This title appears in the Scientific Report :
2021
Please use the identifier:
http://hdl.handle.net/2128/30612 in citations.
Evolving autonomous agents with simulated brains using L2L and Netlogo
Evolving autonomous agents with simulated brains using L2L and Netlogo
Artificial neural networks (ANNs) are popular machine learning techniques used to model autonomous agents. Spiking neural networks (SNNs) provide the ability to reproduce spatio-temporal dynamics by transmitting information through action potentials or spikes. Given their more biologically realistic...
Saved in:
Personal Name(s): | Yu, Jessica (Corresponding author) |
---|---|
Contributing Institute: |
Computational and Systems Neuroscience; INM-6 Jülich Supercomputing Center; JSC |
Imprint: |
Jülich
2021
|
Physical Description: |
76 p |
Dissertation Note: |
Bachelorarbeit, RWTH Aachen, 2021 |
Document Type: |
Bachelor Thesis |
Research Program: |
SimLab Neuroscience Interactive Computing E-Infrastructure for the Human Brain Project Human Brain Project Specific Grant Agreement 3 ohne Topic |
Link: |
OpenAccess |
Publikationsportal JuSER |
Artificial neural networks (ANNs) are popular machine learning techniques used to model autonomous agents. Spiking neural networks (SNNs) provide the ability to reproduce spatio-temporal dynamics by transmitting information through action potentials or spikes. Given their more biologically realistic characteristic, they are particularly attractive for modelling biological systems, including the analysis and understanding of biological self organisation. As with many neural models, the difficulty in achieving the desired performance is finding the appropriate parameters settings. A commonly used autonomous approach is given by genetic algorithms (GAs), which provide an evolution-based search technique inspired by natural adaptation processes. The performance of these meta-heuristic search techniques depends on the settings of its hyperparameters, which present a challenging task on their own.In this work, a multi-agent simulation model embedded in NetLogo is investigated. It simulates an artificial ant navigating through a virtual maze with many obstacles in search of food. Through this process the ant is controlled by an SNN, whose parameter optimisation is examined and optimised in this thesis using GAs of two different tools (L2L and BehaviorSearch). Afterwards, a deeper investigation on the optimized SNNs is covered to understand and explain the observed behavior in the simulation. |