اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدSupervised Compression for Resource-constrained Edge Computing Systems
299
0
0.0
(
0
)
نشر من قبل
Yoshitomo Matsubara
تاريخ النشر
2021
مجال البحث
الهندسة المعلوماتية
والبحث باللغة
English
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
There has been much interest in deploying deep learning algorithms on low-powered devices, including smartphones, drones, and medical sensors. However, full-scale deep neural networks are often too resource-intensive in terms of energy and storage. As a result, the bulk part of the machine learning operation is therefore often carried out on an edge server, where the data is compressed and transmitted. However, compressing data (such as images) leads to transmitting information irrelevant to the supervised task. Another popular approach is to split the deep network between the device and the server while compressing intermediate features. To date, however, such split computing strategies have barely outperformed the aforementioned naive data compression baselines due to their inefficient approaches to feature compression. This paper adopts ideas from knowledge distillation and neural image compression to compress intermediate feature representations more efficiently. Our supervised compression approach uses a teacher model and a student model with a stochastic bottleneck and learnable prior for entropy coding. We compare our approach to various neural image and feature compression baselines in three vision tasks and found that it achieves better supervised rate-distortion performance while also maintaining smaller end-to-end latency. We furthermore show that the learned feature representations can be tuned to serve multiple downstream tasks.
قيم البحث
اقرأ أيضاً
In this paper, we propose a Distributed Intelligent Video Surveillance (DIVS) system using Deep Learning (DL) algorithms and deploy it in an edge computing environment. We establish a multi-layer edge computing architecture and a distributed DL train
ing model for the DIVS system. The DIVS system can migrate computing workloads from the network center to network edges to reduce huge network communication overhead and provide low-latency and accurate video analysis solutions. We implement the proposed DIVS system and address the problems of parallel training, model synchronization, and workload balancing. Task-level parallel and model-level parallel training methods are proposed to further accelerate the video analysis process. In addition, we propose a model parameter updating method to achieve model synchronization of the global DL model in a distributed EC environment. Moreover, a dynamic data migration approach is proposed to address the imbalance of workload and computational power of edge nodes. Experimental results showed that the EC architecture can provide elastic and scalable computing power, and the proposed DIVS system can efficiently handle video surveillance and analysis tasks.
Neural networks training on edge terminals is essential for edge AI computing, which needs to be adaptive to evolving environment. Quantised models can efficiently run on edge devices, but existing training methods for these compact models are design
ed to run on powerful servers with abundant memory and energy budget. For example, quantisation-aware training (QAT) method involves two copies of model parameters, which is usually beyond the capacity of on-chip memory in edge devices. Data movement between off-chip and on-chip memory is energy demanding as well. The resource requirements are trivial for powerful servers, but critical for edge devices. To mitigate these issues, We propose Resource Constrained Training (RCT). RCT only keeps a quantised model throughout the training, so that the memory requirements for model parameters in training is reduced. It adjusts per-layer bitwidth dynamically in order to save energy when a model can learn effectively with lower precision. We carry out experiments with representative models and tasks in image application and natural language processing. Experiments show that RCT saves more than 86% energy for General Matrix Multiply (GEMM) and saves more than 46% memory for model parameters, with limited accuracy loss. Comparing with QAT-based method, RCT saves about half of energy on moving model parameters.
In Federated Learning (FL), a global statistical model is developed by encouraging mobile users to perform the model training on their local data and aggregating the output local model parameters in an iterative manner. However, due to limited energy
and computation capability at the mobile devices, the performance of the model training is always at stake to meet the objective of local energy minimization. In this regard, Multi-access Edge Computing (MEC)-enabled FL addresses the tradeoff between the model performance and the energy consumption of the mobile devices by allowing users to offload a portion of their local dataset to an edge server for the model training. Since the edge server has high computation capability, the time consumption of the model training at the edge server is insignificant. However, the time consumption for dataset offloading from mobile users to the edge server has a significant impact on the total time consumption. Thus, resource management in MEC-enabled FL is challenging, where the objective is to reduce the total time consumption while saving the energy consumption of the mobile devices. In this paper, we formulate an energy-aware resource management for MEC-enabled FL in which the model training loss and the total time consumption are jointly minimized, while considering the energy limitation of mobile devices. In addition, we recast the formulated problem as a Generalized Nash Equilibrium Problem (GNEP) to capture the coupling constraints between the radio resource management and dataset offloading. We then analyze the impact of the dataset offloading and computing resource allocation on the model training loss, time, and the energy consumption.
Mobile-edge computing (MEC) has recently emerged as a prominent technology to liberate mobile devices from computationally intensive workloads, by offloading them to the proximate MEC server. To make offloading effective, the radio and computational
resources need to be dynamically managed, to cope with the time-varying computation demands and wireless fading channels. In this paper, we develop an online joint radio and computational resource management algorithm for multi-user MEC systems, with the objective as minimizing the long-term average weighted sum power consumption of the mobile devices and the MEC server, subject to a task buffer stability constraint. Specifically, at each time slot, the optimal CPU-cycle frequencies of the mobile devices are obtained in closed forms, and the optimal transmit power and bandwidth allocation for computation offloading are determined with the Gauss-Seidel method; while for the MEC server, both the optimal frequencies of the CPU cores and the optimal MEC server scheduling decision are derived in closed forms. Besides, a delay-improved mechanism is proposed to reduce the execution delay. Rigorous performance analysis is conducted for the proposed algorithm and its delay-improved version, indicating that the weighted sum power consumption and execution delay obey an $left[Oleft(1slash Vright),Oleft(Vright)right]$ tradeoff with $V$ as a control parameter. Simulation results are provided to validate the theoretical analysis and demonstrate the impacts of various parameters.
Unmanned Aerial Vehicles (UAVs) have attracted great interest in the last few years owing to their ability to cover large areas and access difficult and hazardous target zones, which is not the case of traditional systems relying on direct observatio
ns obtained from fixed cameras and sensors. Furthermore, thanks to the advancements in computer vision and machine learning, UAVs are being adopted for a broad range of solutions and applications. However, Deep Neural Networks (DNNs) are progressing toward deeper and complex models that prevent them from being executed on-board. In this paper, we propose a DNN distribution methodology within UAVs to enable data classification in resource-constrained devices and avoid extra delays introduced by the server-based solutions due to data communication over air-to-ground links. The proposed method is formulated as an optimization problem that aims to minimize the latency between data collection and decision-making while considering the mobility model and the resource constraints of the UAVs as part of the air-to-air communication. We also introduce the mobility prediction to adapt our system to the dynamics of UAVs and the network variation. The simulation conducted to evaluate the performance and benchmark the proposed methods, namely Optimal UAV-based Layer Distribution (OULD) and OULD with Mobility Prediction (OULD-MP), were run in an HPC cluster. The obtained results show that our optimization solution outperforms the existing and heuristic-based approaches.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
