无标题文章

DEEP-LEARNING systems tend to be one-trick wonders: great atthe task they were trained to do,but pretty awful at everything else. Now a neural network from Google suggests that AI can be multitalented after all.Most deep-learning systems arebuilt to solve specific problems,such as recognising animals in photos of the Serengeti, ortranslating between languages.But if you take, for instance, animage-recognition algorithm and retrain it to do a completely different task, such as recognising speech, it usually becomes worse at its original job.Humans don’t have that issue. We use our knowledge ofone problem to solve new tasks,and don’t usually forget how touse a skill when we start learninganother. Google’s neural networktakes a tiny step in this direction,by simultaneously learning to solve a range of differentproblems without specialising in any one area.