Design of Neuromorphic Spiking Neural Networks for Real-Time Processing

Supervisors:

Primary supervisor Dr Saeed Afshar

Description:

Spiking Neural Networks (SNNs) are the third generation of neural networks and come with several advantages including higher energy efficiency and bio-plausible computational models. Unlike traditional neural networks, SNNs operate based on spikes, which are discrete events that take place at points in time, similar to how neurons in the human brain communicate.

In this project, we will harness the power of neuromorphic engineering to design and develop SNNs for real-time processing tasks. Our primary focus will be on creating networks that not only offer superior performance but also are more energy-efficient and offer faster response times than their conventional counterparts.

Neuromorphic engineering, which designs hardware and algorithms based on the structure and function of the brain, can bring several advantages to this project. For instance, neuromorphic hardware, like event-based sensors and neuromorphic processors, can help significantly reduce power consumption. These devices can operate continuously, consuming power only when processing actual changes in the input, making them ideal for real-time monitoring applications.

Furthermore, we can leverage machine learning algorithms, such as SNNs, designed to run on such neuromorphic hardware. These networks can learn and adapt in a manner that's more analogous to biological brains, providing superior accuracy and adaptability.

Outcomes:

The aim of this project is to design, implement, and validate neuromorphic spiking neural networks (SNNs) that are capable of real-time processing. By capitalizing on the inherent computational efficiency and bio-plausible characteristics of SNNs, we aspire to create systems that can perform complex tasks with lower power consumption and latency compared to conventional neural networks. The developed networks will pave the way towards more efficient, adaptive, and intelligent hardware and software systems. This will include the following tasks:

• Literature Review: Carry out a thorough review of existing research on SNNs, neuromorphic hardware, and real-time processing applications.
• Network Design: Design neuromorphic SNNs based on the review and the requirements of the chosen real-time processing tasks.
• Network Implementation: Implement the designed networks using appropriate neuromorphic hardware and software tools.
• Validation and Optimization: Test the implemented networks on various real-time processing tasks. Depending on the initial results, fine-tune the network parameters for increased accuracy and efficiency.
• Evaluation: Evaluate the performance, power efficiency, and response time of the developed networks against conventional neural networks.
• Communication and Publication: Document the results in a form suitable for publication in a scientific journal. Also, present the results at relevant conferences and workshops.

Eligibility criteria:

Experience in Python or C++ for network design and testing is necessary. Familiarity with neuromorphic hardware and software, SNN models, machine learning and data analysis is also beneficial. Lastly, a background in computer architecture or digital signal processing could prove helpful in the network design and implementation process.