Learning distance-behavioural preferences using a single sensor in a spiking neural network. Ross, M., Berberian, N., Cyr, A., Thériault, F., & Chartier, S. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), volume 10613 LNCS, pages 110-118, 2017.
Learning distance-behavioural preferences using a single sensor in a spiking neural network [pdf]Paper  doi  abstract   bibtex   
Actions from autonomous agents demand adaptive rules rather than being hard coded. Contrary to using multiple pre-calibrated sensors, utilizing a single non-calibrated sensor in combination with neural elements could provide flexibility through learning, to effectively cope with changing environments. The objective of this study was to design an adaptive system with the potential capability of learning behavioural preferences in relation to distinct distances from a wall using only a single ultrasonic sensor. Using spike-timing dependent plasticity (STDP) as a learning mechanism in a spiking neural network (SNN), the agent displayed the correct behaviour and was successful in learning the desired behavioural preference at a medium distance. However, the agent treated far and close distances as ambiguous inputs from the sensory environment, despite the presentation of reinforcement cues during learning.

Downloads: 0