Multi-GPU Development of a Neural Networks Based Reconstructor for Adaptive Optics

Aberrations introduced by the atmospheric turbulence in large telescopes are compensated using adaptive optics systems, where the use of deformable mirrors and multiple sensors relies on complex control systems. Recently, the development of larger scales of telescopes as the E-ELT or TMT has created a computational challenge due to the increasing complexity of the new adaptive optics systems. The Complex Atmospheric Reconstructor based on Machine Learning (CARMEN) is an algorithm based on artificial neural networks, designed to compensate the atmospheric turbulence. During recent years, the use of GPUs has been proved to be a great solution to speed up the learning process of neural networks, and different frameworks have been created to ease their development. The implementation of CARMEN in different Multi-GPU frameworks is presented in this paper, along with its development in a language originally developed for GPU, like CUDA. This implementation offers the best response for all the presented cases, although its advantage of using more than one GPU occurs only in large networks.

[1]  Dumitru Erhan,et al.  Going deeper with convolutions , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[2]  Geoffrey E. Hinton,et al.  Learning representations by back-propagating errors , 1986, Nature.

[3]  G. Rousset,et al.  Open-loop tomography with artificial neural networks on CANARY: on-sky results , 2014, 1405.6862.

[4]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[5]  David Romero-Laorden,et al.  Analysis of Parallel Computing Strategies to Accelerate Ultrasound Imaging Processes , 2016, IEEE Transactions on Parallel and Distributed Systems.

[6]  Alex Krizhevsky,et al.  One weird trick for parallelizing convolutional neural networks , 2014, ArXiv.

[7]  Mohak Shah,et al.  Comparative Study of Deep Learning Software Frameworks , 2015, 1511.06435.

[8]  Eric Gendron,et al.  Adaptive Optics Simulation for the World's Largest Telescope on Multicore Architectures with Multiple GPUs , 2016, PASC.

[9]  Geoffrey E. Hinton,et al.  Speech recognition with deep recurrent neural networks , 2013, 2013 IEEE International Conference on Acoustics, Speech and Signal Processing.

[10]  Thomas Paine,et al.  GPU Asynchronous Stochastic Gradient Descent to Speed Up Neural Network Training , 2013, ICLR.

[11]  L M Mugnier,et al.  Optimal wave-front reconstruction strategies for multiconjugate adaptive optics. , 2001, Journal of the Optical Society of America. A, Optics, image science, and vision.

[12]  David Romero-Laorden,et al.  Análisis de la implementación software de un conformador de señales ultrasónicas para tiempo real , 2016 .

[13]  Francisco Javier de Cos Juez,et al.  Analysing the Performance of a Tomographic Reconstructor with Different Neural Networks Frameworks , 2016, ISDA.

[14]  Marc'Aurelio Ranzato,et al.  Multi-GPU Training of ConvNets , 2013, ICLR.

[15]  Stephen Rolt,et al.  DRAGON, the Durham real-time, tomographic adaptive optics test bench: progress and results , 2014, Astronomical Telescopes and Instrumentation.

[16]  Richard H. Myers,et al.  Modeling a MEMS deformable mirror using non-parametric estimation techniques. , 2010, Optics express.

[17]  Jerry Nelson,et al.  The status of the Thirty Meter Telescope project , 2008, Astronomical Telescopes + Instrumentation.

[18]  R. Shack,et al.  History and principles of Shack-Hartmann wavefront sensing. , 2001, Journal of refractive surgery.

[19]  Hiroshi Terada,et al.  Multi-object adaptive optics on-sky results with Raven , 2014, Astronomical Telescopes and Instrumentation.

[20]  Fernando Rosa,et al.  Atmospheric wavefront phase recovery by use of specialized hardware: graphical processing units and field-programmable gate arrays. , 2005, Applied optics.

[21]  N. Hubin,et al.  The E-ELT instrument roadmap: a status report , 2014, Astronomical Telescopes and Instrumentation.

[22]  Francisco Javier de Cos Juez,et al.  Deformable mirror model for open-loop adaptive optics using multivariate adaptive regression splines. , 2010, Optics express.

[23]  Eric Gendron,et al.  CANARY: the on-sky NGS/LGS MOAO demonstrator for EAGLE , 2008, Astronomical Telescopes + Instrumentation.

[24]  J. Beckers.  ADAPTIVE OPTICS FOR ASTRONOMY: Principles, Performance, and Applications , 1993 .

[25]  Francisco Javier de Cos Juez,et al.  Analysis of the Temporal Structure Evolution of Physical Systems with the Self-Organising Tree Algorithm (SOTA): Application for Validating Neural Network Systems on Adaptive Optics Data before On-Sky Implementation , 2017, Entropy.

[26]  Francisco Javier de Cos Juez,et al.  Comparative Study of Neural Network Frameworks for the Next Generation of Adaptive Optics Systems , 2017, Sensors.

[27]  Francisco Javier de Cos Juez,et al.  An ANN-Based Smart Tomographic Reconstructor in a Dynamic Environment , 2012, Sensors.

[28]  Michael C. Roggemann,et al.  Optical performance of fully and partially compensated adaptive optics systems using least-squares and minimum variance phase reconstructors , 1992 .

[29]  Colin Bradley,et al.  Performance Modeling for the RAVEN Multi-Object Adaptive Optics Demonstrator , 2012 .

[30]  A. Guesalaga,et al.  First on-sky results of a neural network based tomographic reconstructor: Carmen on Canary , 2014, Astronomical Telescopes and Instrumentation.

[31]  Rong Zheng,et al.  Asynchronous stochastic gradient descent for DNN training , 2013, 2013 IEEE International Conference on Acoustics, Speech and Signal Processing.

[32]  G. Rousset,et al.  Tomography approach for multi-object adaptive optics. , 2010, Journal of the Optical Society of America. A, Optics, image science, and vision.

[33]  Alex Graves,et al.  Generating Sequences With Recurrent Neural Networks , 2013, ArXiv.

[34]  Gianni Pasolini,et al.  Simulation platform for , 2006 .

[35]  T. Fusco,et al.  Experience with wavefront sensor and deformable mirror interfaces for wide-field adaptive optics systems , 2016, 1603.07527.

[36]  B. Ellerbroek.  First-order performance evaluation of adaptive optics systems for atmospheric turbulence compensatio , 1994 .

[37]  A. Sevin,et al.  MOAO first on-sky demonstration with CANARY , 2011 .

[38]  Dani Guzman,et al.  Using artificial neural networks for open-loop tomography. , 2011, Optics express.

[39]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[40]  Qiang Wang,et al.  Benchmarking State-of-the-Art Deep Learning Software Tools , 2016, 2016 7th International Conference on Cloud Computing and Big Data (CCBD).

[41]  Trevor Darrell,et al.  Caffe: Convolutional Architecture for Fast Feature Embedding , 2014, ACM Multimedia.

[42]  Francisco Javier de Cos Juez,et al.  Successful sulfur recovery in low sulfurate compounds obtained from the zinc industry: Evaporation-condensation method. , 2017, Journal of hazardous materials.

[43]  Aniruddha Parvat,et al.  A survey of deep-learning frameworks , 2017, 2017 International Conference on Inventive Systems and Control (ICISC).

[44]  José Luís Calvo-Rolle,et al.  Using GPUs to Speed up a Tomographic Reconstructor Based on Machine Learning , 2016, SOCO-CISIS-ICEUTE.

[45]  Xiang Zhang,et al.  OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks , 2013, ICLR.