Cargando…
Accelerating Recurrent Neural Networks for Gravitational Wave Experiments
This paper presents novel reconfigurable architectures for reducing the latency of recurrent neural networks (RNNs) that are used for detecting gravitational waves. Gravitational interferometers such as the LIGO detectors capture cosmic events such as black hole mergers which happen at unknown times...
Autores principales: | , , , , , , , , , , , , |
---|---|
Lenguaje: | eng |
Publicado: |
2021
|
Materias: | |
Acceso en línea: | https://dx.doi.org/10.1109/ASAP52443.2021.00025 http://cds.cern.ch/record/2775808 |
_version_ | 1780971594442604544 |
---|---|
author | Que, Zhiqiang Wang, Erwei Marikar, Umar Moreno, Eric Ngadiuba, Jennifer Javed, Hamza Borzyszkowski, Bartłomiej Aarrestad, Thea Loncar, Vladimir Summers, Sioni Pierini, Maurizio Cheung, Peter Y. Luk, Wayne |
author_facet | Que, Zhiqiang Wang, Erwei Marikar, Umar Moreno, Eric Ngadiuba, Jennifer Javed, Hamza Borzyszkowski, Bartłomiej Aarrestad, Thea Loncar, Vladimir Summers, Sioni Pierini, Maurizio Cheung, Peter Y. Luk, Wayne |
author_sort | Que, Zhiqiang |
collection | CERN |
description | This paper presents novel reconfigurable architectures for reducing the latency of recurrent neural networks (RNNs) that are used for detecting gravitational waves. Gravitational interferometers such as the LIGO detectors capture cosmic events such as black hole mergers which happen at unknown times and of varying durations, producing time-series data. We have developed a new architecture capable of accelerating RNN inference for analyzing time-series data from LIGO detectors. This architecture is based on optimizing the initiation intervals (II) in a multi-layer LSTM (Long Short-Term Memory) network, by identifying appropriate reuse factors for each layer. A customizable template for this architecture has been designed, which enables the generation of low-latency FPGA designs with efficient resource utilization using high-level synthesis tools. The proposed approach has been evaluated based on two LSTM models, targeting a ZYNQ 7045 FPGA and a U250 FPGA. Experimental results show that with balanced II, the number of DSPs can be reduced up to 42% while achieving the same IIs. When compared to other FPGA-based LSTM designs, our design can achieve about 4.92 to 12.4 times lower latency. |
id | cern-2775808 |
institution | Organización Europea para la Investigación Nuclear |
language | eng |
publishDate | 2021 |
record_format | invenio |
spelling | cern-27758082023-01-31T08:12:37Zdoi:10.1109/ASAP52443.2021.00025http://cds.cern.ch/record/2775808engQue, ZhiqiangWang, ErweiMarikar, UmarMoreno, EricNgadiuba, JenniferJaved, HamzaBorzyszkowski, BartłomiejAarrestad, TheaLoncar, VladimirSummers, SioniPierini, MaurizioCheung, Peter Y.Luk, WayneAccelerating Recurrent Neural Networks for Gravitational Wave Experimentsphysics.ins-detDetectors and Experimental Techniquescs.ARComputing and Computerscs.LGComputing and ComputersThis paper presents novel reconfigurable architectures for reducing the latency of recurrent neural networks (RNNs) that are used for detecting gravitational waves. Gravitational interferometers such as the LIGO detectors capture cosmic events such as black hole mergers which happen at unknown times and of varying durations, producing time-series data. We have developed a new architecture capable of accelerating RNN inference for analyzing time-series data from LIGO detectors. This architecture is based on optimizing the initiation intervals (II) in a multi-layer LSTM (Long Short-Term Memory) network, by identifying appropriate reuse factors for each layer. A customizable template for this architecture has been designed, which enables the generation of low-latency FPGA designs with efficient resource utilization using high-level synthesis tools. The proposed approach has been evaluated based on two LSTM models, targeting a ZYNQ 7045 FPGA and a U250 FPGA. Experimental results show that with balanced II, the number of DSPs can be reduced up to 42% while achieving the same IIs. When compared to other FPGA-based LSTM designs, our design can achieve about 4.92 to 12.4 times lower latency.This paper presents novel reconfigurable architectures for reducing the latency of recurrent neural networks (RNNs) that are used for detecting gravitational waves. Gravitational interferometers such as the LIGO detectors capture cosmic events such as black hole mergers which happen at unknown times and of varying durations, producing time-series data. We have developed a new architecture capable of accelerating RNN inference for analyzing time-series data from LIGO detectors. This architecture is based on optimizing the initiation intervals (II) in a multi-layer LSTM (Long Short-Term Memory) network, by identifying appropriate reuse factors for each layer. A customizable template for this architecture has been designed, which enables the generation of low-latency FPGA designs with efficient resource utilization using high-level synthesis tools. The proposed approach has been evaluated based on two LSTM models, targeting a ZYNQ 7045 FPGA and a U250 FPGA. Experimental results show that with balanced II, the number of DSPs can be reduced up to 42% while achieving the same IIs. When compared to other FPGA-based LSTM designs, our design can achieve about 4.92 to 12.4 times lower latency.arXiv:2106.14089oai:cds.cern.ch:27758082021-06-26 |
spellingShingle | physics.ins-det Detectors and Experimental Techniques cs.AR Computing and Computers cs.LG Computing and Computers Que, Zhiqiang Wang, Erwei Marikar, Umar Moreno, Eric Ngadiuba, Jennifer Javed, Hamza Borzyszkowski, Bartłomiej Aarrestad, Thea Loncar, Vladimir Summers, Sioni Pierini, Maurizio Cheung, Peter Y. Luk, Wayne Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title | Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title_full | Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title_fullStr | Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title_full_unstemmed | Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title_short | Accelerating Recurrent Neural Networks for Gravitational Wave Experiments |
title_sort | accelerating recurrent neural networks for gravitational wave experiments |
topic | physics.ins-det Detectors and Experimental Techniques cs.AR Computing and Computers cs.LG Computing and Computers |
url | https://dx.doi.org/10.1109/ASAP52443.2021.00025 http://cds.cern.ch/record/2775808 |
work_keys_str_mv | AT quezhiqiang acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT wangerwei acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT marikarumar acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT morenoeric acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT ngadiubajennifer acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT javedhamza acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT borzyszkowskibartłomiej acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT aarrestadthea acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT loncarvladimir acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT summerssioni acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT pierinimaurizio acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT cheungpetery acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments AT lukwayne acceleratingrecurrentneuralnetworksforgravitationalwaveexperiments |