notesum.ai

Published at December 10

QuantFormer: Learning to Quantize for Neural Activity Forecasting in Mouse Visual Cortex

q-bio.NC

cs.CV

eess.IV

eess.SP

Released Date: December 10, 2024

Authors: Salvatore Calcagno¹, Isaak Kavasidis¹, Simone Palazzo¹, Marco Brondi², Luca Sità², Giacomo Turri², Daniela Giordano¹, Vladimir R. Kostic², Tommaso Fellin², Massimiliano Pontil², Concetto Spampinato¹

Aff.: ¹University of Catania; ²Italian Institute of Technology

Arxiv: http://arxiv.org/pdf/2412.07264v1

Method	MSE ( $\downarrow$ )	SMAPE ( $\downarrow$ )	Corr ( $\uparrow$ )	SSIM ( $\uparrow$ )
LSTM	60349.339 $\pm$ -*	0.943 $\pm$ 0.037*	0.273 $\pm$ 0.129*	0.003 $\pm$ 0.004*
AutoFormer	19.828 $\pm$ 11.728*	0.800 $\pm$ 0.033*	0.312 $\pm$ 0.085*	0.011 $\pm$ 0.005*
Informer	0.285 $\pm$ 0.376**	0.707 $\pm$ 0.051*	0.302 $\pm$ 0.033*	0.022 $\pm$ 0.017*
BrainLM	0.605 $\pm$ 2.852	0.701 $\pm$ 0.158*	0.253 $\pm$ 0.125*	0.008 $\pm$ 0.034*
BrainLM_ft	0.457 $\pm$ 0.825	0.697 $\pm$ 0.183*	0.337 $\pm$ 0.112**	0.001 $\pm$ 0.035*
Cross-former	2.011 $\pm$ 2.749*	0.723 $\pm$ 0.062*	0.292 $\pm$ 0.087*	0.036 $\pm$ 0.020*
QuantFormer	0.247 $\pm$ 0.078	0.656 $\pm$ 0.137	0.338 $\pm$ 0.075	0.069 $\pm$ 0.062