A major focus of the CNET research group is on the development of computational models of brain function and to exploit the properties of neural based systems in the development of new computing paradigms in software.
We specifically use our computational models to deepen our understanding of the specific signalling pathways that brain cells use to communicate with each other.
These pathways are orchestrated by a complex biochemical and biophysical processes occurring at the cellular and network level. Our approach is to implement multi-scale models of neural and neural-glial networks with specific interest in understanding the synaptic and cellular mechanisms underpinning plasticity. CNET is also focused on taking inspiration from these brain models in the development of new intelligent computational systems.
The CNET group is currently developing a novel astro-centric hardware platform (EMBRACE) to support acceleration and emulation of key neural behaviours. The platform will open up a new direction of research whereby neuroscientists and computational neuroscientists will be able to explore the complex and detailed interactions between glial and neurons at the level of networks and also how these interaction play out in both the functional and dysfunctional brain.
The research focus of this team is the development of computational approaches to neural architectures and characteristics inspired from biology.
The unifying theme for team members is the investigation, development and optimisation of large scale neural systems that emulate biological sensory capabilities such as vision, sound and haptics; with the appropriate systems emulated on reconfigurable hardware.
Such motivation has encouraged the exploration of spiking neural networks models, their topologies and training regimes. Current research is targeted at the emulation and modeling of visual processing capabilities of the human brain and will build on the existing collaboration and establish additional links with the leading neuroscience research in this area.
The research demands the development of new learning algorithms, encoding schemes and topologies to facilitate the realisation of the biologically inspired architectures on reconfigurable platforms.Previous research within the group has demonstrated that relatively large architectures can be realised at speeds faster than real-time using a time-multiplexed approach. While these interim results demonstrate impressive performance, further advances are possible by exploiting the full parallel processing capabilities of the target hardware and optimising the designs through the use of event based and multiplier-less approaches.
The research outputs will be exploited via application to sensory fusion in robotics and feature extraction in medical and process control problems. A secondary area of research is the incorporation of the biologically inspired healing characteristic to FPGA implementations of critical processing applications. Ultimately, this self-repair capability will be an inherent characteristic of the large scale neural implementations.