During the Amsterdam AI Highlights Event, Corinna Triebold won the Amsterdam AI thesis award.
Thesis title: Enhancing Generalisability of Evolved Neural Network Controllers
Corinna was supervised by Anil Yaman (Computational Intelligence research group)
Corinna is one of the two bachelor students receiving the award.
Abstract: While artificial neural networks (ANN) have demonstrated effective control on a range of tasks, the study of their robustness and generalizability remained limited. This has a crucial impact in control tasks, especially dealing with morphological variations. Morphological variations concern the physical structure of an embodied agent/robot. This bachelor thesis aims to enhance the generalizability of ANN based controllers by introducing morphological variations during the training process. As a result, the controllers are able to handle a wide range of morphological variations with sufficiently well performance. Tolerance to morphological variations can be crucial in maintaining reliable performance and also help reducing the reality gap that (the transferability of learned behavior in simulation to real life).
Read more about the event.