The purpose of this article is to propose a value-oriented electrical load forecasting (ELF) approach that aims to minimize load variance by leveraging vehicle-to-grid (V2G) technology. To achieve this, it is critical and urgent to design a loss function that can accurately measure the suboptimality of decisions induced by forecast errors, especially since the commonly used mean squared error (MSE) falls short in this regard. The Lagrange multiplier method and Karush–Kuhn–Tucker conditions are initially used to elucidate the impact of forecast errors on the actual charging power of electric vehicles. Building on this foundation, a differentiable loss function called mean relative magnitude error (MRME) is put forward for value-oriented ELF, which allows parameter updates in the long short-term memory (LSTM) model via the gradient descent method during the training stage. Furthermore, the MRME and MSE loss functions are combined using a weighted sum method to leverage their respective advantages. Numerous case studies have demonstrated that the LSTM model, whether using MRME alone or in combination with MSE, achieves superior and more robust V2G scheduling performance compared to using MSE alone. The weight coefficients for combining MRME and MSE loss functions are also discussed.

Visual object tracking has significantly promoted autonomous applications for unmanned aerial vehicles (UAVs). However, learning robust object representations for UAV tracking is especially challenging in complex dynamic environments, when confronted with aspect ratio change and occlusion. These challenges severely alter the original information of the object. To handle the above issues, this work proposes a novel progressive representation learning framework for UAV tracking, i.e., PRL-Track. Specifically, PRL-Track is divided into coarse representation learning and fine representation learning. For coarse representation learning, two innovative regulators, which rely on appearance and semantic information, are designed to mitigate appearance interference and capture semantic information. Furthermore, for fine representation learning, a new hierarchical modeling generator is developed to intertwine coarse object representations. Exhaustive experiments demonstrate that the proposed PRL-Track delivers exceptional performance on three authoritative UAV tracking benchmarks. Real-world tests indicate that the proposed PRL-Track realizes superior tracking performance with 42.6 frames per second on the typical UAV platform equipped with an edge smart camera. The code, model, and demo videos are available at https://github.com/vision4robotics/PRL-Track.

The growing prevalence of AI services on cloud platforms is driving the demand for technologies and tools which enable the integration of multiple AI services to handle intricate tasks. The growing prevalence of AI services on cloud platforms is driving the demand for technologies and tools which enable the integration of multiple AI services to handle intricate tasks. Traditional methods of evaluating intelligent systems focus mainly on the performance of AI components, without providing comprehensive metrics for the system as a whole. Additionally, as these AI components are often sourced from third-party providers, users may face challenges due to inconsistent quality assurance and limitations in further developing AI models, and dealing with third-party service providers’ limitations. These limitations often involve quality assurance and a lack of capability for secondary development and training of services. To address these issues, we have developed a tool based on our previous work. It can autonomously build Intelligent systems from AI services while tackling the issues mentioned above. This tool not only creates service composition solutions that align with user-defined functional requirements and performance metrics but also executes these solutions to verify if the metrics meet user requirements. We have demonstrated the effectiveness of this tool in constructing trustworthy intelligent systems through a series of case studies.

Few-sample modes are easy to appear when a new working condition is triggered in industrial processes especially during the early stages of the new working mode. However, monitoring the early behavior of a new mode is important because engineers and operators are less knowledgeable with such a new mode. Considering the few-sample challenge in this problem, a new multisource transfer learning framework is proposed that leverages historical data under various operating conditions to enrich process monitoring over new mode data. In contrast to existing transfer learning-related work, a new unsupervised domain adaptation framework is designed. The historical modes as the source provide precious knowledge and reference to the new mode so that the features of the new mode are robust to noise and insufficient samples. Mathematically, the historical features play the role of a regularizer for the feature learning in the target domain. A geometrical illustration is given and an iterative optimization algorithm is developed with the convergence analysis. Except for the features guided by historical modes, individual features of the new mode are also extracted from the residual part to form a complete monitoring framework. Finally, the effectiveness of the proposed method is validated through a numerical experiment and a real industrial hydrocracking process.

The extensive development of complex applications in embedded devices has driven the rapid development of edge computing, which provides powerful processing capabilities to the edge network. In this context of development, task offloading has received widespread attention as one of the key issues in edge computing. However, since a task usually consists of multiple subtasks with dependencies, the current computing subtask on the local or edge server must wait for the completion of the previous dependent subtask. As a result, existing offloading schemes are often constrained by the complexity of the task topology. Therefore, this paper is aimed at tasks with dependencies described as Directed Acyclic Graphs (DAGs) in devices. First, we optimize the order of subtasks with dynamic priorities. Meanwhile, we employ task duplication to reduce communication latency and thus overall task completion time. Additionally, to support the edge computing environment of multi-user and multi-server, game theory is used to find the optimal offload position for each user, so as to obtain the minimum average computing delay of all tasks. Experimental results show that the proposed algorithm outperforms existing algorithms in terms of task completion delay.

Existing multi-FPGA architectures often leverage high-speed interconnect technologies to achieve higher performance by exploiting ample communication bandwidth. In this paper, we propose an effective mapping approach for accelerating CNNs on bandwidth-constrained distributed multi-FPGA architectures. We formulate the system-level mapping problem and then introduce a method based on Genetic Algorithm (GA) and Mixed-Integer Nonlinear Programming (MINLP) to attain optimal solutions.