Handwritten Digit Recognition With a Committee of Deep Neural Nets on GPUs

Free registration required

Executive Summary

The competitive MNIST handwritten digit recognition benchmark has a long history of broken records since 1998. The most recent substantial improvement by others dates back 7 years (error rate 0.4%). Recently the authors were able to significantly improve this result, using graphics cards to greatly speed up training of simple but deep MLPs, which achieved 0.35%, outperforming all the previous more complex methods. Here they report another substantial improvement: 0.31% obtained using a committee of MLPs.

  • Format: PDF
  • Size: 342.21 KB