detector data is particularly well suited for representation in the … · 2019. 7. 17. ·...

12
1

Upload: others

Post on 05-Feb-2021

0 views

Category:

Documents


0 download

TRANSCRIPT

  • 1

  • 2

  • • Detector data is particularly well suited for representation in the form of images

    • e.g. calorimeter trigger information often transmitted as pixelated energy

    • Convolutional neural networks provide a well-studied and appropriate machine learning (ML) approach to classifying images of this form

    3arXiv:1707.08600

    https://arxiv.org/abs/1707.08600

  • ••

    • DNNDK – Deep Neural Network Development Kit

    • hls4ml – High Level Synthesis for ML

    4

    Testing model

    Extremely preliminary

    https://www.xilinx.com/products/design-tools/ai-inference/ai-developer-hub.html#edgehttps://hls-fpga-machine-learning.github.io/hls4ml/

  • 5https://www.xilinx.com/support/documentation/user_guides/ug1327-dnndk-user-guide.pdf

    https://www.xilinx.com/support/documentation/user_guides/ug1327-dnndk-user-guide.pdf

  • 6https://www.xilinx.com/support/documentation/ip_documentation/dpu/v1_2/pg338-dpu.pdf

    https://www.xilinx.com/support/documentation/ip_documentation/dpu/v1_2/pg338-dpu.pdf

  • 7

    ResNet50 image classification example – much bigger and more

    complicated than our network

  • 8

    ••

    ••

  • 9https://hls-fpga-machine-learning.github.io/hls4ml/CONCEPTS.html

    https://hls-fpga-machine-learning.github.io/hls4ml/CONCEPTS.html

  • 10Image and work by Jack Huang, UChicago Undergrad

  • 11

  • 12