Google researchers have built a 9-layer neural network that can learn to detect faces using only unlabeled images. The researchers built a neural network, which mimics the working of a biological brain, that worked out how to spot pictures of cats in just three days.

The cat-spotting computer was created as part of a larger project to investigate machine learning.

The model has 1 billion connections and the dataset has 10 million 200x200 pixel images randomly downloaded from the internet. It was trained on a cluster of 1,000 machines (16,000 cores) for three days.

The results show it's possible to train a face detector without having to label images as containing a face or not. The detector is robust to translation, scaling, and out-of-plane rotation. It's also sensitive to other high-level concepts such as cat faces and human bodies.

According to Google, the network obtained 15.8% accuracy in recognizing 20,000 object categories, a leap of 70% relative improvement over the previous state-of-the-art.

The team is presenting a paper on its findings at the International Conference on Machine Learning that is being held in Edinburgh from 25 June to 1 July.