itechitsのブログ

​Artificial intelligence​ can only grasp one set of nicely-described tasks –  they can’t collect extra understanding afterward without losing the thing they had formerly discovered. “The difficulty pops up in case you want to increase structures which might be capable of so-known as lifelong mastering,” says Pavel Sanda on the Czech Academy of Sciences within the Czech Republic. Lifelong studying is how humans acquire the know-how to evolve to and remedy destiny-demanding situations.

Bazhenov, Sanda, and their colleagues educated a spiking neural community – a connected grid of artificial neurons comparable to the human mind’s structure – to analyze two exclusive duties without overwriting connections learned from the primary task. They achieved this by interspersing focused training durations with sleep-like periods.

The researchers simulated sleep inside the neural network by activating the network’s artificial neurons in a noisy pattern. They also ensured that the sleep-inspired noise more or less matched the sample of neuron firing for the duration of the training periods – a manner of replaying and strengthening the connections discovered from each obligation.