Architecture-independent approximation discovery
US11521076B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Jun 29, 2018 |
| Grant date | Dec 6, 2022 |
| Priority date | — |
| Expiry date | Jun 5, 2041 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06N5/01
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
Systems and methods for discovering approximations for compilers to apply through genetic programming and deterministic symbolic regression heuristics are provided. A method for discovering approximations for compilers and runtime optimization can include profiling a program to identify performance critical functions, determining appropriate candidates for approximation and developing application and architecture specific approximations through machine learning techniques, genetic programming, and deterministic heuristics. Such approximations can target any optimization goal, with a primary emphasis on parallelism, or can provide a set of Pareto-optimal tradeoffs.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.