MIT's Technology Review is the only publication to pick up on this, so I'll risk the possible paywall to link to it (though you can see the same point being made on DeepMind's own website). It's this: "The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more..." Or as the MIT article says, "After learning to play Go, it had to forget everything before learning to play chess, and so on. It could not learn to play both games at once. This is what Gato does: it learns multiple different tasks at the same time." I've tried to diagram this in the past, but it's hard to conceptualize without an actual working example. And that's what Gato is. Why is this important? This: direct instruction is the first sort of learning, where you do one thing, then another, etc, while constructivist approaches enable the second sort of elarning, where you're learning multiple things at the same time.
Today: 3 Total: 93 [Direct link] [Share]
