a method for training neural networks by adjusting their internal connections based on errors — was refined and popularized in the 1980s, even as expert systems grabbed all the headlines.