Discipline: Computer Sciences and Information Management
Subcategory: Nanoscience
William Brown III - Hampton University
This poster discussed an investigation on Evolvable Neural Network (ENN) that can be used in nano-robotics. To achieve this, both Artificial Neural Network (ANN) and Evolutionary Algorithm (EA) were explored. ANN is biologically mimicked algorithm to be used to learn from the past experience. EA is a population based algorithm to use the concepts of evolution to evolve the parameters to satisfy specific goal. Simulation results of neurons (15 vs. 10) on two different fitness functions (MEAN of elements in Synapses matrix vs. MEAN of neighboring elements of Synapses matrix) shows self-assembling and learning are helpful in reaching given goals. Artificial Neural Network The neurons ( x1, x2 ,…) take input from other neurons in their connection with the synaptic weights of w1, w2,…. The sum of all of the inputs passes through the activation function ( f ), to make neuron’s output. The output is usually in the range of 0 to 1. The Hopfield Neural Network uses a set of neurons with synaptic weight between every pair of neuron with a matrix of pair of neurons: [wi,j] Evolutionary Algorithm Principle of Biological Evolution Recombination of parents. Small changes from one generation to the next Adaptation with environment – Survival of fittest Mutation to increase survival chance. Evolvable Neural Network shows self-assembling and learning are helpful in reaching goals. This process is helpful in design process of nano-robots (self-assembled learner). Neural Network behaves as the central brain of the robots. Evolvable Neural Network can use the concept of evolution to find the best parameters in the neural network. Next step is to use a graphical simulator to evolve a robot to climb the stairs. Bullet Physics (free open source) – challenges on version incompatibility; lack of documentation; and unsolved bugs. MATLAB’s SIMULINK (paid software).
Funder Acknowledgement(s): Dr. Calvin Lowe, Dean of School of Science, Dr. Michelle Claville, NanoHU Program Director, Dr. Jean Muhammad, Chair, Computer Science Department, Dr. Chutima Boonthum, Computer Science Department, Dr. Moayed Daniel Daneshyari, Computer Science Department, and Mr. Roopchan Ramdon, Program Manager.
Faculty Advisor: Moayed Daniel Daneshyari,