Multi-scale cross-attention transformer encoder for event classification

About

We deploy an advanced Machine Learning (ML) environment, leveraging a multi-scale cross-attention encoder for event classification, taking gg→H→hh→bbbb process at the High Luminosity Large Hadron Collider (HL-LHC) as an example. In the boosted Higgs regime, the final state consists of two fat jets. Our multi-modal network can extract information from the jet substructure and the kinematics of the final state particles through self-attention transformer layers. The learned information is subsequently integrated to improve classification performance using an additional transformer encoder with cross-attention heads. We demonstrate that our approach surpasses in performance current alternative ML methods, whether solely based on kinematic analysis or else on a combination of this with mainstream ML approaches. Then, we employ various interpretive methods to evaluate the network results, including attention map analysis and visual representation of Gradient-weighted Class Activation Mapping (Grad-CAM). The proposed network is generic and can be applied to analyse any process carrying information at different scales.

Mihoko Nojiri is a theorist in Theory Center, KEK, currently working on ML application in HEP. She is also one of the PIs of "Foundation of "Machine Learning Physics"(https://mlphys.scphys.kyoto-u.ac.jp/en/), a Grant-in-Aid for Transformative Research Aria(A), MEXT, Japan. She has various experiences of long-term collaboration with experimental groups, connecting the grant package with experimentalists. She is also involved in Diversity & Inclusion activities in Japan Physics Society and Science Council of Japan.

Coffee will be served at 10:30. .

0:00:00 Slide 1

0:03:03 Slide 2

0:04:33 Slide 3

0:06:00 Slide 4

0:07:26 Slide 5

0:08:40 Slide 6

0:10:29 Slide 7

0:11:44 Slide 8

0:14:17 Slide 9