检索结果-内蒙古大学图书馆

7th Chinese Conference on Pattern Recognition and Computer Vision

作者： Dong, Yiming Lin, Zhouchen Peking Univ Sch Intelligence Sci & Technol State Key Lab Gen Artificial Intelligence Beijing Peoples R China Pazhou Lab Huangpu Guangzhou Guangdong Peoples R China

ISBN: (纸本)9789819784899;9789819784905

As deep learning continues to spearhead transformative breakthroughs across various domains, the computational and memory demands for training state-of-the-art models have surged exponentially. This escalation not only challenges the scalability of deep learning systems but also significantly increases the financial cost associated with training. Memory-intensive operations, particularly during the optimization phase of training large models, can drastically inflate budgets, making cutting-edge research and applications less accessible. In response to this challenge, we introduce a novel technique termed gradient space reutilization, aiming at reducing memory usage in deep network training by repurposing the memory allocated for gradients once it is no longer needed in the later computing process. This approach is implemented across modified versions of popular optimizers, with their names AdamW-R, Adan-R, and Lion-R, respectively, demonstrating appreciable memory savings without compromising the performance as they are equivalent to the original algorithms. Extensive experiments demonstrate that our simple engineering trick can achieve up to 25.60% memory savings at the best, providing a practical solution for efficient resource management in deep learning training environments.

关键词： deep learning optimization Efficient deep learning Memory footprint

来源：评论

学校读者我要写书评

暂无评论

Optimizing YOLOv8 for Real-Time Performance in Humanoid Soccer Robots with OpenVINO 26

Optimizing YOLOv8 for Real-Time Performance in Humanoid Socc...

引用

26th International Electronics Symposium (IES) on Shaping the Future - Society 5.0 and Beyond

作者： Pradana, Erlangga Yudi Aji, Shalahuddin Aditya Abdulrrozaq, Muhammad Amir Alasiry, Ali Husein Risnumawan, Anhar Pitowarno, Endra Politekn Elekt Negeri Surabaya Elect Engn Div Surabaya Indonesia Politekn Elekt Negeri Surabaya Mechatron Engn Div Surabaya Indonesia Politekn Elekt Negeri Surabaya Informat Engn Div Surabaya Indonesia

ISBN: (纸本)9798350392005;9798350391992

Humanoid soccer robots require fast and accurate vision systems for effective real-time decision-making. YOLOv8 (You Only Look Once) is a leading deep learning-based object detection method known for its balance of speed and accuracy. This research explores optimizing YOLOv8 performance using OpenVINO (Open Visual Inference & Neural Network optimization), a toolkit designed to accelerate and enhance deep learning models on Intel hardware. We assessed performance improvements in terms of frames per second (FPS) and accuracy using an NUC1017FNH mini PC. Our experiments utilized a dataset of over 12,000 images categorized into five classes: orange ball, goalpost, L line, T line, and X line. The results show that OpenVINO effectively doubles YOLOv8's inference speed, reducing the average processing time from 36 ms to 16 ms and increasing FPS from 25 to 32, while maintaining detection accuracy. This optimization significantly boosts the speed and responsiveness of vision systems in humanoid robots, ensuring robust performance in real-time scenarios.

关键词： humanoid soccer robots YOLOv8 OpenVINO object detection inference speed deep learning optimization

来源：评论

学校读者我要写书评

暂无评论

Jigsaw: Accelerating SpMM with Vector Sparsity on Sparse Tensor Core 24

Jigsaw: Accelerating SpMM with Vector Sparsity on Sparse Ten...

引用

53rd International Conference on Parallel Processing (ICPP)

作者： Zhang, Kaige Liu, Xiaoyan Yang, Hailong Feng, Tianyu Yang, Xinyu Liu, Yi Luan, Zhongzhi Qian, Depei Beihang Univ Sch Comp Sci & Engn Beijing Peoples R China

ISBN: (纸本)9798400717932

As deep learning models continue to grow larger, model pruning is employed to reduce memory footprint and computation complexity, which generates a large number of sparse matrix-matrix multiplication (SpMM) with unstructured sparsity (e.g., vector sparsity). However, leveraging GPU especially the newly integrated sparse tensor core (SpTC) to accelerate SpMM is quite challenging due to the unstructured sparsity. Unfortunately, existing works fail to fully exploit the SpTC on GPU due to the difficulty of satisfying the stringent requirement for restricted sparsity (e.g., 2:4 sparsity). In this paper, we propose Jigsaw, a novel method to utilize SpTC for accelerating SpMM with vector sparsity. Specifically, we propose the multi-granularity sparsity reorder method to transform the sparse data for satisfying the sparse pattern supported on SpTC. In addition, we propose a reorder-aware storage format for the transformed sparse data to better adapt to the parallelism of SpTC. Moreover, we propose corresponding optimizations to better exploit the SpTC for further accelerating SpMM. The experiment results demonstrate that Jigsaw outperforms state-of-the-art SpMM implementations and achieves promising speedup over cuBLAS.

关键词： sparse matrix reordering sparse tensor core deep learning optimization sparse matrix-matrix multiplication

来源：评论

学校读者我要写书评

暂无评论

Efficient Sparse Tensor Core Networks for Real-Time Insect Classification in Agriculture 4th

Efficient Sparse Tensor Core Networks for Real-Time Insect C...

引用

4th International Conference on Advanced Network Technologies and Intelligent Computing-ANTIC-Annual

作者： Rao, P. Kiran Prakash, P. Suman Kumar, N. Jaswanth Reddy, V. Kartheek Kumar, Arigela Satheesh Rajeev Gandhi Mem Coll Engn & Technol Dept CSE Data Sci Nandyal Andra Pradesh India G Pullaiah Coll Engn & Technol Autonomus Dept CSE Data Sci Kurnool Andra Pradesh India

ISBN: (纸本)9783031837890;9783031837906

deep learning has become a powerful tool for various image-based applications, including the detection of agricultural pests. However, the deployment of deep neural networks in resource-constrained agricultural settings poses significant challenges due to their high computational demands. Traditional pruning methods designed to mitigate this challenge often result in sparse networks that may not perform efficiently on hardware optimized for dense operations. In this study, we present a novel approach called Efficient Sparse Tensor Core Networks (STCN), which leverages the inherent sparsity of neural networks to enhance computational efficiency significantly. Our method integrates a specialized sparse network architecture with hardware-accelerated tensor operations, optimized for real-time analysis on edge devices. We have tailored our model to exploit sparse tensor operations, facilitating faster computations and reduced energy consumption without compromising accuracy. When applied to the IP102 dataset, a comprehensive collection of insect pest images, the STCN method proves effective in classifying various pest species under different environmental conditions. Initial experiments show that our approach not only maintains a high classification accuracy of 92% but also achieves a substantial reduction in computational workload, reducing energy consumption by 30% and improving processing speed by 40% during inference. These findings highlight the potential of STCN for real-time pest detection, paving the way for the practical application of advanced AI in precision agriculture, optimizing resource utilization, and promoting sustainable farming practices.

关键词： Sparse Tensor Core Networks (STCN) Real-time Insect Pest Classification Agricultural AI Computational Efficiency Edge Computing deep learning optimization

来源：评论

学校读者我要写书评

暂无评论

Optimizing GNN Inference Processing on Very Long Vector Processor 1

引用

23rd International Conference on Algorithms and Architectures for Parallel Processing (ICA3PP)

作者： Chen, Kangkang Su, Huayou Liu, Chaorun Li, Yalin Natl Univ Def & Technol Changsha Hunan Peoples R China

ISBN: (数字)9789819708116

ISBN: (纸本)9789819708109;9789819708116

Graph Neural Network (GNN) has shown great success in graph learning. However, within the complexity of the real-world tasks and the big graph datasets, current GNN models become increasingly bigger and more complicated to enhance learning ability and prediction accuracy, which poses a huge challenge to the computation of GNN. The accelerated optimization work for GNN mainly focuses on typical architecture, such as GPU and CPU, with relatively little research on other architectures. In this paper, we focus on accelerating the GNN inference on a very long vector processor. There are several problems in deploying GNN models effectively on long vector architectur, including the lack of an efficient library for GNN on long vector architecture, the very low memory access bandwidth of DDR, and the heterogeneous scheduling issues. To address these challenges, we proposed several strategies to accelerate GNN inference on long vector architecture. Specifically, we build an efficient GNN operator library targeting on the long vector architecture. Secondly, we designed an operator fusion strategy to improve on-chip memory utilization and alleviate the pressure of off-chip memory access. Finally, we implemented a heterogeneous multi-threaded scheduling strategy for long vector architecture to eliminate the overhead of kernel launch. The experimental results show that compared to DGL on Phytium 2000+, the proposed GNN inference on a very long vector processor achieve up to 6.47x speedups.

关键词： GNN Long Vector Architecture Accelerated Inference deep learning optimization

来源：评论

学校读者我要写书评

暂无评论

Adaptive FSP: Adaptive Architecture Search with Filter Shape Pruning 16th

Adaptive FSP: Adaptive Architecture Search with Filter Shape...

引用

16th Asian Conference on Computer Vision (ACCV)

作者： Kim, Aeri Lee, Seungju Kwon, Eunji Kang, Seokhyeong Pohang Univ Sci & Technol Dept Elect Engn Pohang South Korea

ISBN: (纸本)9783031263187;9783031263194

deep Convolutional Neural Networks (CNNs) have high memory footprint and computing power requirements, making their deployment in embedded devices difficult. Network pruning has received attention in reducing those requirements of CNNs. Among the pruning methods, Stripe-Wise Pruning (SWP) achieved a further network compression than conventional filter pruning methods and can obtain optimal kernel shapes of filters. However, the model pruned by SWP has filter redundancy because some filters have the same kernel shape. In this paper, we propose the Filter Shape Pruning (FSP) method, which prunes the networks using the kernel shape while maintaining the receptive fields. To obtain an architecture that satisfies the target FLOPs with the FSP method, we propose the Adaptive Architecture Search (AAS) framework. The AAS framework adaptively searches for the architecture that satisfies the target FLOPs with the layer-wise threshold. The layer-wise threshold is calculated at each iteration using the metric that reflects the filters influence on accuracy and FLOPs together. Comprehensive experimental results demonstrate that the FSP can achieve a higher compression ratio with an acceptable reduction in accuracy.

关键词： deep learning optimization Structured pruning Convolution neural networks

来源：评论

学校读者我要写书评

暂无评论

An integrated machine learning and metaheuristic approach for advanced packed bed latent heat storage system design and optimization

引用

ENERGY 2024年 297卷

作者： Anagnostopoulos, Argyrios Xenitopoulos, Theofilos Ding, Yulong Seferlis, Panos Aristotle Univ Thessaloniki Dept Mech Engn POB 454 Thessaloniki 54124 Greece Univ Silesia Inst Chem PL-40006 Katowice Poland Univ Birmingham Birmingham Ctr Energy Storage Birmingham B15 2TT England Univ Birmingham Sch Chem Engn Birmingham B15 2TT England

To tackle the challenge of waste heat recovery in the industrial sector, this research presents a novel design and optimization framework for Packed Bed Latent Heat Storage Systems (PBLHS). This features a deep learning (DL) model, integrated with metaheuristic algorithms. The DL model was developed to predict PBLHS performance, trained using data generated from a validated Computational Fluid Dynamics (CFD) model. The model exhibited a high performance with an R(2 )value of 0.975 and a low Mean Absolute Percentage Error (<9.14%). To enhance the ML model's efficiency and optimized performance, various metaheuristic algorithms were explored. The Harmony Search algorithm emerged as the most effective through an early screening and underwent further refinement. The optimized algorithm demonstrated its capability by rapidly producing designs that showcased an improvement in total efficiency of up to 85% over available optimized experimental PBLHS designs. This research underscores the potential of ML-integrated approaches in laying the groundwork for generalized design frameworks for TES systems, offering efficient and effective solutions for waste heat recovery.

关键词： Packed bed latent heat storage deep learning optimization Metaheuristic algorithms Thermal energy storage design Phase change materials

来源：评论

学校读者我要写书评

暂无评论

End-to-End Compressive Spectral Classification: A deep learning Approach Applied to the Grading of Tahiti Lime 2nd

End-to-End Compressive Spectral Classification: A Deep Learn...

引用

2nd International Conference on Smart Technologies, Systems and Applications (SmartTech-IC)

作者： Silva-Maldonado, Mauricio Galvis, Laura Arguello, Henry Univ Ind Santander Dept Comp Sci Bucaramanga Colombia

ISBN: (纸本)9783030991708;9783030991692

Compressed sensing (CS) theory enables the reconstruction of spectral images (SI) using a lower number of measurements than the traditional Shannon-Nyquist sampling approach, through compressive spectral imaging (CSI) systems. These CSI systems rely on a dispersive-based optical setup coupled to one or more coded-apertures to capture and compress a spectral scene simultaneously. Afterward, the reconstruction of the underlying scene is obtained through computational algorithms. Then, processing tasks like classification, object detection, or segmentation are performed over the reconstructed images. However, this reconstruction process is computationally expensive, which introduces a time overhead for these tasks. In this paper, spectral classification is directly performed over compressed measurements acquired through an optical architecture following the CS framework. An end-to-end method to optimize both coded-apertures and deep learning model parameters is proposed. This approach has been applied to the grading of Tahiti lime (Citrus latifolia), but can be used for different agricultural materials. In this specific case, the classification accuracy reached 99%. In addition, for the purpose of comparison, our experiments improved up to 7% in classification accuracy over a testing database when the coded-apertures were optimized.

关键词： Compressed sensing deep learning optimization Spectral images Spectral classification Tahiti lime

来源：评论

学校读者我要写书评

暂无评论

Improving Inference Time of deep learning Model with Partial Skip of ReLU-fused Matrix Multiplication Operations

Improving Inference Time of Deep Learning Model with Partial...

引用

International Conference on Electronics, Information, and Communication (ICEIC)

作者： Kim, Sungkyun Kim, Jaemin Kim, Nahun Kang, Mincheal Seo, Jiwon Hanyang Univ Dept Comp Sci Seoul South Korea Hanyang Univ Dept Artificial Intelligence Seoul South Korea McKinsey & Co Inc Tokyo Japan

ISBN: (纸本)9781665409346

deep learning has been expanding its application, while large-scale models tend to perform well. However, as such a model inevitably requires a vast amount of resources and computations, lengthy inference time is a crucial, but essential, consequence that needs to be optimized for the efficient utilization of deep learning. To achieve the goal, we aim at fusing the Rectified Linear Unit and matrix multiplication in the inference process, which we may reduce the total amount of computation by predicting the sign bit of output value. We propose four methods of prediction and statistically choose an optimal method for reducing inference time with low accuracy loss.

关键词： deep learning optimization omitted computation fully-connected layer inference optimization

来源：评论

学校读者我要写书评

暂无评论

Demystify Hyperparameters for Stochastic optimization with Transferable Representations 22

Demystify Hyperparameters for Stochastic Optimization with T...

引用

28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KKD)

作者： Sun, Jianhui Huai, Mengdi Jha, Kishlay Zhang, Aidong Univ Virginia Charlottesville VA 22903 USA

ISBN: (纸本)9781450393850

This paper studies the convergence and generalization of a large class of Stochastic Gradient Descent (SGD) momentum schemes, in both learning from scratch and transferring representations with fine-tuning. Momentum-based acceleration of SGD is the default optimizer for many deep learning models. However, there is a lack of general convergence guarantees for many existing momentum variants in conjunction with stochastic gradient. It is also unclear how the momentum methods may affect the generalization error. In this paper, we give a unified analysis of several popular optimizers, e.g., Polyak's heavy ball momentum and Nesterov's accelerated gradient. Our contribution is threefold. First, we give a unified convergence guarantee for a large class of momentum variants in the stochastic setting. Notably, our results cover both convex and nonconvex objectives. Second, we prove a generalization bound for neural networks trained by momentum variants. We analyze how hyperparameters affect the generalization bound and consequently propose guidelines on how to tune these hyperparameters in various momentum schemes to generalize well. We provide extensive empirical evidence to our proposed guidelines. Third, this study fills the vacancy of a formal analysis of fine-tuning in literature. To our best knowledge, our work is the first systematic generalizability analysis on momentum methods that cover both learning from scratch and fine-tuning. Our codes are available (1).

关键词： deep learning optimization Fine-tuning Stochastic Gradient Descent AutoML Generalization Bound

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：