And for Google X’s Next Trick, It Will Identify Cat Videos
A neural network of 16,000 computers presented with stills from 10 million YouTube videos taught itself to recognize cats.
The Stanford-Google paper identifies these pictures as “the most responsive stimuli on the test set for the cat neuron.”
The seemingly trivial task is actually a huge achievement for machine learning, as John Markoff reports in the New York Times. The point is to train a face detection tool from unlabeled images.
Yup, cat video identification is the height of science!
The research project was designed and studied by a group including Andrew Ng of Stanford University and Jeff Dean of Google. It has now been moved out of Google X and into Google’s regular search and knowledge work.
The Stanford-Google system was 70 percent more accurate at recognizing object categories than previous efforts.
As a remarkable side note, Ng is on leave from Stanford to co-found and run the online learning start-up Coursera. Meanwhile, Sebastian Thrun — who has led Google X projects for self-driving cars and augmented reality glasses — was also until recently a Stanford professor until leaving to start the online learning start-up Udacity. Talk about interesting parallels.
