As the data collection grows larger and computational/statistical techniques become more complex, many physics analysis users are experiencing a "two-language problem"1 without knowing it.
Julia and the ever-growing JuliaHEP2 ecosystem aim to provide end-users the ability to chew through a larger amount of data faster, by being a JIT language from the ground up. And also enable users to do custom machine learning and training because the pure-Julia ecosystem would allow automatic differentiation to propagate freely without foreign-library call barrier.
The poster presenter would talk about:
- Julia and why it's designed precisely for workloads like physics analysis
- How mono language enables effortless parallelization and automatic differentiation
- UnROOT.jl, BAT.jl, pyhf, FHist.jl, etc. and workflow for an end-user analysis in Julia
- Custom training and inference loop based on data before N-tupleized -- enable deeper insight into raw event data, supported by Julia's speed and auto diff ability.
|In-person or Virtual?||In-person|