notesum.ai

Published at November 27

Surveying the space of descriptions of a composite system with machine learning

cs.IT

cs.LG

math.IT

physics.data-an

Released Date: November 27, 2024

Authors: Kieran A. Murphy¹, Yujing Zhang², Dani S. Bassett^{1,3,4,5,6,7,8}

Aff.: ¹Dept. of Bioengineering, School of Engineering & Applied Science, University of Pennsylvania; ²School of Industrial Engineering, Purdue University; ³Dept. of Electrical & Systems Engineering, School of Engineering & Applied Science, University of Pennsylvania; ⁴Dept. of Neurology, Perelman School of Medicine, University of Pennsylvania; ⁵Dept. of Psychiatry, Perelman School of Medicine, University of Pennsylvania; ⁶Dept. of Physics & Astronomy, College of Arts & Sciences, University of Pennsylvania; ⁷The Santa Fe Institute; ⁸Montreal Neurological Institute, McGill University

Arxiv: http://arxiv.org/abs/2411.18579v1

Parameter	Value
Bottleneck embedding space dimension	2
Encoder MLP architecture	[512 leaky_ReLU, 512 leaky_ReLU]
InfoNCE similarity metric $s(u,v)$	Euclidean squared
InfoNCE space dimensionality	32
InfoNCE temperature	1
Batch size	576
Optimizer, latent encodings	SGD
Learning rate, latent encodings	$1\times 10^{-2}$
Optimizer, InfoNCE	Adam
Learning rate, InfoNCE	$1\times 10^{-4}$
Transmitted information coefficient $\gamma_{0}$	1
Transmitted information coefficient $\gamma_{1}$	10
Training steps	$2\times 10^{4}$