75 - Reinforcement / Imitation Learning in NLP, with Hal Daumé III
NLP Highlights
00:00
Structured Prediction Problems as Reinforcement Learning Problems
Imitation learning is becoming almost the de facto way of thinking about these problems. In structured prediction, you can easily produce and best lists,. And so I think this gives you a bit more power and leverage that you can use to make things easier. But if you're treating structured prediction as a reinforcement learning problem, you basically have a fixed known reward at every step.
Transcript
Play full episode
