How to construct a good student
Juan P. Neirotti
Aston University, Birmingham/UK


We will investigate the problem of designing a suitable student for learning from a given teacher. Both student and teacher, are represented by feed-forward neural networks, thus the problem of designing the student presents itself at two levels, the hardware (the architecture) and the software (learning algorithm). We will discuss how the problem can be solved analytically and what are the advantages (and disadvantages) of such an approach. We will also discuss how this problem is solved in nature by studding a simulation based on an evolutionary technique (genetic programming). We will consider a population of students learning from the same teacher and we will assign a ranking order (or fitness) based on the students' learning ability. In this context the main issue will be the tradeoff between the adaptation the students must show to the current conditions and their ability to cope with a sudden change of environment.

back to the list of talks