“数字电子技术基础”MOOC设计与应用

采用线上线下混合教学模式已经成为提升教学质量的必要手段。2021年3月数字电子技术基础在中国大学MOOC平台上线,基于项目教学法,重构知识体系,以4个精心设计的、循序渐进的、侧重不同知识点的、具有代表性的项目为课程脉络主线,讲解过程中广泛采用对比教学法、层次教学法、错误引导法,逐步引领学生建构知识体系。本MOOC课程可有效缓解学时不断减少与挑战度要求不断提高之间的矛盾。     

Light gradient boosting machine with optimized hyperparameters for identification of malicious access in IoT network

In this paper, an advanced and optimized Light Gradient Boosting Machine (LGBM) technique is proposed to identify the intrusive activities in the Internet of Things (IoT) network. The followings are the major contributions: i) An optimized LGBM model has been developed for the identification of malicious IoT activities in the IoT network; ii) An efficient evolutionary optimization approach has been adopted for finding the optimal set of hyper-parameters of LGBM for the projected problem. Here, a Genetic Algorithm (GA) with k-way tournament selection and uniform crossover operation is used for efficient exploration of hyper-parameter search space; iii) Finally, the performance of the proposed model is evaluated using state-of-the-art ensemble learning and machine learning-based model to achieve overall generalized performance and efficiency. Simulation outcomes reveal that the proposed approach is superior to other considered methods and proves to be a robust approach to intrusion detection in an IoT environment.     

基于密度拟合的区域航道挖掘方法

雷达目标跟踪主要利用从回波信号中得到的目标位置、速度等信息,所用的先验信息较少。由于地形、燃料等因素,空中/海面目标通常是会沿着一定航道进行运动,但这一信息通常难以作为先验信息提供给雷达观测,用以辅助雷达跟踪。因此本文提出基于航迹密集程度的区域航道挖掘方法,从航迹数据集提取航道,可为后续相同区域内的目标跟踪提供必要的辅助信息支持。本方法对航迹进行参数拟合,基于参数聚类找到航迹高密集区域并利用最小二乘提取航道;它在无任何先验知识的前提下,仅利用已有航迹数据就可实现对区域内航道快速、准确提取。基于实测数据,通过本方法与其他方法的实验对比,我们验证了本方法提取航迹的有效性和准确性。     

Learning knowledge representation with meta knowledge distillation for single image super-resolution

Although the deep CNN-based super-resolution methods have achieved outstanding performance, their memory cost and computational complexity severely limit their practical employment. Knowledge distillation (KD), which can efficiently transfer knowledge from a cumbersome network (teacher) to a compact network (student), has demonstrated its advantages in some computer vision applications. The representation of knowledge is vital for knowledge transferring and student learning, which is generally defined in hand-crafted manners or uses the intermediate features directly. In this paper, we propose a model-agnostic meta knowledge distillation method under the teacher–student architecture for the single image super-resolution task. It provides a more flexible and accurate way to help teachers transmit knowledge in accordance with the abilities of students via knowledge representation networks (KRNets) with learnable parameters. Specifically, the texture-aware dynamic kernels are generated from local information to decompose the distillation problem into texture-wise supervision for further promoting the recovery quality of high-frequency details. In addition, the KRNets are optimized in a meta-learning manner to ensure the knowledge transferring and the student learning are beneficial to improving the reconstructed quality of the student. Experiments conducted on various single image super-resolution datasets demonstrate that our proposed method outperforms existing defined knowledge representation-related distillation methods and can help super-resolution algorithms achieve better reconstruction quality without introducing any extra inference complexity.     

HSP-MFL: A High-level Semantic Property driven Multi-task Feature Learning Network for unsupervised person Re-ID

Unsupervised person re-identification aims to distinguish different pedestrians from discriminative representations on the basis of unlabeled data. Currently, most unsupervised Re-ID approaches explore visual representations to generate pseudo-labels for model’s training, which may suffer from background noise and semantic loss. To tackle this problem, this paper proposes a High-level Semantic Property driven Multi-task Feature Learning Network (HSP-MFL) to firstly introduce three high-level semantic properties for unsupervised person Re-ID. Technically, we design a novel Multiple Feature Fusion Module (MFFM) to deeply explore the complex correlation among multiple semantic and visual features to capture the discriminative feature cues, as well as a multi-task training scheme to generate robust fusion features. The architecture is quite simple and does not consume extra labeling costs. Extensive experiments on three datasets demonstrate that both high-level semantic properties and multi-task learning are effective in performance improvement, yielding SOTA mAPs for unsupervised person Re-ID.     

Room-Temperature-Processable Highly Reliable Resistive Switching Memory with Reconfigurability for Neuromorphic Computing and Ultrasonic Tissue Classification

Reversible metal-filamentary mechanism has been widely investigated to design an analog resistive switching memory (RSM) for neuromorphic hardware-implementation. However, uncontrollable filament-formation, inducing its reliability issues, has been a fundamental challenge. Here, an analog RSM with 3D ion transport channels that can provide unprecedentedly high reliability and robustness is demonstrated. This architecture is realized by a laser-assisted photo-thermochemical process, compatible with the back-end-of-line process and even applicable to a flexible format. These superior characteristics also lead to the proposal of a practical adaptive learning rule for hardware neural networks that can significantly simplify the voltage pulse application methodology even with high computing accuracy. A neural network, which can perform the biological tissue classification task using the ultrasound signals, is designed, and the simulation results confirm that this practical adaptive learning rule is efficient enough to classify these weak and complicated signals with high accuracy (97%). Furthermore, the proposed RSM can work as a diffusive-memristor at the opposite voltage polarity, exhibiting extremely stable threshold switching characteristics. In this mode, several crucial operations in biological nervous systems, such as Ca²⁺ dynamics and nonlinear integrate-and-fire functions of neurons, are successfully emulated. This reconfigurability is also exceedingly beneficial for decreasing the complexity of systems—requiring both drift- and diffusive-memristors.     

Solid Ink Laser Patterning for High-Resolution Information Labels with Supervised Learning Readout

Tagging, tracking, or validation of products are often facilitated by inkjet-printed optical information labels. However, this requires thorough substrate pretreatment, ink optimization, and often lacks in printing precision/resolution. Herein, a printing method based on laser-driven deposition of solid polymer ink that allows for printing on various substrates without pretreatment is demonstrated. Since the deposition process has a precision of <1 µm, it can introduce the concept of sub-positions with overlapping spots. This enables high-resolution fluorescent labels with comparable spot-to-spot distance of down to 15 µm (444,444 spots cm⁻²) and rapid machine learning-supported readout based on low-resolution fluorescence imaging. Furthermore, the defined thickness of the printed polymer ink spots can be used to fabricate multi-channel information labels. Additional information can be stored in different fluorescence channels or in a hidden topography channel of the label that is independent of the fluorescence.     

基于改进麻雀搜索算法优化BN的变压器故障诊断研究

针对变压器故障诊断准确率低和稳定性差的问题,文中提出了一种改进麻雀搜索算法优化贝叶斯网络的变压器故障诊断方法。首先,通过计算互信息建立最大支撑树并进行定向处理得到贝叶斯网络初始结构即初始种群。然后,在算法中引入一种新的合作机制和正弦余弦算法,提高算法收敛速度和全局搜索能力,并利用油中溶解气体分析,创建基于改进麻雀搜索算法优化贝叶斯网络的变压器故障诊断模型。最后,为了证明所提方法的优越性,将所提的方法与现有变压器故障诊断方法进行对比。结果表明,文中所提出的方法故障诊断率最高,可以更精准地对变压器进行故障诊断。     

道路交通网络节点分配优化策略研究进展

为了进一步提高城市道路交通网络的通行效率,粒子群优化和神经网络等多种智能优化算法受到越来越多的关注。近年来,深度学习技术的普及与应用大幅提升了城市交通网络的节点识别效率,而交通网络的节点调度又扩展了深度学习技术的应用。文中详细分析了交通节点调度所面临的关键问题,归纳并总结了相关网络节点分配的研究现状。在此基础上,深入研讨了城市交通网络节点调度与深度学习的应用前景,并对交通网络节点分配优化策略的未来研究方向进行了展望。     

Direct Freeform Laser Fabrication of 3D Conformable Electronics

3D conformable electronic devices on freeform surfaces show superior performance to the conventional, planar ones. They represent a trend of future electronics and have witnessed exponential growth in various applications. However, their potential is largely limited by a lack of sophisticated fabrication techniques. To tackle this challenge, a new direct freeform laser (DFL) fabrication method enabled by a 5-axis laser processing platform for directly fabricating 3D conformable electronics on targeted arbitrary surfaces is reported. Accordingly, representative laser-induced graphene (LIG), metals, and metal oxides are successfully fabricated as high-performance sensing and electrode materials from different material precursors on various types of substrates for applications in temperature/light/gas sensing, energy storage, and printed circuit board for circuit. Last but not the least, to demonstrate an application in smart homes, LIG-based conformable strain sensors are fabricated and distributed in designated locations of an artificial tree. The distributed sensors have the capability of monitoring the wind speed and direction with the assistance of well-trained machine-learning models. This novel process will pave a new and general route to fabricating 3D conformable electronic devices, thus creating new opportunities in robotics, biomedical sensing, structural health, environmental monitoring, and Internet of Things applications.