User-configurable, high-level transformations with CodeBoost
Abstract
Scalable design of large applications requires domain-specific, high-level abstraction. Classically, there has been a tension between abstraction and good performance. We demonstrate how to overcome this tension by augmenting a numerical library providing domain-specific, high-level abstractions with user-configurable optimizations for matrix operations and differentiation.
Publisher
The University of BergenCopyright
The authorCopyright the author. All rights reserved