检索结果-内蒙古大学图书馆

Proceedings of the SC '24 Workshops of the International Conference on High Performance Computing, Network, Storage, and Analysis

作者： Teo Price-Broncucia Allison H. Baker Michael Duda Department of Computer Science University of Colorado Boulder CO Computational and Information Systems Lab NSF National Center for Atmospheric Research Boulder CO Mesoscale and Microscale Meteorology Lab NSF National Center for Atmospheric Research Boulder CO

ISBN: (纸本)9798350355543

In a series of related works developing an ensemble consistency testing approach for multiple popular global climate models (GCMs), one test scenario has repeatedly stood out. Why does the use of the Fused Multiply-Add (FMA) operation result in model configurations getting flagged as failures, while changes to compiler choice, optimization level, processor type and number, etc. are passed as expected? This work explores the impacts of FMA on GCM simulation output from a distributional perspective and provides directions for future work to enable model developers and users to use numerical optimization techniques with confidence.

关键词：

DACOOP-A: Decentralized Adaptive Cooperative Pursuit via Attention

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Zhang, Zheng Zhang, Dengyu Zhang, Qingrui Pan, Wei Hu, Tianjiang Lab. Sun Yat-Sen University Shenzhen China Department of Computer Science The University of Manchester United Kingdom Department of Cognitive Robotics Delft University of Technology Netherlands

Integrating rule-based policies into reinforcement learning promises to improve data efficiency and generalization in cooperative pursuit problems. However, most implementations do not properly distinguish the influence of neighboring robots in observation embedding or inter-robot interaction rules, leading to information loss and inefficient cooperation. This paper proposes a cooperative pursuit algorithm named Decentralized Adaptive COOperative Pursuit via Attention (DACOOP-A) by empowering reinforcement learning with artificial potential field and attention mechanisms. An attention-based framework is developed to emphasize important neighbors by concurrently integrating the learned attention scores into observation embedding and inter-robot interaction rules. A KL divergence regularization is introduced to alleviate the resultant learning stability issue. Improvements in data efficiency and generalization are demonstrated through numerical simulations. Extensive quantitative analysis and ablation studies are performed to illustrate the advantages of the proposed modules. Real-world experiments are performed to justify the feasibility of deploying DACOOP-A in physical systems. Copyright © 2023, The Authors. All rights reserved.

关键词： Reinforcement learning

Optical tomography-guided system for pre-clinical radiotherapy research

学校读者我要写书评

暂无评论

Optical tomography-guided system for pre-clinical radiothera...

2023 Digital Holography and Three-Dimensional Imaging, DH 2023 in Optica Imaging Congress - Part of Imaging and Applied Optics Congress 2023

作者： Wang, Ken Kang-Hsin Xu, Xiangkun Deng, Zijian Tong, Zhishen Tsai, Yupei Iordachita, Iulian Dehghani, Hamid Wong, John W. Department of Radiation Oncology University of Texas Southwestern Medical Center DallasTX75390 United States Laboratory for Computational Sensing and Robotics Johns Hopkins University BaltimoreMD21218 United States School of Computer Science University of Birmingham Edgbaston BirminghamB15 2TT United Kingdom Department of Radiation Oncology and Molecular Radiation Sciences Johns Hopkins University BaltimoreMD21287 United States

We advance the in vivo cancer imaging capability of small animal irradiators to include bioluminescence and fluorescence tomography. The imaging system is expected to enhance preclinical radiation guidance and assessm... 详细信息

ISBN: (纸本)9781957171289

关键词： Optical tomography

DQ-Whisper: Joint Distillation and Quantization for Efficient Multilingual Speech Recognition

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Shao, Hang Liu, Bei Wang, Wei Gong, Xun Qian, Yanmin Auditory Cognition and Computational Acoustics Lab MoE Key Lab of Artificial Intelligence AI Institute Department of Computer Science and Engineering Shanghai Jiao Tong University Shanghai China

As a popular multilingual and multitask pre-trained speech model, Whisper has the problem of curse of multilinguality. To enhance multilingual capabilities in small Whisper models, we propose DQ-Whisper, a novel joint distillation and quantization framework to compress Whisper for efficient inference. Firstly, we propose a novel dynamic matching distillation strategy. Then, a quantization-aware distillation framework is introduced to integrate quantization with distillation. Experimental results on various multilingual datasets show that our suggested distillation approach can effectively enhance the multilingual capabilities of small Whisper models without increasing computational costs. Up to 5.18x reduction in model size is achieved with marginal performance degradation. In addition, quantization is compatible with distillation, which can result in a higher compression rate. Copyright © 2023, The Authors. All rights reserved.

关键词： Speech recognition

Fetal Ultrasound Standard Plane Extraction using Orthogonal Triple-slice Deep Reinforcement Learning Agent

学校读者我要写书评

暂无评论

Fetal Ultrasound Standard Plane Extraction using Orthogonal ...

IEEE International Symposium on Applications of Ferroelectrics (ISAF)

作者： Baichuan Jiang Keshuai Xu Ernest Graham Russell H. Taylor Jeeun Kang Mathias Unberath Emad M. Boctor Department of Computer Science Johns Hopkins University Baltimore Maryland Department of Gynecology and Obstetrics Johns Hopkins Medical Institute Baltimore Maryland Department of Anesthesiology and Critical Care Medicine Johns Hopkins Medical Institute Baltimore Maryland Laboratory for Computational Sensing and Robotics Johns Hopkins University Baltimore Maryland

ISBN: (数字)9798350371901

ISBN: (纸本)9798350371918

Using ultrasound for fetal anatomical survey and fetal growth monitoring can be challenging and tedious as sonographers need to manually search for a set of standard planes (SPs) using a 2D ultrasound probe. A desirable alternative is using a 3D ultrasound device, either hand-held or wearable, to capture large field-of-view volumetric images and apply an image analysis algorithm to automatically extract the target SPs. Prior work has been conducted to formulate this problem as iteratively moving a 6-degree-of-freedom 2D resampling plane toward the target viewing pose. However, views with insufficient anatomical information can lead to incorrect actions thus poor results for SP *** this work, we propose to extend the 6-degree-of-freedom plane agent and leverage the two other resampling views orthogonal to the original plane to incorporate more context information for adaptive action prediction based on the most informative view. Experiments have been conducted on a preliminary clinical ultrasound dataset and the results show that with adaptive view selection, our algorithm can extract fetal biparietal diameter plane with an average plane localization error of 7.09 mm and 8.01 deg, comparing to the error of 25.35 mm and 35.15 deg when using a single view.

关键词： Surveys Location awareness Ultrasonic imaging Three-dimensional displays Image analysis Prediction algorithms Probes Standards Monitoring Frequency control

Effective Imaging Window Analysis for Wearable Ultrasound Device Using Fetal MRI

学校读者我要写书评

暂无评论

Effective Imaging Window Analysis for Wearable Ultrasound De...

IEEE International Symposium on Applications of Ferroelectrics (ISAF)

作者： Baichuan Jiang Keshuai Xu Yunpu Zhang Ernest Graham Russell H. Taylor Mathias Unberath Jeeun Kang Emad M. Boctor Department of Computer Science Johns Hopkins University Baltimore Maryland Department of Gynecology and Obstetrics Johns Hopkins Medical Institute Baltimore Maryland Department of Anesthesiology and Critical Care Medicine Johns Hopkins Medical Institute Baltimore Maryland Laboratory for Computational Sensing and Robotics Johns Hopkins University Baltimore Maryland

ISBN: (数字)9798350371901

ISBN: (纸本)9798350371918

Wearable ultrasound devices can play a key role and enable revolutionary workflows in fetal monitoring applications in both antepartum and intrapartum phase of pregnancy. However, the fundamental question for developing wearable fetal monitoring solutions is whether the sensor is capable of performing continuous monitoring while being placed at a fixed location with minimum instruction to the user. Therefore, in this work, we developed a validation workflow using fetal MRI images to virtually place the wearable ultrasound scanner and examine the feasible placement window given the scanner design. This allows any scanner design being validated and optimized even before production. We showed that a scanner with a cone-shaped field-of-view with 18cm imaging depth can successfully monitor both the key heart and brain features throughout the critical period of delivery for fetal hypoxic-ischemic encephalopathy detection.

关键词： Pregnancy Performance evaluation Heart Ultrasonic imaging Magnetic resonance imaging Production Feature extraction Biomedical monitoring Monitoring Frequency control

Fast multiplication of random dense matrices with fixed sparse matrices

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Liang, Tianyu Murray, Riley Buluç, Aydın Demmel, James Electrical Engineering and Computer Science Department UC Berkeley United States Computational Research Division Lawrence Berkeley National Lab United States International Computer Science Institute United States Sandia National Laboratories

This work focuses on accelerating the multiplication of a dense random matrix with a (fixed) sparse matrix, which is frequently used in sketching algorithms. We develop a novel scheme that takes advantage of blocking and recomputation (on-the-fly random number generation) to accelerate this operation. The techniques we propose decrease memory movement, thereby increasing the algorithm's parallel scalability in shared memory architectures. On the Intel Frontera architecture, our algorithm can achieve 2x speedups over libraries such as Eigen and Intel MKL on some examples. In addition, with 32 threads, we can obtain a parallel efficiency of up to 45%. We also present a theoretical analysis for the memory movement lower bound of our algorithm, showing that under mild assumptions, it's possible to beat the data movement lower bound of general matrix-matrix multiply (GEMM) by a factor of √M, where M is the cache size. Finally, we incorporate our sketching method into a randomized algorithm for overdetermined least squares with sparse data matrices. Our results are competitive with SuiteSparse for highly overdetermined problems;in some cases, we obtain a speedup of 10x over SuiteSparse. Copyright © 2023, The Authors. All rights reserved.

关键词： Random number generation

Boosting Earth System Model Outputs And Saving PetaBytes in their Storage Using Exascale Climate Emulators

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Abdulah, Sameh Baker, Allison H. Bosilca, George Cao, Qinglei Castruccio, Stefano Genton, Marc G. Keyes, David E. Khalid, Zubair Ltaief, Hatem Song, Yan Stenchikov, Georgiy L. Sun, Ying Extreme Computing & Statistics & Earth Science King Abdullah University of Science and Technology KSA Saudi Arabia Computational and Information Sciences Lab NSF National Center for Atmospheric Research United States NVIDIA United States Department of Computer Science Saint Louis University United States Department of Applied and Computational Mathematics and Statistics University of Notre Dame United States

We present the design and scalable implementation of an exascale climate emulator for addressing the escalating computational and storage requirements of high-resolution Earth System Model simulations. We utilize the spherical harmonic transform to stochastically model spatio-temporal variations in climate data. This provides tunable spatio-temporal resolution and significantly improves the fidelity and granularity of climate emulation, achieving an ultra-high spatial resolution of 0.034◦ (∼3.5 km) in space. Our emulator, trained on 318 billion hourly temperature data points from a 35-year and 31 billion daily data points from an 83-year global simulation ensemble, generates statistically consistent climate emulations. We extend linear solver software to mixed-precision arithmetic GPUs, applying different precisions within a single solver to adapt to different correlation strengths. The PaRSEC runtime system supports efficient parallel matrix operations by optimizing the dynamic balance between computation, communication, and memory requirements. Our BLAS3-rich code is optimized for systems equipped with four different families and generations of GPUs, scaling well to achieve 0.976 EFlop/s on 9,025 nodes (36,100 AMD MI250X multi-chip module (MCM) GPUs) of Frontier (nearly full system), 0.739 EFlop/s on 1,936 nodes (7,744 Grace-Hopper Superchips (GH200)) of Alps, 0.243 EFlop/s on 1,024 nodes (4,096 A100 GPUs) of Leonardo, and 0.375 EFlop/s on 3,072 nodes (18,432 V100 GPUs) of Summit. Copyright © 2024, The Authors. All rights reserved.

关键词： Stochastic systems

Design and Fabrication of a Fiber Bragg Grating Shape Sensor for Shape Reconstruction of a Continuum Manipulator

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Amirkhani, Golchehr Goodridge, Anna Esfandiari, Mojtaba Phalen, Henry Ma, Justin H. Iordachita, Iulian Armand, Mehran Department of Mechanical Engineering Johns Hopkins University BaltimoreMD21218 United States Laboratory for Computational Sensing and Robotics Johns Hopkins University BaltimoreMD21218 United States Department of orthopaedic Surgery The Department of Mechanical Engineering The Department of Computer Science

Continuum dexterous manipulators (CDMs) are suitable for performing tasks in a constrained environment due to their high dexterity and maneuverability. Despite the inherent advantages of CDMs in minimally invasive surgery, real-time control of CDMs’ shape during non-constant curvature bending is still challenging. This study presents a novel approach for the design and fabrication of a large deflection fiber Bragg grating (FBG) shape sensor embedded within the lumens inside the walls of a CDM with a large instrument channel. The shape sensor consisted of two fibers, each with three FBG nodes. A shape-sensing model was introduced to reconstruct the centerline of the CDM based on FBG wavelengths. Different experiments, including shape sensor tests and CDM shape reconstruction tests, were conducted to assess the overall accuracy of the shape sensing. The FBG sensor evaluation results revealed the linear curvature-wavelength relationship with the large curvature detection of 0.045 mm at a 90◦ bending angle and a sensitivity of up to 5.50 nm/mm in each bending direction. The CDM’s shape reconstruction experiments in a free environment demonstrated the shape tracking accuracy of 0.216±0.126 mm for positive/negative deflections. Also, the CDM shape reconstruction error for three cases of bending with obstacles were observed to be 0.436±0.370 mm for the proximal case, 0.485±0.418 mm for the middle case, and 0.312±0.261 mm for the distal case. This study indicates the adequate performance of the FBG sensor and the effectiveness of the model for tracking the shape of the large-deflection CDM with nonconstant-curvature bending for minimally-invasive orthopaedic applications. © 2023, CC BY.

关键词： Fiber Bragg gratings

A neuro-observer-based optimal control for nonaffine nonlinear systems with control input saturations

学校读者我要写书评

暂无评论

Control Theory and Technology 2021年第2期19卷 283-294页

作者： Behzad Farzanegan Mohsen Zamani Amir Abolfazl Suratgar Mohammad Bagher Menhaj Distributed Intelligent Optimization Research Lab Department of Electrical EngineeringAmirkabir University of TechnologyTehranIran Computational Intelligence Lab Department of Electrical EngineeringAmirkabir University of TechnologyTehranIran The School of Electrical Engineering and Computer Science The University of NewcastleNewcastleAustralia Department of Medical Physics and Engineering Shiraz University of Medical SciencesShirazIran

In this study,an adaptive neuro-observer-based optimal control(ANOPC)policy is introduced for unknown nonaffine nonlinear systems with control input ***–Jacobi–Bellman(HJB)framework is employed to minimize a non-quadratic cost function corresponding to the constrained control *** consists of both analytical and algebraic *** the analytical part,first,an observer-based neural network(NN)approximates uncertain system dynamics,and then another NN structure solves the HJB *** the algebraic part,the optimal control input that does not exceed the saturation bounds is *** weights of two NNs associated with observer and controller are simultaneously updated in an online *** ultimately uniformly boundedness(UUB)of all signals of the whole closed-loop system is ensured through Lyapunov’s direct ***,two numerical examples are provided to confirm the effectiveness of the proposed control strategy.

关键词： Input constraints Optimal control Neural networks Nonaffine nonlinear systems Reinforcement learning Unknown dynamics