FORCE training with slow learning rates leads to strongly correlated and swapable decoders across spiking LIF and LIF-matched rate networks.

FORCE training with slow learning rates leads to strongly correlated and swapable decoders across spiking LIF and LIF-matched rate networks.

<p><b>A–C</b> Networks of 2000 neurons were trained on different supervisors over a grid of points in the (<i>Q</i>,<i>G</i>) parameter space for both spiking LIF and LIF-matched rate networks. The learning rates used were: , , and for the pitchfork, Ode to...

Full description

Saved in:

Bibliographic Details
Main Author:	Thomas Robert Newton (21756023) (author)
Other Authors:	Wilten Nicola (4951927) (author)
Published:	2025
Subjects:	Cell Biology Neuroscience Science Policy Biological Sciences not elsewhere classified supervised learning tasks slower learning rates slow learning rates longer fully interchangeable fast learning rates force hyperparameters identical force effectively act trial firing rate spiking network could matched instantaneous rate applied force training fire spiking networks spiking networks approximate force trained spiking decoding solutions learned rate network matched rate force training spiking networks rate networks training networks weights trained spiking neurons weight solutions weight matrices vice versa successful convergence selectively stabilizing results indicate principal components pass filter neural bases little effect leaky integrate inherent spike increasing instability findings suggest analytic derivation
Tags:	Add Tag No Tags, Be the first to tag this record!