PI: Håkan Grahn, Blekinge Institute of Technology
co-PI: Christoph Kessler, Linköping University
co-PI: Flavius Gruian, Lund University
Energy consumption is one of the main challenges of today’s AI systems. Most AI systems today are designed using various forms of neural networks and deep learning. However, training and even inference are very costly in terms of time, computational demands, and energy consumption. One promising alternative is spiking neural networks (SNNs) executed on neuromorphic hardware. Neuromorphic computing tries to mimic how the brain works by relying on changes in signals, i.e., ”spikes”, rather than continuously recalculating numerical values as in traditional artificial neural networks. However, challenges that arise that we will address in this project are, e.g., efficient training of SNN models, programming abstractions for SNNs, development and mapping of applications to SNNs, and balancing system energy, latency, and cost.
Project number: F5