Evolving Neural Network Ensembles with Ensemble-Based Fitness FunctionsPublic Deposited
Downloadable Contentopen in viewer
This study aimed to enhance the capabilities of NeuroEvolution of Augmenting Topologies (NEAT) to evolve diverse ensembles of neural networks that can solve both classification and reinforcement learning tasks. A novel fitness function, Constituent Ensemble Evaluation, rewards networks which perform well in ensembles. The study showed that for simple reinforcement learning tasks, the team’s approach did not improve the capability of NEAT to evolve useful ensembles. However, for classification tasks, when CEE was incrementally introduced into the fitness function, ensembles performed marginally better on average (0.599 ± 0.09) than a similar published technique, Orthogonal Evolution of Teams (0.568 ± 0.02). However, CEE also has a slightly higher variance, making it less reliable.
- This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.
- UN Sustainable Development Goals
- Date created
- Resource type
- Rights statement
- Last modified
- In Collection:
Permanent link to this page: https://digital.wpi.edu/show/2801pk99c