CS 5073 Homework 6 — Neuroevolution
for Problem Solving
Due Thursday, March 25, 2019
1. Motivation
Just as with artificial neural networks or evolutionary computation alone, to use neuroevolution to research questions or address problems, we need to apply them to particular instances. To do that, we need to take the same actions with regard the evolutionary computation aspects of neuroevolution as we do for any other application of evolutionary computation (following De Jong, Evolutionary Computation: A Unified Approach, p 72):
- decide what an individual in the population represents,
- provide a means for computing the fitness of an individual,
- decide how children (new search points) are generated from parents (current search points),
- specify population sizes and dynamics,
- define termination criteria for stopping the evolutionary process, and
- return an answer
In addition, we need to make several decisions regarding what features of our artificial neural networks (ANNs) will be evolved, such as connection weights or various aspects of the structure of the network.
2. Goals
The goals of this assignment are:
- to give you experience with reading papers from the primary, peer-reviewed literature in neuroevolution, and
- to give you experience with dissecting such papers and understanding how evolutionary computation models are implemented for particular problems.
3. Assignment
- List and explain at least 10 distinct features of an ANN that could be evolved using neuroevolution.
Read the paper "The Evolution of Learning:An Experiment in Genetic Connectionism" by Chalmers (located in the files section of the course in Canvas), and answer the questions below.
- What does an individual in the population represent?
- Is this a fixed-length linear object, a fixed-length nonlinear object (such as a tree, graph, or set of production rules), a variable-length linear object, or a nonlinear variable-length
object?
- How is each individual encoded?
- Is this encoding genotypical or phenotypical?
- How is fitness calculated for each individual in the population?
- What reproductive operators are used?
- What parameters are used for each of the reproductive operators?
- What is the parent population size?
- What is the offspring population size?
- Is an overlapping or non-overlapping generation model used?
- What stopping criteria are used?
- What answers are returned?
4. What to Turn In
You will turn in to the appropriate dropbox in Canvas a machine readable electronic copy of your homework that completes the exercises above.