Object Recognition in a Distributed Neural Network

As reported in the New York Times, A team of scientists at Google, including Marc'Aurelio Ranzato (Ph.D. NYU Computer Science, 2009), implemented a distributed neural network with 1 billion connection over a network with 16,000 processors (1000 machines). Applying "deep learning" to a dataset of 10,000,000 unlabelled images, the network achieved an accuracy of 15.8% in identifying 20,000 different categories, an improvement of 70% over the state of the art. The article also quotes Yann LeCun on the application of deep learning techniques to speech recognition.