Learning Beam Search Policies via Imitation Learning
This work addresses a fundamental issue in machine learning for structured prediction, offering a more integrated approach to beam search that could improve decoding performance across various applications.
The paper tackles the disconnect between training and test-time beam search in structured prediction by proposing a meta-algorithm for learning beam search policies via imitation learning, which unifies existing methods and provides novel no-regret guarantees.
Beam search is widely used for approximate decoding in structured prediction problems. Models often use a beam at test time but ignore its existence at train time, and therefore do not explicitly learn how to use the beam. We develop an unifying meta-algorithm for learning beam search policies using imitation learning. In our setting, the beam is part of the model, and not just an artifact of approximate decoding. Our meta-algorithm captures existing learning algorithms and suggests new ones. It also lets us show novel no-regret guarantees for learning beam search policies.