This paper presents a method for automatic generation of a training dataset for a deep convolutional neural network used for playing card detection. The solution allows to skip the time-consuming processes of manual image collecting and labelling recognised objects. The YOLOv4 network trained on the generated dataset achieved an efficiency of 99.8% in the cards detection task. The proposed method is a part of a project that aims to automate the process of broadcasting duplicate bridge competitions using a vision system and neural networks.

This paper presents a closed-loop current balancing control for Modular Multilevel Converters with Interleaved half-bridge Sub-Modules (ISM-MMC). The new control loop solves the well-known problem of proper current balancing among interleaved half-bridge legs in each ISM-MMC submodule while preserving a simple and reliable structure. In addition to that, a novel capacitor voltage balancing strategy for MMCs is developed. The new algorithm contains main advantages of the classical capacitor voltage balancing methods while provides an opportunity to decouple two balancing tasks of ISM-MMC, namely capacitor voltage and interleaved legs current balancing. The proposed control methods feature good dynamic performance and are compliant with a digital processor’s operational constraints. The effectiveness of the new balancing methods was verified during extensive numerical simulations and experimental tests on a laboratory prototype by the corresponding system response under the input/output characteristics variation and interleaved current control perturbation.

In this paper, we propose CSQN, a new Continual Learning (CL) method which considers Quasi-Newton methods, more specifically, Sampled Quasi-Newton methods, to extend EWC. EWC uses a Bayesian framework to estimate which parameters are important to previous tasks, and it punishes changes made to these parameters. However, it assumes that parameters are independent, as it does not consider interactions between parameters. With CSQN, we aim to overcome this.

Accurately sensing AC magnetic field signatures poses a series of challenges to commonly used Hall-effect sensors. In particular, induced voltage and lack of high-frequency spinning methods are bottlenecks in the measurement of AC magnetic fields. We describe a magnetic field measurement technique that can be implemented in two ways: 1) the current driving the Hall-effect sensor is oscillating at the same frequency as the magnetic field, and the signal is measured at the second harmonic of the magnetic field frequency, and 2) the frequency of the driving current is preset, and the measured frequency is the magnetic field frequency plus the frequency of the current. This method has potential advantages over traditional means of measuring AC magnetic fields used in power systems (e.g., motors, inverters), as it can reduce the components needed (subsequently reducing the overall cost and size) and is not frequency bandwidth limited by current spinning. The sensing technique produces no induced voltage and results in a low offset, thus preserving accuracy and precision in measurements. Experimentally, we have shown offset voltage values between 8 and 27 μT at frequencies ranging from 100 Hz to 1 kHz, validating the potential of this technique in both cases

The Routing Protocol for low power Lossy networks (RPL) is a critical operational component of low power wireless personal area networks using IPv6 (6LoWPANs). In this paper we propose a Reinforcement Learning (RL) based IDS to detect various attacks on RPL in 6LoWPANs, including several unaddressed by current research. The proposed scheme can also detect previously unseen attacks and the presence of mobile intruders. The scheme is well suited to the resource constrained environments of our target networks.

In this paper we consider such HAPS-based cellular systems to provide IoT coverage for protected area.

Challenges such as class imbalance, time intensive visual scoring, and limited amounts of labeled data are often encountered while accessing lung cancer screening low-dose computed tomography (LDCT) data for automated emphysema detection. To tackle these issues, we propose a generative adversarial network (GAN) based on unsupervised anomaly detection (UAD) to identify emphysema in LDCT. And to initiate disease specific feature learning, we introduce data augmentation based on minimum intensity projection (minIP) for the adversarial framework. The generator of the UAD model resembles an image latent-image configuration consisting of fully convolutional encoder-decoder architecture with skip connections. The discriminator is an encoder, which acts as a feature extractor and a classifier. We tested the robustness of the minIP based UAD model by adding anomalies to the training set. Furthermore to illustrate the usefulness of minIP and to serve as a comparison methodology for emphysema detection in LDCT, the proposed model was compared to the 2.5D transfer learning model. Visualization techniques like post processed residual images and Grad-cam maps were used to explain the UAD model inference. Our proposed minIP based UAD model showed an area under the curve of 0.91 0.02 for emphysema detection in LDCT. The UAD model was robust to inclusion of anomalies up to 3%. The augmentation improved the sensitivity of the transfer learning model by 2% however the UAD model outperformed the transfer learning model. The UAD model with minIP is a potential computer-aided tool for early detection of emphysema in lung cancer screening data.

This study introduces a model to solve a dynamic network optimization model on a heterogeneous graph. We use this model to optimize the collection and consolidation operations on a cross-country multi-modal distribution network. The model’s dynamic objects are trucks, trailers, orders, unvisited collection and customs check points. Information about dynamic objects is extracted from a real-time database. The model’s static objects include objects that are known in advance, such as warehouses. The constraints of the problem include due dates, vehicle capacity, availability of vehicles, and precedence constraints of visiting locations. We propose a mixed-integer programming model and provide a solution using matheuristics. We decompose the master MIP model into subproblems that can be solved to optimality with LP solvers. We also reduce the graph complexity by variable fixing due to optimized subproblems or by bounding the maximum number of paths to be selected due to the solutions of priority-based bin packing algorithms. Finally, we convert the resulting problem into a bipartite matching problem by expanding the graph nodes which can then be solved in polynomial time. We implement our solution method on real-time data retrieved from the tracking system of a third-party logistics company. Experiments show that our solution method significantly outperforms other heuristics in terms of solution quality which is measured with respect to lateness, empty kilometers traveled, travel times, number of required/used vehicles, load factors, and ratio of served orders.

Deep learning for procedural level generation has been explored in many recent works, however, experimental comparisons with previous works are rare and usually limited to the work they extend upon. This paper’s goal is to conduct an experimental study on four recent deep learning procedural level generators for Sokoban to explore their strengths and weaknesses. The methods will be bootstrapping conditional generative models, controllable & uncontrollable procedural content generation via reinforcement learning (PCGRL) and generative playing networks. We will propose some modifications to either adapt the methods to the task or improve their efficiency and performance. For the bootstrapping method, we propose using diversity sampling to improve the solution diversity, auxiliary targets to enhance the models’ quality and Gaussian mixture models to improve the sample quality. The results show that diversity sampling at least doubles the unique plan count in the generated levels. On average, auxiliary targets increases the quality by 24% and sampling conditions from Gaussian mixture models increases the sample quality by 13%. Overall, PCGRL shows superior quality and diversity while generative adversarial networks exhibit the least control confusion when trained with diversity sampling and auxiliary targets.

Predictive modeling involving brain morphological features and other covariates is of paramount interest in such heterogeneous mental disorders as PTSD. We propose a comprehensive shape analysis framework representing brain substructures, such as the hippocampus, amygdala, and putamen, as parameterized surfaces and quantifying their shape differences using an elastic shape metric. Under this metric, we compute shape summaries (mean, covariance, PCA) of subcortical data and represent individual shapes by their principal scores under a shape PCA basis. These representations are rich enough to allow visualizations of full 3D structures and help understand localized changes. Subsequently, we use these PCs, the auxiliary exposure variables, and their interactions for regression modeling and prediction. We apply our method to data from the Grady Trauma Project (GTP), where the goal is to predict clinical measures of PTSD. The framework seamlessly integrates accurate morphological features and other clinical covariates to yield superior predictive performance when modeling PTSD outcomes. This approach reveals considerably greater predictive power under the elastic shape analysis than the current approaches and helps identify local deformations in brain shapes associated with PTSD severity.

By their very nature, Spin Waves (SWs) with different frequencies can propagate through the same waveguide without affecting each other, while only interfering with their own species. Therefore, more SW encoded data sets can coexist, propagate, and interact in parallel, which opens the road towards hardware replication free parallel data processing. In this paper, we take advantage of these features and propose a novel data parallel spin wave based computing approach. To explain and validate the proposed concept, byte-wide 2-input XOR and 3-input Majority gates are implemented and validated by means of Object Oriented MicroMagnetic Framework (OOMMF) simulations. Furthermore, we introduce an optimization algorithm meant to minimize the area overhead associated with multifrequency operation and demonstrate that it diminishes the byte-wide gate area by 30% and 41% for XOR and Majority implementations, respectively. To get inside on the practical implications of our proposal we compare the byte-wide gates with conventional functionally equivalent scalar SW gate based implementations in terms of area, delay, and power consumption. Our results indicate that the area optimized 8-bit 2-input XOR and 3-input Majority gates require 4.47x and 4.16x less area, respectively, at the expense of 5% and 7% delay increase, respectively, without inducing any power consumption overhead. Finally, we discuss factors that are limiting the currently achievable parallelism to 8 for phase based gate output detection and demonstrate by means of OOMMF simulations that this can be increased 16 for threshold based detection based gates.

Data driven approaches have been widely employed in recent years to detect electricity thefts. Although many techniques have been proposed in the literature, they mainly focus on electricity thefts by consumers of power from the grid. Existing studies do not consider electricity thefts by prosumers, who act as both supplier and consumer in the energy system. This is of great importance as inaccurate reports of prosumers’ behaviours can disturb power system operation. Here, the paper examines the role prosumers may play in subverting the energy system and propose a novel means of detecting such malfeasance. Specifically, this work introduces a new electricity theft attack scenarios called balance attacks, where an attacker concurrently modifies his readings along with neighbouring meters in an attempt to balance the total aggregated reading. Such attacks can be difficult to detect by existing solutions that reach detection decisions based on aggregated readings. A novel electricity theft detector is proposed that can detect thefts in the presence of prosumers. Current approaches use either a single model for all users across the system or else a model for each user. Here, a half-way house approach is adopted where a cluster-based detection model is used. In each cluster, the power time series for a user is decomposed into trend, cyclical and residual components. Residual data, along with different features from multiple data sources, are fed in an ML classification algorithm to detect anomalous readings. Simulations have been conducted using a newly generated dataset and results have shown that the proposed model can detect electricity theft with high detection and low error rates. The results also shows that the proposed model can detect thefts with great accuracy from new users.

Please cite this work as: Huan Zheng, Zhao Zhang, Haijun Zhang, Yi Yang, Shuicheng Yan and Meng Wang, “Deep Multi-Resolution Mutual Learning for Image Inpainting,” In: Proceedings of the 30th ACM International Conference on Multimedia (ACM MM), Lisbon, Portugal, June 2022.

Deep data analysis for latent information prediction has been an important research area. Many of the existing solutions have used the textual data and have obtained an accurate results for predicting users’ interests and other latent attributes. However, little attention has been paid to visual data that is becoming increasingly popular in recent times. In this paper, we addresses the problem of discovering the attributed interest and of analyzing the performance of the automatic prediction using a comparison with the self assessed topics of interest (topics of interest provided by the user in a proposed questionnaire) based on data analysis techniques applied on the users visual data. We analyze the content of each user’s images to aggregate the image-level users’ interests distribution in order to obtain the user-level users’ interest distribution. To do this, we employ the pretrained ImageNet convolutional neural networks architectures for the feature extraction step and to construct the ontology, as the users’ interests model, in order to learn the semantic representation for the popular topics of interests defined by social networks (e.g., Facebook). Our experimental studies show that this analysis, on the most relevant features, enhances the performance of the prediction framework. In order to improve our framework’s robustness and generalization with unknown users’ profiles, we propose a novel database evaluation. Our proposed framework provided promising results which are competitive to state-of-the-art techniques with an accuracy of 0.80.

Space-air-ground integrated network (SAGIN) has emerged as a paradigm shifting architecture that offers large-scale, flexible wireless coverage and seamless, high-rate connectivity to complement terrestrial communications. Nonetheless, unlocking the potentials of SAGIN is subject to addressing non-trivial challenges associated with their inherent time-variability, self-organization, and heterogeneity. Meanwhile, the concept of reconfigurable intelligent surfaces (RISs) is recognized as a disruptive technology that supports an unprecedented control of electromagnetic waves propagation and potentially offers significant enhancements in spectral efficiency, coverage expansion, and security, all achieved in a low-cost and energy-efficient manner. We anticipate that the integration of RISs into SAGIN will not only play a fundamental role in improving the quality of inter- and intra- layer communications, but will also provide complex interactions among the three network segments, and hence, opens the horizons for a new breed of applications across all industries. In this article, we explore the underlying opportunities and challenges of integrating RIS-enabled communications into SAGIN, and present a forward-looking overview of the cross-layer operational elements supported by RISs. Finally, we outline major enabling technologies and present a look ahead toward addressing open research issues.

This paper deals with a method of demodulation for oversampled LoRa signal. The usual maximum likelihood (ML) based demodulation method for LoRa chirp spread spectrum (CSS) waveform is dedicated to signals sampled at Nyquist rate, whereas considering oversampled signals may improve the performance of the LoRa demodulation process. In this respect, when an oversampling rate (OSR) 2 is assumed, the method suggested in this paper consists in applying two demodulation processes to the even and odd samples of the oversampled LoRa signal, and then combining the results. This principle is then generalized to any OSR, and we show that the complexity of the method is low since it only involves discrete Fourier transforms (DFT). Moreover, a performance analysis in terms of symbol and bit error rate (SER and BER) is presented considering both additive white Gaussian noise (AWGN) and Rayleigh channel models. Simulations show the relevance of the method and the performance analysis as a gain of 3 dB is achieved by the demodulation at OSR 2 compared with OSR 1.

Any predictive software deployed with this hypothesis that test data distribution will not differ from training data distribution. Real time scenario does not follow this rule, which results inconsistent and non-transferable observation in various cases. This makes the dataset shift, a growing concern. In this paper, we’ve explored the recent concept of Label shift detection and classifier correction with the help of Black Box shift detection(BBSD), Black Box shift estimation(BBSE) and Black Box shift correction(BBSC). Digits dataset from ”sklearn” and ”LogisticRegression” classifier have been used for this investigation. Knock out shift was clearly detected by applying Kolmogorov–Smirnov test for BBSD. Performance of the classifier got improved after applying BBSE and BBSC from 91% to 97% in terms of overall accuracy.

Recent results in language processing show that language models are capable of performing several natural language tasks without the need of supervised learning. A challenging task for pre-trained language models is dialogue summarization. One way of generating summaries is engineering prompt templates for few-shot training. However, a static approach of creating prompts leads to unreliable outcomes between different classes of dialogues. Focusing on the dialogues structure properties, we propose a scoring system to improve the few-shot training performances by building tuned prompts composed by the highest scored dialogue samples. Our evaluation based on ROUGE scores and human evaluation shows that there is improvement for the experiments where we made use of the score system. The positive results are validated by all the three large-scale datasets we used in testing. All experiments were performed within the framework of the GPT-3 API.