You are being asked to perform pattern recognition with English alphabet symbols. The architecture of your neural network is presented on page 60, with n = 63. You are to solve problem 2.1 on page 100. This problem refers back to example 2.15 on page 73. Seven alphabetic characters are presented on 9 x 7 grids. The input to your neural network is a vector with 63 inputs. You are asked to use bipolar values for more robust recognition. The alphabetic characters are A, B, C, D, E, J, and K. You are asked to create additional samples of each letter where noise has been added, i.e., ink might be missing at a particular pixel where it should have been present, or ink may be present where it should not be. Read the details on representing these instances. Part a) asks you to vary threshold and noise level. Part b) should also be attempted. And as usual, include a discussion explaining your results. This material is from Fundamentals of Neural Networks by Laurene Fausett.
0 comments