h07-4 - noisy versions of each of these images (randomly...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
CIS6930/4930 Intro to Computational Neuroscience Fall 2007 Home Work Assignment 4: Due Thursday 11/08/07 before class 1. Code and test a feed forward net of sigmoidal nodes with two input units, ten hidden units and one output unit that learns the concept of a circle in 2D space. The concept is: h x, y i is labeled “+” if ( x - a ) 2 + ( y - b 2 ) < r 2 and is labeled “-” otherwise. Set a = 2 , b = 3 , r = 1 . Generate 100 random samples uniformly distributed from a cell in R 2 to train the network using error backpropagation and 100 random samples to test it. Repeat the procedure multiple times and with multiple initial weights. Report the changing accuracy and the hyperplanes corresponding to the hidden nodes. 2. Get 20 images from the internet and reduce them to black and white dithered images of size 50 × 50 pixels. Now train a Hopfield type recurrent net to retrieve these images. Finally, initialize the net with
Background image of page 1
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: noisy versions of each of these images (randomly inverted pixels or cropped versions of the image) and see whether the net converges to the original images. Images can be displayed easily in pbm format. For example, copy the next few lines into a le called test.pbm and use your favourite viewer (irfanview, xv, eog) to view it. P1 # Width, height of image. Pixel values follow: &lt; 70 chars per line 10 20 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 1...
View Full Document

Ask a homework question - tutors are online