A new machine-learning training method developed at the University of Toronto enables neural networks to learn directly from human-defined rules. The achievement supports new possibilities for artificial intelligence in medical diagnostics and self-driving cars.
The algorithm learns directly from human instructions, rather than an existing set of examples. According to the team's researchers, the AI algorithm outperformed conventional methods of training neural networks by 160 percent.
Researchers Parham Aarabi and Wenzhi Guo trained the technology to identify a person’s hair in photographs — a much more challenging task for computers than it is for humans.
“Our algorithm learned to correctly classify difficult, borderline cases — distinguishing the texture of hair versus the texture of the background,” said Aarabi. “What we saw was like a teacher instructing a child, and the child learning beyond what the teacher taught her initially.”
Instead of the traditional way to "teach" neural networks — providing a set of labeled data and asking the neural network to make decisions based on the samples — the University of Toronto algorithm uses heuristic training: Humans provide direct instructions that pre-classify training samples rather than a set of fixed examples.
With a heuristic approach, trainers program the algorithm with guidelines such as “Sky is likely to be varying shades of blue,” and “Pixels near the top of the image are more likely to be sky than pixels at the bottom.”
By addressing the challenge of making correct classifications of previously unknown or unlabeled data, the new development could improve machine learning's ability to correctly identify cancerous tissues for medical diagnostics, or classify all the objects surrounding and approaching a self-driving car.
Also: Learn about NASA's shape-sensing algorithms.