Chapter 2: GAs in Problem Solving

Section 2.2: Data Analysis and Prediction

Predicting Dynamical Systems

An abstract version of the problem can be thought of as trying to predict x_n+1, given x₁, x₂, ... x_n where the subscripts often indicate time.

Meyer and Packard used the following GA to predict blood flow:

1. Randomly generate some condition sets.
2. Evaluate the fitness of each condition set.
3. Rank the population fitness.
4. Replace some fraction of the lower-fitness individuals with new condition sets obtained by applying uniform crossover and mutation.
5. Go to step 2.

A typical condition set might look something like this: [(2.0 ≤ x₁ ≤ 3.0) and (10.0 ≤ x₇ ≤ 15.0)].

• What might crossover look like?
• What might mutation look like? (Add a condition, delete a condition, broaden or shrink a range, shift a range up or down)

Experiment

• Figure 2.12 shows some high fitness condition sets
• Figure 2.13 shows trajectories of the data points that satisfied one of the previous condition sets.

Predicting Protein Structure

You are not responsible for this subsection.

Section 2.3: Evolving Neural Networks

GAs have been used to

• Evolve weights in a fixed network
• Evolve the network architecture, both through direct encoding and grammatical encoding
• Evolve a learning rule

You are responsible for understanding the details of this section. We do not have time to cover the details in class.