AI uses artificial sleep to learn new task without forgetting the last

Jeremy Hsu in New Scientist:

Artificial intelligence can learn and remember how to do multiple tasks by mimicking the way sleep helps us cement what we learned during waking hours.

“There is a huge trend now to bring ideas from neuroscience and biology to improve existing machine learning – and sleep is one of them” says Maxim Bazhenov at the University of California, San Diego.

Many AIs can only master one set of well-defined tasks –  they can’t acquire additional knowledge later on without losing everything they had previously learned. “The issue pops up if you want to develop systems which are capable of so-called lifelong learning,” says Pavel Sanda at the Czech Academy of Sciences in the Czech Republic. Lifelong learning is how humans accumulate  knowledge to adapt to and solve future challenges.

Bazhenov, Sanda and their colleagues trained a spiking neural network – a connected grid of artificial neurons resembling the human brain’s structure – to learn two different tasks without overwriting connections learned from the first task. They accomplished this by interspersing focused training periods with sleep-like periods.

More here.