Yusoff, Nooraini and Yusof, Yuhanis and Siraj, Fadzilah and Ahmad, Farzana Kabir (2017) Fish Motion Trajectories Detection Algorithm Based on Spiking Neural Network (S/O: 12893). Technical Report. UUM. (Submitted)

PDF - Submitted Version Restricted to Registered users only Download (16MB) | Request a copy

Abstract

This study proposes sequence learning for the fish motion trajectory by using the Spike-Time dependent Plasticity (STDP) in the Spiking Neural Networks (SNNs). In the most predictive application, Backpropagation (BP) has been used to learn the behavioural motion pattern. In terms of plausibility, BP has several drawbacks due to the lack in functionalities in complex data (e.g. spatio-temporal data). Hence, the spiking neural network for reward-based learning is proposed because the SNNs models simulate the biological neural system more closely and they can also process the spatio-temporal data. In this study, the Izikevich spiking model and recurrent neural network are involved in this network learning. For the fish dataset, fish4knowledge was used. The spike encoding was used for feature extraction. The algorithm for this learning model adopted the reward-modulated STDP. The response group in the learning model was determined by the firing rate in the STDP. The experiment performed in the learning rules comes in two types. The first experiment studied the sequence learning 2 points of the fish motion. In the first experiment, the study was divided into two- no repeated points in a sequence and sequence with repeating point. The result for the first experiment got high average performances (correct recall rates) compared to the sequence with repeating point, where 80.43% and 85.57% were for training and testing. The results for the sequence with repeating point were 69.87% and 65.15% for training and testing. For the second experiment, we studied the sequence learning 3 points of the fish motion. The experiment demonstrates that the result for different points without repeated point was significant compared to other experiments. The significant contribution for this study is that the learning rule (e.g. STDP algorithm) has learning capability in memory recall. This learning rule can lead to a novel brain-inspired computer system

Item Type:	Monograph (Technical Report)
Additional Information:	GERAN: FRGS
Subjects:	T Technology > T Technology (General)
Divisions:	Research and Innovation Management Centre (RIMC)
Depositing User:	Mdm. Sarkina Mat Saad @ Shaari
Date Deposited:	12 Dec 2024 08:51
Last Modified:	12 Dec 2024 08:51
URI:	https://repo.uum.edu.my/id/eprint/31752

Actions (login required)

View Item