BSc AI Thesis: Chaotic Dynamic in the HTR model using Reservoir Computing

Keywords
Loading...
Thumbnail Image
Issue Date
2022-06-20
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Reservoir computing is a promising field of research in recurrent neural networks. RC has multiple benefits compared to normal RNN’s such as biological plausibility, accuracy, and others. The reservoirs of the network are randomly generated and only a linear output layer is trained. However, there is not much information about how the reservoirs process information and how the parameters influence the results. This research intends to explain the effect of the spectral radius and how ratios between layers in the HTR model influence the performance on a dynamical system. This will be done by implementing the HTR model using reservoir computing on a hierarchical time series task (POS tagging). By experimenting with slower and faster dynamics throughout the layers, the hypothesis that the spectral radius needs to increase while retaining the ’effective spectral radius’ has shown to be relatively incorrect. Although respecting the ’effective spectral radius’ is important, the spectral radius is of little influence and does not have significant impact on the results, only when the ’effective spectral radius’ is violated, results become error prone.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen