Related references
Note: Only part of the references are listed.
Article
Physics, Particles & Fields
Simone Francescato et al.
Summary: The article introduces a multi-stage compression approach developed for implementing fast real-time inference for deep neural networks on the latest generation of hardware accelerators, with application in high energy physics use cases, and summarizes the effectiveness of the method.
EUROPEAN PHYSICAL JOURNAL C
(2021)
Correction
Physics, Particles & Fields
Simone Francescato et al.
EUROPEAN PHYSICAL JOURNAL C
(2021)
Article
Computer Science, Artificial Intelligence
Thea Aarrestad et al.
Summary: An automated tool is introduced for deploying ultra low-latency, low-power deep neural networks with convolutional layers on FPGAs. Through model compression techniques, significant reduction in FPGA critical resource consumption can be achieved with minimal to no loss in model accuracy.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2021)
Article
Computer Science, Information Systems
Yutaro Iiyama et al.
Summary: This paper discusses the design of distance-weighted graph networks that can be executed with a latency of less than one microsecond on an FPGA for crucial tasks in particle physics. By using a graph network architecture developed for specific tasks and applying additional simplifications to match the computing constraints of Level-1 trigger systems, including weight quantization.
FRONTIERS IN BIG DATA
(2021)
Article
Computer Science, Artificial Intelligence
Claudionor N. Coelho et al.
Summary: The paper discusses a quantization method for deep learning models that can reduce energy consumption and model size while maintaining high accuracy, suitable for efficient inference on edge devices.
NATURE MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Jennifer Ngadiuba et al.
Summary: The research presents the implementation of binary and ternary neural networks in the hls4ml library, which aims to automatically convert deep neural network models into digital circuits with FPGA firmware. By reducing the numerical precision of network parameters, the binary and ternary implementation achieves similar performance to higher precision implementations while using drastically fewer FPGA resources. The study discusses the trade-off between model accuracy, resource consumption, and the balance between latency and accuracy.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2021)
Article
Instruments & Instrumentation
J. Duarte et al.
JOURNAL OF INSTRUMENTATION
(2018)
