold) of TCOTI = Input information deficiency * local information deficiency (does not pass threshold)
May th, 2003 School of Electrical Engineering and Computer Science17 Data Preprocessing Missing data recovery All features are independent Some features are dependent Ref: [Liu] & [Starzyk & Zhu] Symbolic values assignment Number of numerical feature = 1 Number of numerical features > 1
May th, 2003 School of Electrical Engineering and Computer Science18 Symbolic value – numerical feature =1 1) 2) 3) 4)
May th, 2003 School of Electrical Engineering and Computer Science19 Symbolic value – numerical feature =1 Symbolic value – numerical feature =1 X s = [ ] T
May th, 2003 School of Electrical Engineering and Computer Science20 Data Preprocessing (Cont’d) 1) 2) 3) 4) 5)
May th, 2003 School of Electrical Engineering and Computer Science21 Data Preprocessing (Cont’d) Symbolic value – numerical feature > 1 X s = [ ] T
May th, 2003 School of Electrical Engineering and Computer Science22 Software Simulation Result
May th, 2003 School of Electrical Engineering and Computer Science23 Software Simulation Result (Cont’d) FSS Naïve Bayes NBTree C4.5-auto IDTM (Decision table) HOODG / SOLAR C4.5 rules OC C Voted ID3 (0.6) CN Naïve-Bayes Voted ID3 (0.8) T R Nearest-neighbor (3) Nearest-neighbor (1) PeblsCrashed
May th, 2003 School of Electrical Engineering and Computer Science24 Conclusion and Future Work Conclusion Local learning Self-organizing Data preprocessing Future work VHDL simulation FPGA machine VLSI design
May th, 2003 School of Electrical Engineering and Computer Science25 Reference Information & Computer Science (ICS), University of California at Irvine (UCI). (1995, December), Machine Learning Repository, Available FTP: Hostname: ftp.ics.uci.edu Directory: /pub/machine-learning- databases/ Liu T. H. (2002), Thesis, Future Hardware Realization of Self- Organizing Learning Array and Its Software Simulation. School of Electrical Engineering and Computer Science, Ohio University. Starzyk A. J. and Zhu Z. (2002), Software Simulation of a Self- Organizing Learning Array. Int. Conf. on Artificial Intelligence and Soft Computing.