An algorithm that learns directly from human instructions instead of an existing set of examples has been designed. This new type of algorithm has already outperformed conventional methods of training neural networks by an amazing 160%. More importantly, it has also outperformed its own training by 9%.
Researchers from the University of Toronto Engineering, Parham Aarabi and Wenzhi Guo, have trained their new algorithm to identify people’s hair in photographs, which is a rather challenging task for computers. But surprisingly, their algorithm learned to recognize hair in pictures with greater reliability than that enabled by the training, which is a significant leap forward for artificial intelligence and algorithms in general.
“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.”
The new model of algorithm uses a method called heuristic training, where humans provide direct instructions used to pre-classify training samples. Here, there is no set of fixed examples. In other words, this method holds promise for addressing one of the main challenges for neural networks, which is making accurate classifications of unknown and/or unlabeled data.
U of T Engineering News (http://news.engineering.utoronto.ca/new-ai-algorithm-taught-humans-learns-beyond-training/?_ga=1.72517523.1645278805.1479309944)