Automatic differentiation in machine learning: a survey

Atılım Güneş Baydin, Barak A. Pearlmutter, Alexey Radul, Jeffrey Mark Siskind

2015Maynooth University ePrints and eTheses Archive (Maynooth University)2,097 citationsOpen Access PDF

Abstract

Derivatives, mostly in the form of gradients and Hessians, are ubiquitous in machine learning. Automatic differentiation (AD) is a technique for calculating derivatives of numeric functions expressed as computer programs efficiently and accurately, used in fields such as computational fluid dynamics, nuclear engineering, and atmospheric sciences. Despite its advantages and use in other fields, machine learning practitioners have been little influenced by AD and make scant use of available tools. We survey the intersection of AD and machine learning, cover applications where AD has the potential to make a big impact, and report on some recent developments in the adoption of this technique. We aim to dispel some misconceptions that we contend have impeded the use of AD within the machine learning community.

Topics & Concepts

Computer scienceArtificial intelligenceRelevance (law)Machine learningAutomatic differentiationDifferentiable functionCLARITYToolboxField (mathematics)Algorithmic learning theoryActive learning (machine learning)Theoretical computer scienceAlgorithmProgramming languageMathematicsMathematical analysisBiochemistryLawComputationPure mathematicsPolitical scienceChemistryGaussian Processes and Bayesian InferenceNeural Networks and ApplicationsComputational Physics and Python Applications