My main research interests revolve entirely around Neural Computation. Together with several PhD students and collaborators, I am trying to unravel what computations are being carried out by different neural systems, and how and why they are being implemented. Therefore, I am interested in both biological and artificial aspects of neural computation.
On the biological side, my work falls mostly within the field of Computational Neuroscience concerned with how different biological neural systems solve different types of computational problems. In terms of problem domains, I have mostly been involved in the research of visual functions and therefore Computer Vision and Image Processing are central to my investigations. I am currently focusing on the simulation-modelling of the biological retina, through which I hope to:
1) answer several questions pertaining to the structure and function of different retinae,
2) propose new image processing techniques for low-level functions such as illumination normalization and colour correction, and
3) propose new designs for retinal prostheses.
So far, our models of the outer retina have revealed that the relatively simple micro-circuits established between photoreceptors, horizontal cells and bipolar cells are capable of a significant number of distinct image processing functions such as denoising, contrast and saturation enhancement, edge detection and colour normalization (see Figure 1 for an example of local illumination normalization and denoising). We are currently in the process of extending our models of the retina, by adding several amacrine and ganglion cell types, with the aim of understanding retinal colour coding and processing. This understanding should guide us towards achieving the goal of enhancing retinal prostheses for allowing users to experience coherent colour perception.
On the artificial or applied side we are developing new types of hybrid artificial neural networks with a particular emphasis on functional diversity, i.e., Neural Diversity Machines (NDM). The notion of diversity is borrowed from biological neural systems where we observe a significant diversity of neurons in terms of morphology, connectivity, electrophysiology, and other properties. The case of the retina, where on average (across species) there are approximately 55 different types of neuron, is one simple illustration of this point. Initial experiments have shown some promising results and our current priority is to improve the speed and reliability of optimization methods for NDM.
Although NDMs are intended to be general and have already been tested on several unrelated benchmarks from the UCI Machine Learning Repository, one of the target domains which ties NDMs with our more biologically oriented research pertains to the segmentation of histological images of the retina for automated or semi-automated reconstruction of retinal circuits (i.e. connectomics research). The motivation behind this application is to bridge one gap in the chain from biological specimens through data acquisition up to computational modelling. To completely bridge the gaps in this chain means that one day we will be able to scan a neural tissue (e.g., retina) and automatically generate computational insights. As collaboration is essential (e.g., biologists and computer scientists), I look forward to receiving any potential queries from interested parties.
Figure 1: Examples of image processing by a neural model of the outer retina.
Scientific Malaysian Profile:
 Research page. http://baggins.nottingham.edu.my/~kcztm/Research.html
 Interdisciplinary Computing and Complex Systems research group. http://icos.cs.nott.ac.uk/
 Cognitive and Sensory Systems research group. http://www.nottingham.edu.my/Psychology/Research/CognitiveandSensorySystemsGroup.aspx