Program Synthesis meets Machine Learning

Опубликовано 21 марта 2019, 16:56

We give a tutorial overview of program synthesis, from its first formulation by Church in 1957, through its pragmatic evolution through sketching and programing-by-examples, and compare program synthesis with supervised machine learning. We then present our recent efforts in combining program synthesis and machine learning techniques to solve the problem of synthesizing extractors from heterogeneous data. Finally, we explore several opportunities at the intersection of program synthesis (and more broadly the PL community) and machine learning, such as pruning and ranking programs during synthesis, neural program synthesis and automatic differentiation.

See more at microsoft.com/en-us/research/v...