Updated
AAD in three short videos: theory, tools, large libraries
Three short videos on algorithmic differentiation: the theory and a manual implementation, how automatic tools work, and how to bring AAD into a large quant library.
Sensitivities drive the heaviest calculations in finance, and bumping each input is often impractical. These three short videos introduce algorithmic differentiation from the ground up.
Part one covers the theory and a manual implementation, part two the two main families of automatic tools, operator overloading and source transformation, and part three the practical questions of bringing AAD into a library of a million lines.
- Format
- Three videos, about 40 minutes in total
- For
- Quants and quant developers
- Part one: AD theory and a manual implementation about 11 minutes
- Part two: automatic AD tools, operator overloading and source transformation about 9 minutes
- Part three: AAD in a large quant library about 22 minutes
More on AAD and XAD
- Case studyDaily model calibration with AADHow adjoint gradients with XAD took a tier-one bank's morning stochastic local volatility calibration across about 300 underlyings from three hours to 25 minutes, with risk through the calibration.
- Case studyFrom AAD training to AAD in productionHow a bank's quant team went from the AAD course to its first XAD pricer checked against bumping within six weeks, and into production about four months later, after validation.
Related
- XADXAD, the AAD library for C++ and PythonAutomatic differentiation for pricing and risk code, open source or by subscription.
- TrainingAAD trainingThe AAD course for quant teams, with a workshop on your own code.
- ExplainerWhat is AAD (adjoint algorithmic differentiation)?The idea in plain words, with a worked example.