Jürgen Schmidhuber at AGI-2011: Fast Deep/Recurrent Nets for AGI Vision

The Fourth Conference on Artificial General Intelligence Mountain View, California, USA August 3-6, 2011 Jürgen Schmidhuber's short talk on fast deep neural networks at AGI 2011 at Google Headquarters, CA. Co-authors: Dan Ciresan, Ueli Meier, Jonathan Masci, Alex Graves. The deep / recurrent neural networks of Schmidhuber's team keep winning important visual pattern recognition competitions, and are starting to achieve human-competitive results: 9. August 2011: IJCNN 2011 on-site Traffic Sign Recognition Competition (0.56% error rate, nearly three times better than 2nd best algorithm - the only method outperforming humans) 8. June 2011: ICDAR 2011 offline Chinese Handwriting Recognition Competition (1st & 2nd rank) 7. MNIST Handwritten Digit Recognition Benchmark (perhaps the most famous machine learning benchmark). New record (0.35% error rate) in 2010, improved to 0.31% in March 2011, then 0.27% for ICDAR 2011 6. NORB Object Recognition Benchmark. New record (2.53% error rate) in 2011 5. CIFAR-10 Object Recognition Benchmark. New records in 2011, now down to 12% error rate 4. January 2011: Online German Traffic Sign Recognition Contest (1st & 2nd rank; 1.02% error rate) 3. ICDAR 2009 Arabic Connected Handwriting Competition, like the others below won by LSTM recurrent nets (deep by nature). 2. ICDAR 2009 Handwritten Farsi/Arabic Character Recognition Competition 1. ICDAR 2009 French Connected Handwriting Competition based on data from the RIMES campaign Overview sites with more information and scientific papers: Computer vision with fast deep / recurrent neural networks: http://www.idsia.ch/~juergen/vision.html Handwriting recognition with fast deep / recurrent neural nets: http://www.idsia.ch/~juergen/handwriting.html Formal Theory of Fun & Creativity & Intrinsic Motivation http://www.idsia.ch/~juergen/creativity.html Artificial curiosity - how to build artificial scientists and artists: http://www.idsia.ch/~juergen/interest.html Optimal Universal Artificial Intelligence: http://www.idsia.ch/~juergen/unilearn.html Self-referential Gödel Machines as universal problem solvers: http://www.idsia.ch/~juergen/goedelmachine.html Artificial Evolution: http://www.idsia.ch/~juergen/evolution.html Unsupervised Learning: http://www.idsia.ch/~juergen/ica.html Hierarchical Learning: http://www.idsia.ch/~juergen/subgoals.html Reinforcement Learning: http://www.idsia.ch/~juergen/rl.html Robot Learning: http://www.idsia.ch/~juergen/learningrobots.html Source code of machine learning algorithms at Pybrain: http://pybrain.org/ Home page: http://www.idsia.ch/~juergen/ What's new: http://www.idsia.ch/~juergen/whatsnew.html