notesum.ai

Published at December 3

Reproduction of AdEx dynamics on neuromorphic hardware through data embedding and simulation-based inference

cs.NE

Released Date: December 3, 2024

Authors: Jakob Huhle¹, Jakob Kaiser¹, Eric Müller¹, Johannes Schemmel¹

Aff.: ¹Kirchhoff Institute for Physics, Heidelberg University, Germany

Arxiv: http://arxiv.org/pdf/2412.02437v1

Refer to caption

Layer	Activation	Output Shape	Kernel Shape	Trainable Parameters
Encoder
Conv 1D	ReLU	$(32,\,1024)$	$5$	$192$
BatchNorm 1D	–	$(32,\,1024)$		$64$
MaxPool 1D	–	$(32,\,512)$	$2$
Conv 1D	ReLU	$(16,\,512)$	$3$	$1552$
BatchNorm 1D	–	$(16,\,512)$		$32$
MaxPool 1D	–	$(16,\,256)$	$2$
Conv 1D	ReLU	$(64,\,256)$	$11$	$11\,328$
MaxPool 1D	–	$(64,\,128)$	$2$
Conv 1D	ReLU	$(128,\,128)$	$13$	$106\,624$
MaxPool 1D	–	$(128,\,64)$	$2$
Conv 1D	ReLU	$(1,\,64)$	$3$	$385$
MaxPool 1D	–	$(1,\,32)$	$2$
Decoder
Conv 1D	ReLU	$(128,\,32)$	$3$	$512$
Upsample	–	$(128,\,64)$
Conv 1D	ReLU	$(64,\,64)$	$13$	$106\,560$
Upsample	–	$(64,\,128)$
Conv 1D	ReLU	$(16,\,128)$	$11$	$11\,280$
Upsample	–	$(16,\,256)$
Conv 1D	ReLU	$(32,\,256)$	$3$	$1568$
Upsample	–	$(32,\,512)$
Conv 1D	–	$(1,\,512)$	$5$	$161$
Upsample	–	$(1,\,1024)$
Total				$240\,258$