notesum.ai

Published at November 8

physics.optics

cs.AI

cs.AR

Released Date: November 8, 2024

Authors: Ziang Yin¹, Yu Yao¹, Jeff Zhang¹, Jiaqi Gu¹

Aff.: ¹Arizona State University

Tasks

Digital 2-Layer CNN

C32K3-C32K3

Pre-fab

Initial Training

P

=4,

L

=4,

\alpha

\Phi

\beta

Skip

s

\lambda

Post-fab

Task Adaptation

P

=4,

L

=4,

\alpha

\beta

Skip

\lambda

\mathcal{O}

\mathcal{P}

FMNIST

0.9122

0.9014

Quickdraw

0.8978

0.8646

0.7992 (adapt from FMNIST)

CIFAR10

0.7633

0.6182

CIFAR100

0.4347

0.3815

0.2513 (adapt from CIFAR10)

Darcy

0.0466

0.055

NavierStokes

0.1085

0.1007

0.1753 (adapt from Darcy)