Metalab researcher, Zheng Wang, has won the best paper award in The 26th International Conference on Parallel Architectures and Compilation Techniques (PACT), 2017.
The paper is co-authored by Zheng with Chris Cummins, Pavlos Petoumenos and Hugh Leather from University of Edinburgh.
This work develops an end-to-end deep learning based system to automatically construct compiler optimisation heuristics without human involvement.
The tool is opensourced and can be downloaded from https://github.com/ChrisCummins/paper-end2end-dl
PACT is a flagship conference in parallel architectures and computing. “The purpose of PACT 2017 is to bring together researchers from architecture, compilers, applications and languages to present and discuss innovative research of common interest. PACT started as a Data Flow Workshop in conjunction with ISCA 1989 but quickly evolved into a unique venue at the intersection of classical parallel architecture and compilers. Recently, PACT widened its scope to include insights useful for the design of machines and compilers from applications such as, but not limited to, machine learning, data analytics and computational biology.”
PACT 2017 accepts 25 papers out of 108 submission with an acceptance rate of 23%.