Natural Language to Structured Query Generation via Meta-Learning

  • 2018-07-18 21:40:45
  • Po-Sen Huang, Chenglong Wang, Rishabh Singh, Wen-tau Yih, Xiaodong He
  • 0

Abstract

In conventional supervised training, a model is trained to fit all thetraining examples. However, having a monolithic model may not always be thebest strategy, as examples could vary widely. In this work, we explore adifferent learning protocol that treats each example as a unique pseudo-task,by reducing the original learning problem to a few-shot meta-learning scenariowith the help of a domain-dependent relevance function. When evaluated on theWikiSQL dataset, our approach leads to faster convergence and achieves1.1%-5.4% absolute accuracy gains over the non-meta-learning counterparts.

 

Quick Read (beta)

loading the full paper ...