Researchers at Google‘s X lab and Stanford University revealed that a large neural network had taught itself to recognise objects, including cats and human faces, from unlabelled images. The work was led by researchers including Andrew Ng and Jeff Dean.
What happened
- The system ran on 1,000 computers with a combined 16,000 processor cores.
- It was shown 10 million still images taken from YouTube videos over three days, without being told what they contained.
- The network developed features that responded to faces, human bodies, and cat faces.
- Researchers reported it detected human faces with around 82% accuracy in tests, although overall accuracy across thousands of object categories was far lower.
Why it mattered
The ‘cat paper’ became one of the best-known early demonstrations of deep learning at scale, showing machines could learn useful concepts without labelled data. It helped build momentum behind Google Brain and modern AI research.
Lessons for organisations
AI capability grows with data and computing power, so organisations adopting AI should understand what data their models learn from and document it. Early governance helps avoid later surprises around bias and intellectual property.
Part of our Top stories archive of headline-making events in information security, privacy, and AI. If you would like help applying the lessons to your organisation, contact us.
