Deep Big Simple Neural Nets Excel on Handwritten Digit Recognition
IDSIA, Galleria 2, 6928 Manno-Lugano, Switzerland
arXiv:1003.0358v1 [cs.NE] (1 Mar 2010)
@article{2010arXiv1003.0358C,
author={Claudiu Ciresan}, D. and {Meier}, U. and {Gambardella}, L.~M. and {Schmidhuber}, J.},
title={“{Deep Big Simple Neural Nets Excel on Handwritten Digit Recognition}”},
journal={ArXiv e-prints},
archivePrefix={“arXiv”},
eprint={1003.0358},
keywords={Computer Science – Neural and Evolutionary Computing, Computer Science – Artificial Intelligence},
year={2010},
month={mar},
adsurl={http://adsabs.harvard.edu/abs/2010arXiv1003.0358C},
adsnote={Provided by the SAO/NASA Astrophysics Data System}
}
Good old on-line back-propagation for plain multi-layer perceptrons yields a very low 0.35% error rate on the famous MNIST handwritten digits benchmark. All we need to achieve this best result so far are many hidden layers, many neurons per layer, numerous deformed training images, and graphics cards to greatly speed up learning.
February 9, 2011 by hgpu
