Title :
Better Digit Recognition with a Committee of Simple Neural Nets
Author :
Meier, Ueli ; Ciresan, Dan Claudiu ; Gambardella, Luca Maria ; Schmidhuber, Jürgen
Author_Institution :
IDSIA, USI, Manno-Lugano, Switzerland
Abstract :
We present a new method to train the members of a committee of one-hidden-layer neural nets. Instead of training various nets on subsets of the training data we preprocess the training data for each individual model such that the corresponding errors are decor related. On the MNIST digit recognition benchmark set we obtain a recognition error rate of 0.39%, using a committee of 25 one-hidden-layer neural nets, which is on par with state-of-the-art recognition rates of more complicated systems.
Keywords :
error analysis; learning (artificial intelligence); neural nets; pattern recognition; set theory; MNIST digit recognition benchmark set; digit recognition error rate; nets training; one hidden layer neural nets; state-of-the-art recognition rate; training data preprocess; training data subsets; Benchmark testing; Error analysis; Neural networks; Shearing; Text analysis; Training; Committee; Handwritten Digits Recognition; MNIST; Neural Networks;
Conference_Titel :
Document Analysis and Recognition (ICDAR), 2011 International Conference on
Conference_Location :
Beijing
Print_ISBN :
978-1-4577-1350-7
Electronic_ISBN :
1520-5363
DOI :
10.1109/ICDAR.2011.252
