Non-disruptive spike-based learning

• Applicant: Professor Dr. Raoul-Martin Memmesheimer
• Subject Area: Experimental and Theoretical Network Neuroscience
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 557450678

Neurons in biological neural networks and some of the most recent artificial neural networks communicate with short electrical impulses, called spikes. This communication with discrete all-or-nothing events renders the network dynamics generically discontinuous. Therefore, exact gradient descent learning, which forms the basis of the current success of machine learning, seemed to be inept for spiking neural networks. In contrast, we have recently shown that exact non-disruptive, spike-based gradient descent learning of neural networks, including the addition and removal of spikes, is possible. We now want to establish our approach in neuroscience and engineering and solve several tasks that are relevant on its own: For this we will first compare the performance of our approach to that of existing ones and explore the range of its applicability. Then, we will use it for the building of novel neurobiological models, for data analysis and to solve challenges in the application of spiking neural networks in engineering.

• DFG Programme: Research Grants