Lab 3
MCS 394: Topics in Computer Science
Karl Knight, Spring 2000
This lab is not to be handed in.
The goal of this lab is to make you familiar with PDP++, the next
generation version of the neural-network simulation system originally
released with James L. McClelland's and David E. Rumelhart's books
Parallel Distributed Processing and Explorations in Parallel
Distributed Processing.
Specifically, you will work through part of the tutorial that is
available at The PDP++
Software Users Manual. (Please note that I only guarantee that
this link is accessible on the lab computers.)
In particular, you should:
- Look at the already-constructed simulator of a neural network that
solves the XOR problem using a single hidden layer. To do this, you
should go to chapter 4: Tutorial Introduction (using Bp), skim
through the background section (4.1: Backpropagation and XOR),
and work through the actual simulator (section 4.2: Using the
Simulator to run BP on the XOR Example).
By the way, to run this simulation, you should launch an xterm,
cd to the directory /usr/share/pdp/demo/bp,
and then type bp++.
- If you have time, you may continue with the tutorial and
construct a neural network that solves the 4-2-4 encoder problem.
This is described in section 4.3: Configuring the Encoder
Problem.
Recommendation: Do not record a script file as the
suggest in section 4.3.1: Recording a Script File of Your
Actions. You may want to do this with later projects, but it gets
in the way here.