Researchers from New York University have created an algorithm that mimics human learning abilities and allows computers to view and draw visual concepts at a level that is almost indistinguishable from humans. The findings mark a significant advancement in a field that continues to focus on designing robots and computer algorithms that can learn and adapt in the same way as the human brain.
"Our results show that by reverse engineering how people think about a problem, we can develop better algorithms," Brenden Lake, the paper's lead author, said in a press release. "Moreover, this work points to promising methods to narrow the gap for other machine learning tasks."
When humans encounter a new concept, such as a new dance move or new number, they often need to be exposed to only a few examples in order to understand it and recognize it in new situations. While machines can conduct pattern-replication tasks, they typically need to be given thousands of examples in order to perform them with the same level of accuracy as humans.
"It has been very difficult to build machines that require as little data as humans when learning a new concept," study co-author Ruslan Salakhutdinov said. "Replicating these abilities is an exciting area of research connecting machine learning, statistics, computer vision and cognitive science."
The researchers developed a "Bayesian Program Learning" (BPL) framework, which represents concepts within robots as simple computer programs, in order to shorten the learning process and make robot learning more similar to that of humans in the way that they acquire and apply new knowledge.
The results showed that fewer than 25 percent of human participants performed significantly better than chance when determining whether drawings were created by a machine or human.
"We are still far from building machines as smart as a human child, but this is the first time we have had a machine able to learn and use a large class of real-world concepts - even simple visual concepts such as handwritten characters - in ways that are hard to tell apart from humans," fellow co-author Joshua Tenenbaum said.
The study was published in the Dec. 11 issue of the journal Science.
