Physics-Based Encoding for Spiking Neural Networks
Supervisors
Dr Nimrod Kruger & Prof Paul Hurley
Description
Efficient encoding of spike-based time-surfaces in neuromorphic computing remains an open challenge. While understanding biological neural encoding may eventually shed light on potential schemes, these are unlikely to provide compatibility to digital systems or standardized interfaces.
This project aims to explore the use of mathematical primitives and spike-based neural operators to define interfaces between spike-to-spike or spike-to-digital communication. These primitives and operators can offer interoperability by developers, maintaining the unique benefits of spike-based computing without compromising on compatibility and standardization.
Outcomes
- Literature survey – state-of-the-art interfaces of spike-to-digital and spike-to-spike modalities.
- Formulate spiking neural networks operators and/or mathematical primitives.
- Functional and comparative testing of these elements in context of transfer efficiency, signal fidelity, and robustness.
- Build a model interface for a set of tasks (robot arm control, navigation, user interface, etc.)
Eligibility Criteria
Background in Mathematics, Computer Science, or Mechanical / Electrical Engineering.
