Automatic Differentiation: State of the Art and Collaboration Opportunities

Speaker: Paul Hovland (ANL)
Date: Thursday, 3 December 2015, 13:30-15:00
Session: Double Feature: Automatic Differentiation
Talk type: Keynote talk (45 min)

Abstract: We provide an overview of the state of the art in automatic differentiation (AD), also called algorithmic differentiation. AD is a methodology for transforming code for computing a mathematical function into code for computing the derivatives of that function. AD implementations typically rely on compiler analysis, heuristic solution of combinatorial problems, and analysis of storage/recomputation tradeoffs. We survey available AD tools, discuss implementation details, and describe several remaining challenges that may present collaboration opportunities.