University of Twente Student Theses


Representation of Spatial Knowledge in an artificial system

Stienen, J. (2014) Representation of Spatial Knowledge in an artificial system.

[img] PDF
Abstract:In this bachelor thesis research was conducted as to how spatial knowledge should be represented in an artificial system. This representation was to be inspired on the way this works in the human body. Literature studies pointed out that when people learn, they make use of an interaction between the hippocampus and the neocortex, where several input signals will be labeled as ‘being connected’. They also pointed out that as these signals would appear together more often, they would get an even stronger connection. This paradigm is called reinforcement learning in the scientific world and is also the learning mechanism being used in this research. For data storage this study is using a reservoir of neurons which is constructed according to the principles of reservoir computing; the user specifies an input and an output layer, the system itself will generate a random reservoir in which it will store the relationships between different signals. This program is written in the python programming language using the PyBrain Library. To be able to verify the system had actually learned something simulations were conducted using three different learning algorithms based on reinforcement learning. For all three conditions a significant effect was found for the effect the number of interactions had on the learning capabilities of the system. This shows our system actually learned its way around the maze, which is also illustrated in the visual representations the program also produces.
Item Type:Essay (Bachelor)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:06 documentary information
Programme:Psychology BSc (56604)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page