Conference Proceedings
The use of adversaries for optimal neural network training
Anton Hawthorne-Gonzalvez, Martin Sevior, A Forti (ed.), L Betev (ed.), M Litmaath (ed.), O Smirnova (ed.), P Hristov (ed.)
23RD INTERNATIONAL CONFERENCE ON COMPUTING IN HIGH ENERGY AND NUCLEAR PHYSICS (CHEP 2018) | E D P SCIENCES | Published : 2019
Abstract
B-decay data from the Belle experiment at the KEKB collider have a substantial background from e+e- -h> qq¯ events. To suppress this we employ deep neural network algorithms. These provide improved signal from background discrimination. However, the deep neural network develops a substantial correlation with the ∆E kinematic variable used to distinguish signal from background in the final fit due to its relationship with input variables. The effect of this correlation is reduced by deploying an adversarial neural network. Over-all the adversarial deep neural network performs better than a Boosted Decision Tree algorithimn and a commercial package, NeuroBayes, which employs a neural net with ..
View full abstractGrants
Awarded by Australian Research Council
Funding Acknowledgements
We gratefully acknowledge our colleagues within the Belle collaboration for all the work required to make this investigation possible and for their permission to employ Belle data and tools for this study. We would particularly like to thank Thomas Keck of the Karlsruher Institut fiir Technologie for very helpful discussions about adversarial neural networks. We thank the KEKB group for the excellent operation of the accelerator; the KEK cryogenics group for the efficient operation of the solenoid; and the KEK computer group, the National Institute of Informatics, and the PNNL/EMSL computing group for valuable computing and SINET5 network support. We acknowledge financial support from the Australian Research Council grant DP180102629.