Task-Driven Model Evaluation in Large-Scale Spiking Neural Networks
Supervisors
Dr Paul Kirkland
Description
This research focuses on evaluating the effectiveness of a modular platform for large-scale spiking neural networks, aiming to enhance neuromorphic systems for complex task-driven applications. The project investigates how combining small-scale, multi-task networks with dendritic connections can detect temporal and spatial activation patterns. It also explores the use of large-scale spiking neural network models to solve task-oriented problems. A key aspect of the research is examining structural plasticity to understand how network structures adapt during task execution.
Insights from machine learning and deep learning will be utilised alongside principles from neuroscience to optimise network performance. The final stage involves mapping these solutions onto a large-scale neural network simulator to assess the platform’s ability to efficiently handle complex spiking neural network tasks.
Key Components
- Small-scale multi-task networks: Investigating how dendritic structures detect temporal and spatial patterns.
- Large-scale network models: Applying spiking neural networks to solve task-oriented problems in a computationally efficient manner.
- Structural plasticity: Exploring adaptive neural network structures informed by neuroscience and deep learning insights.
- Simulation integration: Mapping solutions onto a large-scale neural network simulator to evaluate task performance.
This is a multi-partner collaborative project supported by the Volkswagen Foundation, in collaboration with the Universities of Osnabrück, Sussex, and Forschungszentrum Jülich, ensuring an interdisciplinary and innovative approach.
