[Comp-neuro] Neural Network ReNNaissance / Visual Pattern Recognition Benchmarks / Jobs

Schmidhuber Juergen juergen at idsia.ch
Tue Feb 8 11:50:27 CET 2011

Recently our neural computer vision team collected a string of 1st  
ranks in the following visual pattern recognition competitions and  

1. Online Traffic Sign Recognition Competition, U. Bochum (1st & 2nd  
rank; 1.02% error rate, January 2011).
2. NORB Object Recognition Benchmark, NYU, 2004. New record (2.53%  
error rate) in January 2011 [1].
3. CIFAR-10 Object Recognition Benchmark, U. Toronto, 2009. New record  
(19.51% error rate) in January 2011 [1].
4. MNIST Handwritten Digit Recognition Benchmark, NYU, 1998. New  
record (0.35% error rate) in 2010, tied in January 2011 [1].
5.-7. Three Connected Handwriting Recognition Competitions at ICDAR  
2009, all won by our multi-dimensional LSTM recurrent neural networks.

None of 1-7 requires the traditional sophisticated computer vision  
techniques developed over the past six decades or so. Instead, our  
biologically rather plausible neural architectures learn to recognize  
objects from numerous training examples. We use supervised feedforward  
or recurrent neural networks with many non-linear processing stages.  
GPUs speed up learning by a factor of up to 50. No need for  
unsupervised pre-training, which is a bit depressing, as we have  
developed unsupervised learning algorithms for two decades.   
Nevertheless, we think our results are contributing to a second Neural  
Network ReNNaissance (the first one happened in the 1980s and early  
90s). Most recent paper:
[1] D. C. Ciresan, U. Meier, J. Masci, L. M. Gambardella, J.  
Schmidhuber. High-Performance Neural Networks for Visual Object  
Classification. Preprint arXiv:1102.0183v1 [cs.AI], 1 Feb 2011.

Overview web sites:

We are hiring postdocs and PhD students in
neural vision & decision making and related fields:
Postdoc SFR 72,000/year ~ US$ 75,400/year as of 4/2/2011;
PhD student SFR 40,800/year ~ US$ 42,700/year. Instructions:
BTW, Switzerland is the best country for scientists:
http://www.idsia.ch/~juergen/switzerland.html     :-)

Juergen Schmidhuber
Director of the Swiss AI Lab IDSIA, Lugano
Professor of Artificial Intelligence, Univ. Lugano
Professor SUPSI, Manno-Lugano, Switzerland
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.neuroinf.org/pipermail/comp-neuro/attachments/20110208/4b057013/attachment.html

More information about the Comp-neuro mailing list