Skip to content

Latest commit

 

History

History
20 lines (13 loc) · 1.06 KB

index.md

File metadata and controls

20 lines (13 loc) · 1.06 KB
Raw

Ampersand

The Ampersand project aims to build a new generation of neurosymbolic algorithms for the automatic synthesis of programs and mathematical proofs from high-level goals. Classical methods for these problems are based on rule-based search. In contrast, recent learning-based approaches treat code and proofs as text to which large statistical models can be directly applied. We aim to offer the best of the two worlds by exposing machine learning models to the formal semantics of proofs and code, and by guiding rule-based searches using learned models.

People

  • Yeming Wen (UT Austin)
  • Mingchao Jiang (Rice University)
  • Thomas Logan (UT Austin)
  • Dipak Chaudhari (UT Austin)
  • Rohan Mukherjee (Amazon)
  • Swarat Chaudhuri (UT Austin)
  • Chris Jermaine (Rice University)
  • Thomas W. Reps (University of Wisconsin)

Publications

  • Rohan Mukherjee, Yeming Wen, Dipak Chaudhari, Thomas W. Reps, Swarat Chaudhuri, and Chris Jermaine. Neural Program Generation Modulo Static Analysis. Neural Information Processing Systems (NeurIPS), 2021. (Spotlight paper)