Education: Computer Sciences B.S. + M.S , University of Cincinnati, OH
Extend the Automatic Differentiation Support in RooFit
In terms of minimization time, Roofit offers faster results even with numerical differentiation techniques as compared to minimizing a likelihood function that is written by hand in C++, due its complex caching logic. Automatic differentiation gives an additional speedup and more accuracy and scalability for problems with large number of parameters. The purpose of this project will be to firstly use Minuit as an optimization algorithm with externally provided gradients, extend support to cover HistFactory and other parts of RooFit, and finally to optimize Clad generated derivatives a nd further explore how they can be parallelized (OpenMP or CUDA).
Project Proposal: URL
Mentors: Vassil Vassilev, David Lange