A practical issue of edge AI systems is that data distributions of trained dataset and deployed environment may differ due to noise and environmental changes over time. Such a phenomenon is known as a concept drift, and this gap degrades the performance of edge AI systems and may introduce system failures. To address this gap, retraining of neural network models triggered by concept drift detection is a practical approach. However, since available compute resources are strictly limited in edge devices, in this paper we propose a fully sequential concept drift detection method in cooperation with an on-device sequential learning technique of neural networks. In this case, both the neural network retraining and the proposed concept drift detection are done only by sequential computation to reduce computation cost and memory utilization. We use three datasets for experiments and compare the proposed approach with existing batch-based detection methods. It is also compared with a DNN-based approach without concept drift detection. The evaluation results of the proposed approach show that the proposed method is capable of detecting each of four concept drift types. The results also show that, while the accuracy is decreased by up to 0.9% compared to the existing batch-based detection methods, it decreases the memory size by 88.9%-96.4% and the execution time by 45.0%-87.6%. As a result, the combination of the neural network retraining and the proposed concept drift detection method is demonstrated on Raspberry Pi Pico that has 264 kB memory.

We explore the maximum total number of edge crossings and the maximum geometric thickness of the Euclidean minimum-weight (k, ℓ)-tight graph on a planar point set P. In this paper, we show that (10/7－ε)|P| and (11/6－ε)|P| are lower bounds for the maximum total number of edge crossings for any ε ＞ 0 in cases (k,ℓ)=(2,3) and (2,2), respectively. We also show that the lower bound for the maximum geometric thickness is 3 for both cases. In the proofs, we apply the method of arranging isomorphic units regularly. While the method is developed for the proof in case (k,ℓ)=(2,3), it also works for different ℓ.

Simple and quick tests at medical clinics have become increasingly important. Magnetic sensing techniques have been developed to detect biomarkers using magnetic nanoparticles in liquid-phase assays. We developed a biomarker assay that involves using an alternating current (AC) susceptibility measurement system that uses functional magnetic particles and magnetic sensing technology. We also developed compact biomarker measuring equipment to enable quick testing. Our assay is a one-step homogeneous assay that involves simply mixing a sample with a reagent, shortening testing time and simplifying processing. Using our compact measuring equipment, which includes anisotropic magneto resistance (AMR) sensors, we conducted high-sensitivity measurements of extremely small amounts of two biomarkers (C-reactive protein, CRP and α-Fetoprotein, AFP) used for diagnosing arteriosclerosis and malignant tumors. The results indicate that an extremely small amount of CRP and AFP could be detected within 15 min, which demonstrated the possibility of a simple and quick high-sensitivity immunoassay that involves using an AC-susceptibility measurement system.

In this work, the core size estimation technique of magnetic nanoparticles (MNPs) using the static magnetization curve obtained from a high-Tc SQUID magnetometer and a metaheuristic inversion technique based on the Particle Swarm Optimizer (PSO) algorithm is presented. The high-Tc SQUID magnetometer is constructed from a high-Tc SQUID sensor coupled by a flux transformer to sense the modulated magnetization signal from a sample. The magnetization signal is modulated by the lateral vibration of the sample on top of a planar differential detection coil of the flux transformer. A pair of primary and excitation coils are utilized to apply an excitation field parallel to the sensitive axis of the detection coil. Using the high-Tc SQUID magnetometer, the magnetization curve of a commercial MNP sample (Resovist) was measured in a logarithmic scale of the excitation field. The PSO inverse technique is then applied to the magnetization curve to construct the magnetic moment distribution. A multimodal normalized log-normal distribution was used in the minimization of the objective function of the PSO inversion technique, and a modification of the PSO search region is proposed to improve the exploration and exploitation of the PSO particles. As a result, a good agreement on the Resovist magnetic core size was obtained between the proposed technique and the non-negative least square (NNLS) inversion technique. The estimated core sizes of 8.0484 nm and 20.3018 nm agreed well with the values reported in the literature using the commercial low-Tc SQUID magnetometer with the SVD and NNLS inversion techniques. Compared to the NNLS inversion technique, the PSO inversion technique had merits in exploring an optimal core size distribution freely without being regularized by a parameter and facilitating an easy peak position determination owing to the smoothness of the constructed distribution. The combination of the high-Tc SQUID magnetometer and the PSO-based reconstruction technique offers a powerful approach for characterizing the MNP core size distribution, and further improvements can be expected from the recent state-of-the-art optimization algorithm to optimize further the computation time and the best objective function value.

In recent years, measuring biomagnetic fields in the Earth’s field by differential measurements of scalar-mode OPMs have been actively attempted. In this study, the sensitivity of the scalar-mode OPMs under the geomagnetic environment in the laboratory was studied by numerical simulation. Although the noise level of the scalar-mode OPM in the laboratory environment was calculated to be 104 pT/$\sqrt{\mathrm{Hz}}$, the noise levels using the first-order and the second-order differential configurations were found to be 529 fT/cm/$\sqrt{\mathrm{Hz}}$ and 17.2 fT/cm2/$\sqrt{\mathrm{Hz}}$, respectively. This result indicated that scalar-mode OPMs can measure very weak magnetic fields such as MEG without high-performance magnetic shield roomns. We also studied the operating conditions by varying repetition frequency and temperature. We found that scalar-mode OPMs have an upper limit of repetition frequency and temperature, and that the repetition frequency should be set below 4 kHz and the temperature should be set below 120°C.

A 0.13mJ/prediction with 68.6% accuracy wired-logic deep neural network (DNN) processor is developed in a single 16-nm field-programmable gate array (FPGA) chip. Compared with conventional von-Neumann architecture DNN processors, the energy efficiency is greatly improved by eliminating DRAM/BRAM access. A technical challenge for conventional wired-logic processors is the large amount of hardware resources required for implementing large-scale neural networks. To implement a large-scale convolutional neural network (CNN) into a single FPGA chip, two technologies are introduced: (1) a sparse neural network known as a non-linear neural network (NNN), and (2) a newly developed raster-scan wired-logic architecture. Furthermore, a novel high-level synthesis (HLS) technique for wired-logic processor is proposed. The proposed HLS technique enables the automatic generation of two key components: (1) Verilog-hardware description language (HDL) code for a raster-scan-based wired-logic processor and (2) test bench code for conducting equivalence checking. The automated process significantly mitigates the time and effort required for implementation and debugging. Compared with the state-of-the-art FPGA-based processor, 238 times better energy efficiency is achieved with only a slight decrease in accuracy on the CIFAR-100 task. In addition, 7 times better energy efficiency is achieved compared with the state-of-the-art network-optimized application-specific integrated circuit (ASIC).

The subject must be motionless for conventional radar-based non-contact vital signs measurements. Additionally, the measurement range is limited by the design of the radar module itself. Although the accuracy of measurements has been improving, the prospects for their application could have been faster to develop. This paper proposed a novel radar-based adaptive tracking method for measuring the heart rate of the moving monitored person. The radar module is fixed on a circular plate and driven by stepping motors to rotate it. In order to protect the user’s privacy, the method uses radar signal processing to detect the subject’s position to control a stepping motor that adjusts the radar’s measurement range. The results of the fixed-route experiments revealed that when the subject was moving at a speed of 0.5 m/s, the mean values of RMSE for heart rate measurements were all below 2.85 beat per minute (bpm), and when moving at a speed of 1 m/s, they were all below 4.05 bpm. When subjects walked at random routes and speeds, the RMSE of the measurements were all below 6.85 bpm, with a mean value of 4.35 bpm. The average RR interval time of the reconstructed heartbeat signal was highly correlated with the electrocardiography (ECG) data, with a correlation coefficient of 0.9905. In addition, this study not only evaluated the potential effect of arm swing (more normal walking motion) on heart rate measurement but also demonstrated the ability of the proposed method to measure heart rate in a multiple-people scenario.

Handover is an important property of cellular communication that enables the user to move from one cell to another without losing the connection. It is a very crucial process for the quality of the user’s experience because it may interrupt data transmission. Therefore, good handover management is very important in the current and future cellular systems. Several techniques have been employed to improve the handover performance, usually to increase the probability of a successful handover. One of the techniques is predictive handover which predicts the target cell using some methods other than the traditional measurement-based algorithm, including using machine learning. Several studies have been conducted in the implementation of predictive handover, most of them by modifying the internal algorithm of existing network elements, such as the base station. We implemented a predictive handover algorithm using an intelligent node outside the existing network elements to minimize the modification of the network and to create modularity in the system. Using a recently standardized Open Radio Access Network (O-RAN) Near Realtime Radio Intelligent Controller (Near-RT RIC), we created a modular application that can improve the handover performance by determining the target cell using machine learning techniques. In our previous research, we modified The Near-RT RIC original software that is using vector autoregression to determine the target cell by predicting the throughput of each neighboring cell. We also modified the method using a Multi-Layer Perceptron (MLP) neural network. In this paper, we redesigned the neural network using Long Short-Term Memory (LSTM) that can better handle time series data. We proved that our proposed LSTM-based machine learning algorithms used in Near-RT RIC can improve the handover performance compared to the traditional measurement-based algorithm.

With maturation of 5G technology in recent years, multimedia services such as live video streaming and online games on the Internet have flourished. These multimedia services frequently require low latency, which pose a significant challenge to compute the high latency requirements multimedia tasks. Mobile edge computing (MEC), is considered a key technology solution to address the above challenges. It offloads computation-intensive tasks to edge servers by sinking mobile nodes, which reduces task execution latency and relieves computing pressure on multimedia devices. In order to use MEC paradigm reasonably and efficiently, resource allocation has become a new challenge. In this paper, we focus on the multimedia tasks which need to be uploaded and processed in the network. We set the optimization problem with the goal of minimizing the latency and energy consumption required to perform tasks in multimedia devices. To solve the complex and non-convex problem, we formulate the optimization problem as a distributed deep reinforcement learning (DRL) problem and propose a federated Dueling deep Q-network (DDQN) based multimedia task offloading and resource allocation algorithm (FDRL-DDQN). In the algorithm, DRL is trained on the local device, while federated learning (FL) is responsible for aggregating and updating the parameters from the trained local models. Further, in order to solve the not identically and independently distributed (non-IID) data problem of multimedia devices, we develop a method for selecting participating federated devices. The simulation results show that the FDRL-DDQN algorithm can reduce the total cost by 31.3% compared to the DQN algorithm when the task data is 1000 kbit, and the maximum reduction can be 35.3% compared to the traditional baseline algorithm.

In this study, we propose a novel dataset distillation method based on parameter pruning. The proposed method can synthesize more robust distilled datasets and improve distillation performance by pruning difficult-to-match parameters during the distillation process. Experimental results on two benchmark datasets show the superiority of the proposed method.

Direct-current biased optical orthogonal frequency division multiplexing (DCO-OFDM) converts bipolar OFDM signals into unipolar non-negative signals by introducing a high DC bias, which satisfies the requirement that the signal transmitted by intensity modulated/direct detection (IM/DD) must be positive. However, the high DC bias results in low power efficiency of DCO-OFDM. An adaptively biased optical OFDM was proposed, which could be designed with different biases according to the signal amplitude to improve power efficiency in this letter. The adaptive bias does not need to be taken off deliberately at the receiver, and the interference caused by the adaptive bias will only be placed on the reserved subcarriers, which will not affect the effective information. Moreover, the proposed OFDM uses Hartley transform instead of Fourier transform used in conventional optical OFDM, which makes this OFDM have low computational complexity and high spectral efficiency. The simulation results show that the normalized optical bit energy to noise power ratio (Eb(opt)/N0) required by the proposed OFDM at the bit error rate (BER) of 10-3 is, on average, 7.5 dB and 3.4 dB lower than that of DCO-OFDM and superimposed asymmetrically clipped optical OFDM (ACO-OFDM), respectively.

The steady-state and convergence performances are important indicators to evaluate adaptive algorithms. The step-size affects these two important indicators directly. Many relevant scholars have also proposed some variable step-size adaptive algorithms for improving performance. However, there are still some problems in these existing variable step-size adaptive algorithms, such as the insufficient theoretical analysis, the imbalanced performance and the unachievable parameter. These problems influence the actual performance of some algorithms greatly. Therefore, we intend to further explore an inherent relationship between the key performance and the step-size in this paper. The variation of mean square deviation (MSD) is adopted as the cost function. Based on some theoretical analyses and derivations, a novel variable step-size algorithm with a dynamic limited function (DLF) was proposed. At the same time, the sufficient theoretical analysis is conducted on the weight deviation and the convergence stability. The proposed algorithm is also tested with some typical algorithms in many different environments. Both the theoretical analysis and the experimental result all have verified that the proposed algorithm equips a superior performance.

The automatic defect detection for fabric images is an essential mission in textile industry. However, there are some inherent difficulties in the detection of fabric images, such as complexity of the background and the highly uneven scales of defects. Moreover, the trade-off between accuracy and speed should be considered in real applications. To address these problems, we propose a novel model based on YOLOv4 to detect defects in fabric images, called Feature Augmentation YOLO (FA-YOLO). In terms of network structure, FA-YOLO adds an additional detection head to improve the detection ability of small defects and builds a powerful Neck structure to enhance feature fusion. First, to reduce information loss during feature fusion, we perform the residual feature augmentation (RFA) on the features after dimensionality reduction by using 1×1 convolution. Afterward, the attention module (SimAM) is embedded into the locations with rich features to improve the adaptation ability to complex backgrounds. Adaptive spatial feature fusion (ASFF) is also applied to output of the Neck to filter inconsistencies across layers. Finally, the cross-stage partial (CSP) structure is introduced for optimization. Experimental results based on three real industrial datasets, including Tianchi fabric dataset (72.5% mAP), ZJU-Leaper fabric dataset (0.714 of average F1-score) and NEU-DET steel dataset (77.2% mAP), demonstrate the proposed FA-YOLO achieves competitive results compared to other state-of-the-art (SoTA) methods.

This paper discusses the resilience of networks based on graph theory and stochastic process. The electric power network where edges may fail simultaneously and the performance of the network is measured by the ratio of connected nodes is supposed for the target network. For the restoration, under the constraint that the resources are limited, the failed edges are repaired one by one, and the order of the repair for several failed edges is determined with the priority to the edge that the amount of increasing system performance is the largest after the completion of repair. Two types of resilience are discussed, one is resilience in the recovery stage according to the conventional definition of resilience and the other is steady state operational resilience considering the long-term operation in which the network state changes stochastically. The second represents a comprehensive capacity of resilience for a system and is analytically derived by Markov analysis. We assume that the large-scale disruption occurs due to the simultaneous failure of edges caused by the common cause failures in the analysis. Marshall-Olkin type shock model and α factor method are incorporated to model the common cause failures. Then two resilience measures, “operational resilience” and “operational resilience in recovery stage” are proposed. We also propose approximation methods to obtain these two operational resilience measures for complex networks.

This study investigates the efficacy of a partial decomposition approach in concept map recomposition tasks to reduce cognitive load while maintaining the benefits of traditional recomposition approaches. Prior research has demonstrated that concept map recomposition, involving the rearrangement of unconnected concepts and links, can enhance reading comprehension. However, this task often imposes a significant burden on learners’ working memory. To address this challenge, this study proposes a partial recomposition approach where learners are tasked with recomposing only a portion of the concept map, thereby reducing the problem space. The proposed approach aims at lowering the cognitive load while maintaining the benefits of traditional recomposition task, that is, learning effect and motivation. To investigate the differences in cognitive load, learning effect, and motivation between the full decomposition (the traditional approach) and partial decomposition (the proposed approach), we have conducted an experiment (N=78) where the participants were divided into two groups of “full decomposition” and “partial decomposition”. The full decomposition group was assigned the task of recomposing a concept map from a set of unconnected concept nodes and links, while the partial decomposition group worked with partially connected nodes and links. The experimental results show a significant reduction in the embedded cognitive load of concept map recomposition across different dimensions while learning effect and motivation remained similar between the conditions. On the basis of these findings, educators are recommended to incorporate partially disconnected concept maps in recomposition tasks to optimize time management and sustain learner motivation. By implementing this approach, instructors can conserve cognitive resources and allocate saved energy and time to other activities that enhance the overall learning process.

When new users join a recommender system, traditional approaches encounter challenges in accurately understanding their interests due to the absence of historical user behavior data, thus making it difficult to provide personalized recommendations. Currently, two main methods are employed to address this issue from different perspectives. One approach is centered on meta-learning, enabling models to adapt faster to new tasks by sharing knowledge and experiences across multiple tasks. However, these methods often overlook potential improvements based on cross-domain information. The other method involves cross-domain recommender systems, which transfer learned knowledge to different domains using shared models and transfer learning techniques. Nonetheless, this approach has certain limitations, as it necessitates a substantial amount of labeled data for training and may not accurately capture users’ latent preferences when dealing with a limited number of samples. Therefore, a crucial need arises to devise a novel method that amalgamates cross-domain information and latent preference extraction to address this challenge. To accomplish this objective, we propose a Cross-domain Recommender System based on Domain Knowledge Transferor and Latent Preference Extractor (TECDR).  In TECDR, we have designed a Latent Preference Extractor that transforms user behaviors into representations of their latent interests in items. Additionally, we have introduced a Domain Knowledge Transfer mechanism for transferring knowledge and patterns between domains. Moreover, we leverage meta-learning-based optimization methods to assist the model in adapting to new tasks. The experimental results from three cross-domain scenarios demonstrate that TECDR exhibits outstanding performance across various cross-domain recommender scenarios.

Traditional intuitionistic fuzzy sets and hesitant fuzzy sets will lose some information while representing vague information, to avoid this problem, this paper constructs weighted generalized hesitant fuzzy sets by remaining multiple intuitionistic fuzzy values and giving them corresponding weights. For weighted generalized hesitant fuzzy elements in weighted generalized hesitant fuzzy sets, the paper defines some basic operations and proves their operation properties. On this basis, the paper gives the comparison rules of weighted generalized hesitant fuzzy elements and presents two kinds of aggregation operators. As for weighted generalized hesitant fuzzy preference relation, this paper proposes its definition and computing method of its corresponding consistency index. Furthermore, the paper designs an ensemble learning algorithm based on weighted generalized hesitant fuzzy sets, carries out experiments on 6 datasets in UCI database and compares with various classification algorithms. The experiments show that the ensemble learning algorithm based on weighted generalized hesitant fuzzy sets has better performance in all indicators.

Graph dissimilarities provide a powerful and ubiquitous approach for applying machine learning algorithms to edge-attributed graphs. However, conventional optimal transport-based dissimilarities cannot handle edge-attributes. In this paper, we propose an optimal transport-based dissimilarity between graphs with edge-attributes. The proposed method, multi-dimensional fused Gromov-Wasserstein discrepancy (MFGW), naturally incorporates the mismatch of edge-attributes into the optimal transport theory. Unlike conventional optimal transport-based dissimilarities, MFGW can directly handle edge-attributes in addition to structural information of graphs. Furthermore, we propose an iterative algorithm, which can be computed on GPUs, to solve non-convex quadratic programming problems involved in MFGW.  Experimentally, we demonstrate that MFGW outperforms the conventional optimal transport-based dissimilarity in several machine learning applications including supervised classification, subgraph matching, and graph barycenter calculation.

The study proposes a personalised session-based recommender system that embeds items by using Word2Vec and sequentially updates the session and user embeddings with the hierarchicalization and aggregation of item embeddings. To process a recommendation request, the system constructs a real-time user embedding that considers users’ general preferences and sequential behaviour to handle short-term changes in user preferences with a low computational cost. The system performance was experimentally evaluated in terms of the accuracy, diversity, and novelty of the ranking of recommended items and the training and prediction times of the system for three different datasets. The results of these evaluations were then compared with those of the five baseline systems. According to the evaluation experiment, the proposed system achieved a relatively high recommendation accuracy compared with baseline systems and the diversity and novelty scores of the proposed system did not fall below 90% for any dataset. Furthermore, the training times of the Word2Vec-based systems, including the proposed system, were shorter than those of FPMC and GRU4Rec. The evaluation results suggest that the proposed recommender system succeeds in keeping the computational cost for training low while maintaining high-level recommendation accuracy, diversity, and novelty.

Forecasting time-series data is useful in many fields, such as stock price predicting system, autonomous driving system, weather forecast, etc. Many existing forecasting models tend to work well when forecasting short-sequence time series. However, when working with long sequence time series, the performance suffers significantly. Recently, there has been more intense research in this direction, and Informer is currently the most efficient predicting model. Informer’s main drawback is that it does not allow for incremental learning. In this paper, we propose a Fast Informer called Finformer, which addresses the above bottleneck by reducing the training/predicting time of Informer. Finformer can efficiently compute the positional/temporal/value embedding and Query/Key/Value of the self-attention incrementally. Theoretically, Finformer can improve the speed of both training and predicting over the state-of-the-art model Informer. Extensive experiments show that Finformer is about 26% faster than Informer for both short and long sequence time series prediction. In addition, Finformer is about 20% faster than InTrans for the general Conv1d, which is one of our previous works and is the predecessor of Finformer.