To realise the scientific and clinical benefits of machine learning (ML) in a multi-centre research collaboration, a common issue is tire need to bring high-volnme data, complex analytical algorithms, and large-scale ...
详细信息
In this paper, we propose a computational offloading strategy tailored for multiple industrial equipment (IE) within an Industrial Internet of things (IIoT) framework. Our primary focus is on addressing the computatio...
详细信息
ISBN:
(数字)9798350390643
ISBN:
(纸本)9798350390650
In this paper, we propose a computational offloading strategy tailored for multiple industrial equipment (IE) within an Industrial Internet of things (IIoT) framework. Our primary focus is on addressing the computational offloading problem, specifically by modeling the associative relationship between tasks as a Directed Acyclic Graph (DAG). Initially, we transform the DAG task offloading problem into a Markov decision process (MDP). Subsequently, to derive the optimal offloading decision, we propose a novel approach termed DAG Task offloading with Message Passing Neural Network (MPNN) and Double Deep Q-Network (DAGTO-MAD). the proposed algorithm aims to min-imize the completion time of the task set. Moreover, we enhance the performance of the MPNN by adapting it to extract structural information from the DAG more efficiently. Simulation results demonstrate that our proposed algorithm achieves a shorter completion time compared to existing offloading algorithms across various environments. Additionally, our algorithm exhibits robust convergence and versatility, highlighting its effectiveness in real-application scenarios.
In the rapidly evolving machine learning (ML) and distributed systems realm, the escalating concern for data privacy naturally comes to the forefront of discussions. Federated learning (FL) emerges as a pivotal techno...
详细信息
ISBN:
(纸本)9798400702341
In the rapidly evolving machine learning (ML) and distributed systems realm, the escalating concern for data privacy naturally comes to the forefront of discussions. Federated learning (FL) emerges as a pivotal technology capable of addressing the inherent issues of centralized data privacy. However, FL architectures with centralized orchestration are still vulnerable, especially in the aggregation phase. A malicious server can exploit the aggregation process to learn about participants' data. this study proposes MPCFL, a secure FL algorithm based on secure multi-party computation (MPC) and secret sharing. the proposed algorithm leverages the Sharemind MPC framework to aggregate local model updates for securely formulating a global model. MPCFL provides practical mitigation of trending FL concerns, e.g., inference attack, gradient leakage attack, model poisoning, and model inversion. the algorithm is evaluated on several benchmark datasets and shows promising results. Our results demonstrate that the proposed algorithm is viable for developing secure and privacy-preserving FL applications, significantly improving all performance metrics while maintaining security and reliability. this investigation is a precursor to deeper explorations to craft robust FL aggregation algorithms.
Significant progress in clustering has been achieved by algorithmsthat are based on pairwise affinities between the datapoints. In particular, spectral clustering methods have the advantage of being able to divide ar...
Significant progress in clustering has been achieved by algorithmsthat are based on pairwise affinities between the datapoints. In particular, spectral clustering methods have the advantage of being able to divide arbitrarily shaped clusters and are based on efficient eigenvector calculations. However, spectral methods lack a straightforward probabilistic interpretation which makes it difficult to automatically set parameters using training *** this paper we use the previously proposed typical cut framework for pairwise clustering. We show an equivalence between calculating the typical cut and inference in an undirected graphical model. We show that for clustering problems with hundreds of datapoints exact inference may still be possible. For more complicated datasets, we show that loopy belief propagation (BP) and generalized belief propagation (GBP) can give excellent results on challenging clustering problems. We also use graphical models to derive a learning algorithm for affinity matrices based on labeled data.
this paper develops an integrated optical sensing, communication, and computation system for quadruped robot with virtual-real interaction based on digital twin technology. the system is designed to address key challe...
详细信息
ISBN:
(数字)9798350390643
ISBN:
(纸本)9798350390650
this paper develops an integrated optical sensing, communication, and computation system for quadruped robot with virtual-real interaction based on digital twin technology. the system is designed to address key challenges in robotics technology, including real-time performance, low-latency control, high-precision multi-sensor data fusion, stable network transmission, user-friendly interaction interface, system scalability, and maintainability. the system comprises a number of functional modules, including a three-dimensional (3D) modeling module, a positioning perception module, a virtual interaction module, a wise sensing-transmission module, and a cloud server. the 3D modeling module is responsible for constructing the virtual quadruped robot and motion space scenarios. the positioning perception module integrates LiDAR and inertial measurement unit (IMU) data, utilizing Point-light detection and ranging inertial odometry (LIO) and HDL-localization algorithms for high-precision environmental perception and positioning. the virtual interaction module provides a user-friendly control inter-face through computer software and the HoloLens headset. the wise sensing-transmission module employs WiFi and 5G links to ensure low-latency and high-bandwidth data transmission. the optical signal, sensed via camera and LiDAR, is computed at the quadruped robot and subsequently transmitted to the cloud server via the wise sensing-transmission module, which facilitates integrated optical sensing, communication, and computation. the system is designed to run on Ubuntu 20.04 platform, offering excellent scalability and maintainability.
Tracking of multiple ground targets with airborne Ground Moving Target Indicator (GMTI) sensor measurements is a challenging problem where heavy and dense false alarms with high target density are inevitably encounter...
详细信息
ISBN:
(纸本)9781479902842
Tracking of multiple ground targets with airborne Ground Moving Target Indicator (GMTI) sensor measurements is a challenging problem where heavy and dense false alarms with high target density are inevitably encountered in the surveillance scenes. Hence, optimal approaches require heavy computational load where the duration of overall computation rises exponentially withthe number of target tracks and measurements in observation per scan. Consequently, more practical suboptimal approaches, such as Linear Multi-Target (LM) approach, is explored due to linear number of operations in the number of target tracks with a negligible performance loss compared to optimal approaches. Although LM approach performs modestly adequate with significantly less computation duration than optimal approaches, it is highly susceptible to track loss, as in the rest of suboptimal approaches, when the targets are closely spaced and the number of targets and measurements are considerably high. Simulations are carried out in realistic test scenarios to compare single target tracking algorithms including IMM-PDA and IMM-IPDA algorithms;Optimal approaches in multitarget tracking including IMM-JPDA, IMM-IJPDA and IMM-JIPDA algorithms and an example of Linear Multi-target approaches in multitarget tracking including IMM-LMIPDA algorithm. Benchmarkings of these algorithms are done under RMSE performance, track loss and computation time evaluation results.
暂无评论