CERN Accelerating science

Preprint
Report number	arXiv:2103.05579 ; FERMILAB-CONF-21-080-SCD
Title	hls4ml: An Open-Source Codesign Workflow to Empower Scientific Low-Power Machine Learning Devices
Author(s)	Fahim, Farah (Northwestern U. ; Fermilab) ; Hawks, Benjamin (Fermilab) ; Herwig, Christian (Fermilab) ; Hirschauer, James (Fermilab) ; Jindariani, Sergo (Fermilab) ; Tran, Nhan (Fermilab) ; Carloni, Luca P. (Columbia U.) ; Di Guglielmo, Giuseppe (Columbia U.) ; Harris, Philip (MIT) ; Krupa, Jeffrey (MIT) ; Rankin, Dylan (MIT) ; Valentin, Manuel Blanco (Northwestern U.) ; Hester, Josiah (Northwestern U.) ; Luo, Yingyi (Northwestern U.) ; Mamish, John (Northwestern U.) ; Orgrenci-Memik, Seda (Northwestern U.) ; Aarrestad, Thea (CERN) ; Javed, Hamza (CERN) ; Loncar, Vladimir (CERN) ; Pierini, Maurizio (CERN) ; Pol, Adrian Alan (CERN) ; Summers, Sioni (CERN) ; Duarte, Javier (UC, San Diego) ; Hauck, Scott (Washington U., Seattle) ; Hsu, Shih-Chieh (Washington U., Seattle) ; Ngadiuba, Jennifer (Caltech) ; Liu, Mia (Purdue U.) ; Hoang, Duc (Rhodes Coll.) ; Kreinar, Edward (Unlisted, US, VA) ; Wu, Zhenbin (Illinois U., Chicago)
Imprint	2021-03-09
Number of pages	10
Note	10 pages, 8 figures, TinyML Research Symposium 2021
Subject category	physics.ins-det ; Detectors and Experimental Techniques ; cs.AR ; Computing and Computers ; cs.LG ; Computing and Computers
Abstract	Accessible machine learning algorithms, software, and diagnostic tools for energy-efficient devices and systems are extremely valuable across a broad range of application domains. In scientific domains, real-time near-sensor processing can drastically improve experimental design and accelerate scientific discoveries. To support domain scientists, we have developed hls4ml, an open-source software-hardware codesign workflow to interpret and translate machine learning algorithms for implementation with both FPGA and ASIC technologies. We expand on previous hls4ml work by extending capabilities and techniques towards low-power implementations and increased usability: new Python APIs, quantization-aware pruning, end-to-end FPGA workflows, long pipeline kernels for low power, and new device backends include an ASIC workflow. Taken together, these and continued efforts in hls4ml will arm a new generation of domain scientists with accessible, efficient, and powerful tools for machine-learning-accelerated discovery.
Other source	Inspire
Copyright/License	preprint: (License: CC BY-SA 4.0)