Part II. Technology

We now take a deep dive into the technical underpinnings of artificial intelligence. The first three articles in this section discuss issues of broad AI significance before we shift our focus to deep learning. To kick things off, Mike Loukides discusses the balance of supervised and unsupervised learning—in both the human and machine contexts. Surprisingly, although we think of humans as the paradigm unsupervised learners, he points out that as a species, most of our learning is quite supervised. Junling Hu then gives an overview of reinforcement learning, with tips for implementation and a few signal examples. Ben Lorica then dives into compressed representations of deep learning models, in both mobile and distributed computing environments, and Song Han picks up that theme with a deep dive into compressing and regularizing deep neural networks. Next, we shift to deep learning with tutorials for Tensorflow from Aaron Schumacher and Justin Francis, and Mike Loukides’ account of learning “Tensorflow for poets”—as a poet! An alternative to Tensorflow that offers a “define by run” approach to deep learning is Chainer, the subject of Shohei Hido’s contribution. Finally, some thoughts from Rajat Monga on deep learning, both at Google and in any business model, round out the section.