Radar-based human activity recognition (HAR) is a popular area of research. In this paper, we investigate methods to improve the generalization of micro-Doppler-based swimming activity recognition. We identify three main challenges to this task: a small dataset lacking motion diversity, inaccurate period estimation, and inefficient network design that does not take into account the unique characteristics of spectrograms. To address the limited motion diversity, we propose a spectral data augmentation tailored for micro-Doppler spectrograms, including positive augmentations that account for physical fidelity and negative augmentations that penalize the unrealistic examples. We also investigate self-supervised pre-training to effectively use these negative augmentations. To address inaccurate period estimation, we introduce a segmentation approach based on energy distribution to handle temporal period variation. To exploit the spreading pattern of limb motion in the Doppler dimension and the continuous properties of torso motion in the temporal dimension, we design a module consisting of both 2D convolution and 1D temporal dynamic convolution to serve as a feature extractor. Our evaluation on a self-collected swimming activity recognition dataset shows that our model achieves the best classification accuracy and robustness to corruptions, even compared to much larger models and multi-domain fusion models.

A document by Federica Bellizio . Click on the document to view its contents.

Effective model compression plays a pivotal role in mitigating the computational and interpretational challenges inherent in the domain of time series forecasting. In this study, we introduce an innovative data-centric methodology tailored to identify a representative data subset from the entirety of the dataset. This chosen representative segment forms the cornerstone for the training of proficient time series forecasting models. Furthermore, our investigation unveils a compelling outcome of this approachâ\euro”a substantial reduction in the size of time series forecasting models when trained with this selected representative data segment. This model compression strategy results in a remarkable 56.31% decrease in storage consumption, a discovery of considerable significance for optimizing resources and enhancing scalability in time series forecasting. By distilling the dataset to its fundamental components through our data-centric approach, we aim to enhance both computational efficiency and the interpretability of the resultant models. This paper introduces a pioneering technique to tackle the challenges associated with data volume and model complexity in the field of time series forecasting, offering potential pathways for more efficient and insightful modelling in this domain.

The kVA limit of a power electronic converter (PEC) is often limited to keep the junction temperature Tj within a safe operating area (SOA). Therefore, controlling Tj with advanced algorithms such as Active Thermal Control (ATC) methods is a promising way to increase the kVA limit of PEC without sacrificing their reliability, which has not been explored enough in the literature. Therefore, this paper presents a novel approach combining the discontinuous pulse width modulation (DPWM) strategy with a system-level current controller. The proposed method enables the grid-connected inverter-based resources (IBR) to increase its kVA limit without increasing the Tj. This method also allows the converters to operate at higher ambient temperature Ta for more efficient power transfer without sacrificing its power handling capacity. Additionally, this paper systematically explains the effect of DPWM on switching losses Psw by establishing an analytical relationship between the bus clamp angle α with respect to Psw. Furthermore, for controlling the Tj, a simple Foster thermal model-based temperature estimation method is described, and its accuracy has been shown. Additionally, the impact of the proposed ATC method on maintaining the Tj of a 3Ph 2L-VSC with respect to the conventional sinusoidal PWM method is also compared.

We present a novel high-accuracy rotary magnetic sensor system composed of a two-track coded multi-pole magnet and a dual-spot multi-axes magnetic sensor. The novelty is the measurement of two orthogonal magnetic field components in each of the two sensing spots of the sensor, one associated with each magnet track. As the two orthogonal signals in each spot are naturally in quadrature, i.e. they represent a sine and a cosine signal, the measurement principle is virtually independent of the magnet pole size and pitch. We can therefore design a magnet with much larger poles which in turn generate stronger magnetic flux, allowing for an increased air gap between sensor and magnet. The calibrated encoder system was characterized to deliver 13 bits of absolute accuracy and 17 bits of resolution over the full 360° range.

In order for tokenized asset networks to be accountable as Web3 marketplaces for token-related transactions, identity verification must be conducted by gateways into those token networks. This includes the identity validation and legal status verification of the originators and beneficiaries, the gateway owners/operators, and other relevant service providers. The classic identity provider model could be enhanced to support anti-money laundering regulations, notably the Travel Rule. A privacy-preserving IdP model in combination with a legal service provider is explored where the IdP becomes the issuer of a blinded attestation regarding the user attribute, and where the legal representative with attorney-client privilege becomes the first point of contact for requests for the disclosure of the blinded attestations.

A document by Tianyi Wang . Click on the document to view its contents.

In the rapidly evolving landscape of software development, Artificial Intelligence (AI) emerges as a pivotal force, reshaping traditional methodologies and introducing groundbreaking efficiencies. This research delves into the integration of AI in software development, aiming to elucidate its transformative impact on coding practices, project management, and quality assurance. We commence with a comprehensive literature review, highlighting the historical progression and current state of AI applications in software development. Our methodology encompasses a blend of qualitative and quantitative analyses, including case studies of AI-driven projects and comparative evaluations of AI tools in real-world scenarios. The core of our research focuses on key AI technologies such as machine learning algorithms and natural language processing, examining their roles in automated code generation, testing, and debugging. The results reveal significant enhancements in efficiency, accuracy, and complexity management, attributable to AI integration. However, our discussion also addresses the challenges and ethical considerations that accompany AI adoption, emphasizing the need for balanced and responsible approaches. Looking forward, we explore emerging trends and potential future advancements in AI, proposing avenues for further research. This study not only underscores the transformative potential of AI in software development but also serves as a guiding framework for practitioners and researchers in navigating this dynamic field. Â

Radar-based human activity recognition (HAR) is a popular research field. Despite claims of high accuracy on self-collected datasets, the ability of these models to handle unexpected scenarios has been largely overlooked. This work introduces a framework for analyzing corruption robustness of radar micro-Doppler spectrogram classification. A set of corruptions are categorized, applied, and systematically tested on common model architectures. Diverse training methods, including adversarial training, cadence velocity diagram (CVD) transformation and data augmentation, are explored. The performance is evaluated on two tasks: indoor HAR and continuous aquatic HAR. Our study unveils several insights. Firstly, relying solely on accuracy may not adequately assess model performance due to dataset limitations. All well-trained models exhibit sensitivity to corruptions. Secondly, deeper convolutional neural network (CNN) models excel in both accuracy and robustness, but confront the problem of overfitting to background. Thirdly, adversarial training enhances robustness against corruptions, albeit at the cost of a slight decrease in accuracy. Lastly, combining data augmentation and adversarial training achieves a balance between accuracy and robustness. In essence, our study contributes to a more profound understanding of the complex interplay between model architecture, classification accuracy, and corruption robustness in radar HAR tasks.

Multi-objective hyperparameter optimization (MO-HPO) is applied to find optimal artificial neural network (ANN) architectures used for optimal feedforward torque control (OFTC) of synchronous machines. The proposed framework allows to systematically identify Pareto optimal ANNs with respect to multiple (partly) contradictory objectives such as approximation accuracy and computational burden of the considered ANNs. The obtained Pareto optimal ANNs are trained and implemented on a realtime system and tested experimentally for a nonlinear reluctance synchronous machine (RSM) against non-Pareto optimal ANN designs and a state-of-the-art OFTC approach. Finally, based on the most recent results from ANN approximation theory, guidelines for Pareto optimal ANN-based OFTC design and implementation are provided.Â

In-depth evaluations of an electric drive's behavior typically result from co-simulation, combining models of motor, power electronics and control software. This approach can be used to evaluate pulse patterns for a given operating point, e.g., for electric vehicle applications, preventing real-world experiments. Here, the co-simulation model is fed with offline calculated optimized pulse patterns (OPPs) that are used to increase, among others, the drive's efficiency. Due to the large motor time constant compared to the fundamental wave period, reaching steady state using a simple open-loop control requires unnecessary long simulation times. Hence, a model predictive closed-loop control implementation of the OPPs is proposed which reduced the overall computational effort significantly. However, it turns out that the OPP evaluation using a finite-element-method-based co-simulation for a permanent magnet synchronous motor remains largely uncertain in terms of the predicted power losses which led to the development of a semi-analytical model to further reduce the required computational time. As none of the approaches was able to deliver a suitable trade-off between model accuracy and calculation time, this investigation highlights the remaining challenges of evaluating OPPs based on drive co-simulations motivating further research towards the surrogate-assisted or direct experimental OPP optimization in the future.

Microservice-based architectures enable different aspects of web applications to be created and updated independently, even after deployment. Associated technologies such as service mesh provide application-level fault resilience through attribute configurations that govern the behavior of request - response service -- and the interactions among them -- in the presence of failures. While this provides tremendous flexibility, the configured values of these attributes -- and the relationships among them -- can significantly affect the performance and fault resilience of the overall application. Furthermore, it is impossible to determine the best and worst combinations of attribute values with respect to fault resiliency via testing, due to the complexities of the underlying distributed system and the many possible attribute value combinations. In this paper, we present a model-based reinforcement learning workflow towards service mesh fault resiliency. Our approach enables the prediction of the most significant fault resilience behaviors at a web application-level, scratching from single service to aggregated multi-service management with efficient agent collaborations.

In the evolving landscape of the IT industry, the dichotomy between Work-From-Office (WFH) and Return-To-Office (RTO) models has become a focal point, especially in IT Consulting and Software Development. This study embarks on a comparative analysis of these models in the United States and Germany, delving into their impacts on productivity, commuting, stress, and the economic costs associated with maintaining office spaces. The research is grounded in a comprehensive literature review that explores the historical and current work models in IT, assesses productivity metrics, analyzes commuting patterns, and evaluates the influence of work-related stress on IT professionals. Additionally, it considers the economic implications of office spaces, a factor critical to organizational decision-making. Employing a case study methodology, the research scrutinizes an IT consulting firm in the US and a software development company in Germany. These case studies are instrumental in examining the multifaceted aspects of WFH and RTO models. The productivity analysis is conducted through quantitative measures like project completion rates and qualitative assessments from employee feedback. Commuting impacts are evaluated in terms of time, cost, environmental footprint, and employee satisfaction. Stress levels are measured through well-being surveys and turnover rates, providing insights into the psychological impacts of different work models. A pivotal aspect of the study is the economic analysis of office costs, encompassing real estate expenses, utilities, maintenance, and the potential savings from remote work models. The comparative analysis aims to draw parallels and contrasts between the US and German contexts, highlighting how cultural and economic factors shape the adoption and effectiveness of WFH and RTO models. It also explores the environmental considerations of commuting in the IT sector and the complex interplay between work models, stress, job satisfaction, and economic efficiency. The discussion section extrapolates the broader implications of these findings, offering policy recommendations for IT companies navigating the post-pandemic work environment. It addresses the challenges of balancing productivity with employee well-being and the economic realities of office maintenance. In conclusion, the study synthesizes the findings from the case studies, advocating for a flexible, economically viable approach that harmonizes productivity with employee well-being and company costs. It underscores the necessity for IT companies to adopt adaptable work models that consider not only the productivity and well-being of employees but also the economic realities of office maintenance. The study also highlights the need for ongoing research to understand the long-term economic impacts of WFH and RTO models on the IT industry.

It’s important to note that the classification of epileptic seizures can be complex, and an individual may experience more than one type of  seizure. Additionally, advancements in understanding epilepsy may lead  to refinements in the classification system over time. If someone is  experiencing seizures, it’s crucial for them to consult with a  healthcare professional for a proper diagnosis and appropriate  treatment.Â

A document by Shilian Zheng . Click on the document to view its contents.

Named Entity Recognition (NER) task aims at automatically recognising and classifying named entities in a given natural language input. Majority of the studies related to Named Entity Recognition are focused on English language. Named Entity Recognition in Indian languages is challenging due to the complex grammar and morphology of Indian languages and the scarcity of good quality labelled data. The dissimilarity between the literary and spoken versions of these languages is also a big challenge regarding usability of NER models. Kannada is such an Indian language for which the task of NER is still an active area of research. Usage of Deep learning, especially Transfer learning using Transformer models has drastically improved performance of many NLP tasks at scale. However, Transfer learning still requires considerable amount of data for the required task. In case of low resource language like Kannada, very few labelled datasets are available publicly and creating one from scratch is expensive in terms of time and labor. Active learning (AL) aims to tackle the labelled data acquisition problem by having the learning model and an Oracle to cooperate. Active Learning iteratively builds an optimally labelled and sufficiently large dataset.  This study focuses on Named Entity Recognition in Kannada language. We explore the application of Active Learning technique to Named Entity Recognition problem in the low-resource language Kannada. Results show that AL can be used to boost a multilingual models performance in fine-tuning for NER. We also try to mitigate the gap between formal and colloquial dialects of Kannada in NER datasets.

There have been significant efforts devoted to developing advanced mixed-integer linear programming (MILP) solvers, which are powerful tools for solving various real-world optimization problems. Despite the achievements, the limited availability of real-world instances often results in sub-optimal decisions and biased evaluations, which motivates a suite of MILP instance generation techniques. However, these approaches either rely on expert-designed formulations or struggle to capture the rich features of real-world instances. Moreover, the task of generating challenging MILP instances—which are valuable resources for evaluating solvers and motivating more efficient algorithms—remains underexplored. To tackle these problems, we propose G2MILP, which to the best of our knowledge is the first deep generative framework for MILP instances. Specifically, G2MILP represents MILP instances as bipartite graphs and employs a masked variational autoencoder to iteratively corrupt and replace parts of the original graphs to generate new ones. We then propose a hardness-oriented scheme, which iteratively augments the generator by learning from the hardest instances, to enhance G2MILP to construct challenging MILP instances. Experiments demonstrate that G2MILP can generate realistic MILP instances to effectively facilitate downstream tasks. Moreover, G2MILP can generate difficult instances initializing from given datasets, and the boost of hardness can be orders of magnitude.

Since tokenization serves a fundamental preprocessing step in numerous language models, tokens naturally constitute the basic embedding units for generative linguistic steganography. However, token-based methods encounter challenges including limited embedding capacity and possible segmentation ambiguity. Despite existing character-level linguistic steganographic approaches, they neglect the problem of generating unknown or out-of-vocabulary words, potentially compromising steganographic imperceptibility. In this letter, we focus on both embedding capacity and imperceptibility for a tokenization-free linguistic steganographic approach. Firstly, we suggest that unknown words mainly stem from low-entropy distributions and rigid coding rules within candidate pools, and thus we propose an entropy-based selection approach to flexibly construct candidate pools. Further, we present a lexical emphasis approach, prioritizing characters within candidate pools capable of forming in-vocabulary words. Experiments show that, across a range of high embedding rates, our approaches achieve considerably higher imperceptibility and about 17% higher anti-steganalysis capacity than the baseline method without our approaches.