Lightweight jet reconstruction and identification as an object detection task

Abstract We apply object detection techniques based on deep convolutional blocks to end-to-end jet identification and reconstruction tasks encountered at the CERN large hadron collider (LHC). Collision events produced LHC represented as an image composed of calorimeter tracker cells are given input a Single Shot Detection network. The algorithm, named PFJet-SSD performs simultaneous localization, classification regression cluster jets reconstruct their features. This all-in-one single feed-forward pass gives advantages in terms execution time improved accuracy w.r.t. traditional rule-based methods. A further gain is obtained from network slimming, homogeneous quantization, optimized runtime for meeting memory latency constraints typical real-time processing environment. experiment with 8-bit ternary benchmarking inference against single-precision floating-point. show that closely matches performance its full-precision equivalent outperforms state-of-the-art algorithm. Finally, we report different hardware platforms discuss future applications.