Living organisms are complex systems, and yet they possess an extremely high degree of reliability. Failure mechanisms in nature are often local and their repair is also undertaken at this local level.
In engineering however, we have traditionally approached the problem of unreliability from the system or sub-system level. That is, we have incorporated redundancies by replicating entire systems or sub-systems, in the hope that at least one would still function faultlessly when the others fail.
It has been suggested recently that interaction between neurons and astrocytes may hold the key to repair in large networks of neurons. We could therefore, justifiably ask the question whether it might not be more effective, efficient and less costly to draw inspiration from nature, seeking to learn how it deals with the complexity vs. unreliability issue in such a remarkable way.
This project was an inter-disciplinary collaboration that arose naturally from the combined expertise of the Intelligent Systems Group (ISG) at the University of York (UY) and the Intelligent Systems Research Centre (ISRC) at Ulster University (UU).
The fundamental astrocyte-neuron computer model for self-repair proposed in this project was proven in an earlier EPSRC eFutures funded project (EFXD12011) where it was demonstrated that the co-existence of astrocytes with spiking neurons in a network can yield a fault diagnostic and repair capability at the cellular level, which addresses current hardware reliability challenges [1-6].
This project demonstrated that the self-repairing spiking neural network is capable of diagnosing faults and subsequently performing repairs beyond existing levels, where the repair capability was showcased in hardware using real-time robotic applications.
Projected key outcomes
- Developed a modular astrocyte-neuron self-repair software model
- Hardware translation and evolvable hardware infrastructure developed
- Robotic systems demonstrator and performance evaluated
- Significant number of pier reviewed publications