Anand Subramoney

This paper shows that biologically realistic spiking neural networks can learn from few examples by storing priors in synaptic weights and using network state for rapid information absorption.

The paper shows spike frequency adaptation improves spiking neural network performance on temporally dispersed tasks, reaching human brain levels.

The paper proposes a normative model where top-down inputs in apical dendrites provide probabilistic priors for sensory information, consistent with synaptic plasticity and NMDA-spikes.

This paper introduces a new learning method for spiking neural networks that enables biologically plausible online learning through gradient descent, achieving performance comparable to BPTT.