Kai Mason

Rapidly Reconfigurable Dynamic Computing in Neural Networks with Fixed Synaptic Connectivity

This paper proposes that neural networks can learn by modulating neuron excitability instead of changing synaptic strengths, enabling task switching and robustness against interference.

Direct Training of Networks of Morris-Lecar Neurons withBackprop

This paper trains biophysically detailed spiking neural networks using Morris-Lecar neurons and differentiable synapses, showing that they can be effectively trained with backpropagation for various tasks.