Spiking neural network for energy efficient learning and recognition

Lo, Wang Ning (2020) Spiking neural network for energy efficient learning and recognition. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Spiking neural network for energy efficient learning and recognition.pdf - Submitted Version Download (2MB)

Abstract

People are confronted with an increasingly large amount of data and a tremendous change of human-machine interaction modes. It is a challenging and time-consuming task for traditional computing system to deal with the content of information. The use of applications consumes energy and hard to perform through standard programmed algorithms. Spiking neural networks have emerged that achieve favourable advantages in terms of energy and time efficiency by using spikes for computation and communication as well as solving different problems such as pattern classification and image processing. Therefore, an energy-efficient spiking feedforward computing system is designed to evaluate its performances. Common building blocks and techniques used to implement a spiking neural network are investigated to identify design parameters for hardware-based neuron implementations. IZH neuron, AER system and STDP module are developed by using Vivado software. Hardware implementation for digit recognition is implemented on FPGA to fit the behaviour of biological neural networks, showing the potential for training neural cells into biological processors. The energy consumption of the system is only 136mW and low hardware resource utilization. Hence, spiking feedforward computing system exhibits its significant properties with low energy consumption for this project.

Actions (login required)

View Item