The brain’s processing ability to solve complex problems has inspired many researchers to investigate how the brain processes information, reasons and its learning mechanisms. Inspired by the biological brain, Artificial neural networks (ANNs) have emerged as a machine learning tool to process information. In recent years, deep neural networks have achieved impressive results, especially in computer vision. The convolutional neural network (CNN) is a famous deep neural network, and it usually applies a supervised learning method using backpropagation approaches. Although, it needs a large volume of labelled data, its classification accuracy is impressive. In the convolutional neural network, each neuron has a continuous output value. However, biological neurons generate spikes to process and transmit information. This has resulted in a new area of artificial neural networks, where the focus is placed on more biologically plausible neuronal models known as Spiking Neural Networks (SNNs). Thanks to their ability to capture the rich dynamics of biological neurons and to represent and integrate different information dimensions such as time, frequency, phase etc., SNNs offer a promising computing paradigm more aligned to biological signal processing and their underlying computational models are potentially capable of modelling complex information processing in the brain  .
Training a deep spiking neural network is considerably more challenging as compared to a classical deep neural network . The non-continuous nature of spiking neuron activities makes it difficult to use usual backpropagation method to train SNNs. A bio-inspired local learning rule, like Spike Timing Dependent Plasticity (STDP), can be used to overcome the training difficulty of a multilayer spiking neural network. This project, which is based on our previous published works,   , seeks to develop a new learning algorithm for SNNs.
The aim of this project is to design novel biologically plausible algorithms for deep SNNs using existing computational models. The designed approaches will be evaluated on a range of benchmark datasets and real world information processing applications. To achieve this aim the following objectives are required:
- To investigate and evaluate representative biologically plausible SNN learning algorithms.
- To explore synaptic plasticity features such as short term plasticity and dynamic synapses along with Spike Time-Dependent Plasticity (STDP) to design new supervised and unsupervised learning algorithms for SNNs.
- To explore convolutional SNNs.
- To develop new learning method for deep SNNs.
A suitable candidate should have a background in Machine learning, and Artificial Neural Networks. It is essential that the candidate has good programming skills. The ability to code with Python is an advantage.
Entrants must have a first/undergraduate Honours degree, with an Upper Second Class or a First Class grade, in computer science. Entrants with a Lower Second Class grade at first degree must also have a postgraduate Masters Degree at Merit or Commendation.
Fees and funding
This is a self-funded PhD opportunity.
Guidance and support
Further guidance and support on how to apply can be found on this page.