LIST OF FIGURES

2. Theory

Figure 2.1: Feed-forward, layered, network architecture

Figure 2.2: Node activation

Figure 2.3: Modular network architecture

Figure 2.4: The gating network

Figure 2.5: An example of the crossover operation

Figure 2.6: An example of sub-task identification

3. Design

Figure 3.1: The abstract Network class

Figure 3.2: The FeedForwardLayeredNetwork class

Figure 3.3: The ConnectedNode class

Figure 3.4: The Node class

Figure 3.5: Some implementations of Function and ConnectivityPattern

Figure 3.6: The ModularNetwork class

Figure 3.7: The classes that implement the training algorithms

Figure 3.8: The FFLNBreeder class

Figure 3.9: Some of the implementations of the classes used by FFLNBreeder

Figure 3.10: The ModularBreeder class

Figure 3.11: The ActivityMonitor class, and some implementations of it

Figure 3.12: The Graphical Interface’s FFLNBreeder system panel

Figure 3.13: The FeedForwardLayeredNetworkMonitor monitor methods panel

Figure 3.14: The FFLNBreeder train method panel

4. Testing

Figure 4.1: The suitable neural network structures

Figure 4.2: Learning rates of the networks (difference error)

Figure 4.3: Learning rates of the modular network (sum-squared error)

Figure 4.4: Changes is best fitness

Figure 4.5: Average sizes of networks

Figure 4.6: Changes in best fitness

Figure 4.7: Average sizes of networks

Figure 4.8: Modular breeder results

Figure 4.9: Modular breeder results

Figure 4.10: Modular breeder results