Sitzungen | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

FM: Fall Meeting

FM 82: Quantum & Information Science: Neural Networks, Machine Learning, and Artificial Intelligence III

FM 82.5: Talk

Donnerstag, 26. September 2019, 15:15–15:30, 3044

DSEA+: Deconvolution by Machine Learning — •Tim Ruhe¹, Mirko Bunse², Kai Brügge¹, and Tobias Hoinka¹ — ¹Lehrstuhl Experimentelle Physik 5, TU Dortmund — ²LS8, Fakultät Informatik, TU Dortmund

The reconstruction of an experimentally inaccessible quantity, e.g. a particle's energy, is a common challenge in particle- and astroparticle physics, where correlated observables are measured instead. The transfer from the variable of interest into an experimentally observable quantity is, however, usually governed by stochastical processes, leading to the Fredholm integral equation of the first kind. Additional smearing, stemming from particle propagation and the detector itself, complicate the problem even further. We present a novel machine learning-based approach, DSEA+, which sidesteps certain limitations of existing algorithms by interpreting deconvolution as a multinominal classification task. We discuss the algorithm and show results obtained from simulations provided by the FACT Open Data Project.