Parallel-in-time algorithms provide an additional layer of concurrency for the numerical integration of models based on time-dependent differential equations. Methods like Parareal, which parallelize across multiple t...
详细信息
ISBN:
(纸本)9783031396977;9783031396984
Parallel-in-time algorithms provide an additional layer of concurrency for the numerical integration of models based on time-dependent differential equations. Methods like Parareal, which parallelize across multiple time steps, rely on a computationally cheap and coarse integrator to propagate information forward in time, while a parallelizable expensive fine propagator provides accuracy. Typically, the coarse method is a numerical integrator using lower resolution, reduced order or a simplified model. Our paper proposes to use a physics-informed neuralnetwork (PINN) instead. We demonstrate for the Black-Scholes equation, a partial differential equation from computational finance, that Parareal with a PINN coarse propagator provides better speedup than a numerical coarse propagator. Training and evaluating a neuralnetwork are both tasks whose computing patterns are well suited for GPUs. By contrast, mesh-based algorithms with their low computational intensity struggle to perform well. We show that moving the coarse propagator PINN to a GPU while running the numerical fine propagator on the CPU further improves Parareal's single-node performance. This suggests that integrating machine learning techniques into parallel-in-time integration methods and exploiting their differences in computing patterns might offer a way to better utilize heterogeneous architectures.
Efficient resource allocation among slices/users with different Service Level Agreements (SLAs) is a critical task in 5G+ networks, which has prompted recent research into Deep neuralnetworks (DNNs). However, challen...
详细信息
Brand logo image examines how a critical dimension of logo design, namely the naturalness of the logo color, influences brand design induced by the logo. However, high complexity is observed, requiring higher quality ...
详细信息
Deep neuralnetwork (DNN) inference poses unique challenges in serving computational requests due to high request intensity, concurrent multi-user scenarios, and diverse heterogeneous service types. Simultaneously, mo...
详细信息
ISBN:
(纸本)9798400708435
Deep neuralnetwork (DNN) inference poses unique challenges in serving computational requests due to high request intensity, concurrent multi-user scenarios, and diverse heterogeneous service types. Simultaneously, mobile and edge devices provide users with enhanced computational capabilities, enabling them to utilize local resources for deep inference processing. Moreover, dynamic inference techniques allow content-based computational cost selection per request. This paper presents Dystri, an innovative framework devised to facilitate dynamic inference on distributed edge infrastructure, thereby accommodating multiple heterogeneous users. Dystri offers a broad applicability in practical environments, en-compassing heterogeneous device types, DNN-based applications, and dynamic inference techniques, surpassing the state-of-the-art (SOTA) approaches. With distributed controllers and a global coordinator, Dystri allows per-request, per-user adjustments of qualityof-service, ensuring instantaneous, flexible, and discrete control. The decoupled workflows in Dystri naturally support user heterogeneity and scalability, addressing crucial aspects overlooked by existing SOTA works. Our evaluation involves three multi-user, heterogeneous DNN inference service platforms deployed on distributed edge infrastructure, encompassing seven DNN applications. Results show Dystri achieves near-zero deadline misses and excels in adapting to varying user numbers and request intensities. Dystri outperforms baselines with accuracy improvement up to 95x.
Object detection is a fundamental component of computer vision, playing a pivotal role in various applications. However, the accurate detection of small and densely distributed objects remains a challenging problem in...
详细信息
ISBN:
(纸本)9798350359329;9798350359312
Object detection is a fundamental component of computer vision, playing a pivotal role in various applications. However, the accurate detection of small and densely distributed objects remains a challenging problem in this field. This challenge is particularly exacerbated in the context of aerial imagery, characterized by its distinctive bird's-eye view, intricate backgrounds, and the variability in object appearances. This paper addresses these persistent challenges in object detection, with a focus on the specific difficulties posed by aerial images. We propose a deformable end-to-end object detection with transformers (DETR)-based framework to enhance small object detection accuracy, ultimately contributing to improved computer vision capabilities in domains like remote sensing, surveillance, and autonomous aerial systems. Firstly, in order to aggregate the entire input sequence information in the backbone network and improve the detection accuracy of small objects, we propose DMCA based on deformable features and attention mechanisms. Secondly, in order to capture and model the relationships between samples for dense pixel-level representations in small objects and improve the detection accuracy of small objects, we try to introduce BatchEncoder by implementing an encoder in the batch dimension. Experimental results show that, compared to the baseline, our method significantly improves the accuracy of small object detection in aerial images. The processing and analysis of a large amount of remote sensing data require powerful computing power. The computing power Internet can integrate the scattered computing resources to provide flexible and scalable computing power to meet the computing needs of different sizes and types.
Learning modality-fused representations and processing unaligned multimodal sequences are meaningful and challenging in multimodal emotion *** approaches use directional pairwise attention or a message hub to fuse lan...
详细信息
Learning modality-fused representations and processing unaligned multimodal sequences are meaningful and challenging in multimodal emotion *** approaches use directional pairwise attention or a message hub to fuse language,visual,and audio ***,these fusion methods are often quadratic in complexity with respect to the modal sequence length,bring redundant information and are not *** this paper,we propose an efficient neuralnetwork to learn modality-fused representations with CB-Transformer(LMR-CBT)for multimodal emotion recognition from unaligned multi-modal ***,we first perform feature extraction for the three modalities respectively to obtain the local structure of the ***,we design an innovative asymmetric transformer with cross-modal blocks(CB-Transformer)that enables complementary learning of different modalities,mainly divided into local temporal learning,cross-modal feature fusion and global self-attention *** addition,we splice the fused features with the original features to classify the emotions of the ***,we conduct word-aligned and unaligned experiments on three challenging datasets,IEMOCAP,CMU-MOSI,and *** experimental results show the superiority and efficiency of our proposed method in both *** with the mainstream methods,our approach reaches the state-of-the-art with a minimum number of parameters.
In this paper, an architecture of a photonic neuralnetwork chip with distributed multi-scenarios is presented. The distributed computing chip has five computing modules, which are characterized by three-dimensional i...
详细信息
The Internet of Things (IoT) requires sophisticated security measures because of heterogeneity and resource constraints. Current approaches in Anomaly Detection (AD) do not meet both challenges. Device-specific AD mod...
详细信息
To address the communication congestion and high energy consumption issues in IoT networks, prior research has proposed a distributed information processing system called WiBIC. As an exploratory step towards implemen...
详细信息
ISBN:
(纸本)9798350327038
To address the communication congestion and high energy consumption issues in IoT networks, prior research has proposed a distributed information processing system called WiBIC. As an exploratory step towards implementing the basic functionalities of the WiBIC system, this study focuses on the wireless integration of SNNs. Considering the data transmission and learning characteristics of SNNs, we propose using the APCMA protocol to effectively stabilize the operation of the wireless SNN. By modeling Pavlovian conditioning on this network, the learning capabilities of the implemented system can be determined.
The article contains results of training and testing machine learning models with captured network traffic data. The main goal is to perform classification of video traffic in computer networks. Multiple performance m...
详细信息
The article contains results of training and testing machine learning models with captured network traffic data. The main goal is to perform classification of video traffic in computer networks. Multiple performance metrics have been evaluated for commonly used classic supervised machine learning algorithms, as well as more advanced convolutional neuralnetwork model (for comparison). The article describes in detail the experimental setup, traffic pre-processing procedure, features extraction with different traffic window length and model parameters for training/testing. The article provides some experimental results in the form of tables and 3D surface plots. The conclusion of the article summarises the main findings and outlines the future study directions.
暂无评论