Neural Networks Report: Digit Classification with MLP and Cross-Validation from MNIST, Study Guides, Projects, Research of Computer Science

Download Neural Networks Report: Digit Classification with MLP and Cross-Validation from MNIST and more Study Guides, Projects, Research Computer Science in PDF only on Docsity! Neural Networks, CAP 6615 Program 4, Due 18 March 2009 You shall use an MLP in crossvalidation to classify digits from the mnist data set. The mnist data set is available from leCun at http://yann.lecun.com/exdb/mnist/. I have also provided some resources in http://www.cise.ufl.edu/~jnw/CAP6615/Resources/, namely • The data files themselves: t10k-images-idx3-ubyte, and t10k-labels-idx1-ubyte. • A reader that converts the images and labels into vectors in the [X,d] format used in class: mnist_reader.m. Note that X is a 10,000 by 784 matrix. So, get X and d by doing [X,d] = mnist_reader(‘t10k-images-idx3-ubyte’,’t10k-labels-idx1-ubyte’); • A program to display any of the rows of the data vector X as an image: showNistImage.m. • A matrix of features (10,000 by 45) corresponding to the mnist data : featvecs.mat. This differs from your earlier project in several ways: • There are more than 2 classes, thus you clearly need more than one output node. • You need to convert the labels to a desired output matrix. You can consider your earlier MLP to have operated on a desired output matrix of size N by 1. It now needs to have more than one element per row. • You probably need lots more hidden units. You shall write a brief report (no more than 7 pages) with the following sections: 1. Overview (describing the contents of the report). 2. Methods (describing what your system does and how it is trained). 3. Experiments (telling all the experiments you performed to figure out how to get a solution for each of the data sets). 4. Experimental results (explaining how well your work solved the problems provided). 5. Conclusion (telling what you learned).