Paul Grobstein, Bryn Mawr College Department of Biology
Clare Congdon, Bryn Mawr College Computer Science Program

Lab 6: Neural networks: complex, distributed systems

Brains are an obvious, perhaps the paradigmatic example, of complex systems. And they are, of course, the agents of the trying to make sense of complex systems which we call thinking, learning, and understanding. As such, brains are worthy of study both in their own right, and as exemplars of the kinds of capacities which one would like to instantiate in both computer programs and information processing machines themselves. The labs, both this week and next, are intended to provide insights into how brains work, and how aspects of such understandings can be used in other settings.

At the most basic level, brains consist of (unimaginably) large numbers of interconnected elements (neurons) each of which receives signals from other elements, processes them in rather simple (but generally non-linear) ways, and transmits them to other elements. In so doing, they achieve (so far as we know) all of the information processing capabilities with which we are familiar, from our own lives as well as the achievements of others. In this lab we will focus on two aspects of this remarkable reality. One is the reality that any well-defined process (roughly, any computable function) can be carried out by an appropriately structured neural network (in fact, by an appropriately structured neural network made of elements substantially similar than those that actually exist in the brain). The other is the reality that the rules responsible for the carrying out of the process are not explicitly stated in any element or well-defined location but are instead a "distributed" property to which many elements, connections, and locations contribute.

The distributed character of neural processing will be illustrated for you using a "lateral inhibition network", and you may want to review this and play with the available simulator during the week (Tricks of the Eye, Wisdom of the Brain. You will also be shown a software package which you can use to build small circuits of neurons yourself and test them to see what processes they carry out. You should use this package to create two networks for class demonstration and discussion next week. One network should do an "exclusive or" calculation, that is it should have an output element which comes on if one or the other of two input elements is on but not come on if both input elements are on (or off). For the second network, you should first define some calculation or process, and then build a network appropriate to it.

As in previous weeks, you should record in your general your experiences with the lab assignment, as well as your thoughts about the readings and the relation of both to the larger issues of better understanding information processing in complex systems.

Maintained by: