We investigate the growth dynamics of microalgae under intermittent light regimes whose frequency is modeled as a continuous random variable. We incorporated randomness through the frequency of the light/dark cycles by extending the deterministic Photosynthetic Factory model in a Random Differential Equation framework. This is an important modeling step because there is experimental evidence that this frequency is random. We derive the probability density functions for key model variables: the proportion of active photosynthesis units at a given time and the average growth rate. Moreover, we study the effects of the light-dark ratio, showing that when the system spends 9 times more time in the dark than in the light, even the high-frequency choice cannot perform near the growth in the ideal continuous light experiment. Such randomized modeling approaches could lead to more comprehensive insights into the interplay between mixing-induced intermittent lighting and microalgal growth, ultimately guiding the design of more energyefficient photobioreactors in industrial applications.

Five-phase synchronous machines enable increased torque, power density, and reliability, but the increased number of degrees of freedom makes the optimization of current waveforms a complex task. In this paper, the optimal current waveforms for the field weakening region are proposed by solving the optimization problem. Since the arrangement of the rotor magnets can affect the shape of the back emf, the comparison of optimal solutions for different 1 st and 3 rd harmonic of the rotor magnetic flux is performed. The obtained results are verified by experiments on a laboratory prototype of the five-phase permanent magnet synchronous machine.

Especially in noisy intermediate-scale quantum (NISQ) devices, fault-tolerant quantum computing cannot be realized without quantum error correction. High error rates in quantum systems stemming from decoherence and gate defects call for effective and flexible QEC systems. This paper presents a novel FPGA-based architecture using dynamic reconfiguration to maximize hardware resources in response to real-time error patterns. Supporting bit-flip, phase-flip, and depolarizing faults among several error correction techniques guarantees high accuracy and little hardware overhead. Scaled for surface code distances up to and beyond, the design combines modular pipelines, adaptive hardware allocation, and error pattern detection. Designed on a Xilinx XC7A50T-CPG236 FPGA, the system showed notable gains over static architecture, averaging 30% lower latency and 20% less resource use overall. Low latency and high throughput are maintained while this dynamic technique enables flexible adaptability to several quantum error models. The outcomes show how well this structure may improve quantum error-correcting systems and enable the incorporation of QEC into useful quantum computer configurations.

Generative modeling has seen significant advancements with the rise of diffusion models, which leverage iterative denoising processes to generate high-quality data. Originally developed for image synthesis, these models have now been extended to a wide range of domains, including text, audio, and video. Beyond their generative capabilities, diffusion models have also demonstrated strong potential for representation learning, enabling the extraction of meaningful features useful for downstream tasks. This survey provides a comprehensive review of diffusion models in the context of representation learning. We first introduce the theoretical foundations and key architectural advancements, such as improved noise schedules, latent diffusion, and conditional modeling. Next, we examine their role in self-supervised and contrastive learning, highlighting how the denoising process facilitates structured feature representations. We also explore the integration of diffusion models with other generative frameworks, including VAEs and GANs, to enhance efficiency and representation quality. Despite their growing success, diffusion models face several challenges, including high computational costs, slow inference, and limited theoretical understanding. We discuss ongoing efforts to address these limitations, including accelerated sampling techniques, model compression strategies, and the development of more interpretable representations. Additionally, we highlight ethical concerns, such as bias, robustness, and the potential misuse of generated content. Finally, we outline promising research directions, including multimodal and crossdomain learning, improved integration with large-scale language models, and real-world deployment considerations. By synthesizing recent developments, this survey aims to provide a structured perspective on the evolving role of diffusion models in both generative modeling and representation learning.

Electricity service interruptions compromise households' ability to meet energy demand, and thus increase energy insecurity. Interruptions can take the form of power outages, or involuntary disconnections due to nonpayment. To achieve equitable electricity access, it is crucial to understand how electricity loss varies across demographics. However, utilities often report outages at the system or county level and face few incentives to report disconnection metrics, making spatially granular analysis of interruptions challenging. This paper presents a new method of quantifying electricity reliability by classifying unlabeled advanced metering infrastructure (AMI) data, hence disaggregating power outages from disconnections in residential electricity consumption data. We use the Commonwealth Edison utility service territory in Illinois as a case study and evaluate how the duration and frequency of power outages (outage burden) change across ZIP codes. We find that the most outage-burdened regions are concentrated in rural counties, and regions with a higher percentage of low-income residents and people of color often carry higher outage burdens than the rest of the service territory. Our proposed method is the first to disaggregate AMI data into outages and disconnections, enabling more granular metrics for electricity reliability, highlighting gaps in energy accessibility and informing the identification of energy-insecure communities.

Non-contact manipulation of floating objects at the air-liquid interface holds significant potential applications in microrobotics, non-invasive microassembly, and lab-on-a-chip systems. However, achieving simultaneous position and orientation (pose) control of floating objects remains a considerable challenge. Here, we present the Vision-based Adaptive Laser Gripper (VALG) system to address these challenges. A modelfree pose control strategy based on equidistant contour scanning lasers (laser grippers) that leverage distributed thermocapillary flow induced by patterned laser scanning is proposed. The adaptive laser gripper can be generated within seconds and relies solely on visual recognition, making it applicable to a wide range of floating objects without the need for prior modeling or additional information. The effectiveness and control precision of the VALG system is experimentally validated through discrete pose verification and continuous pose control tests. The maximum position and orientation errors are 0.05 mm and 2.6 • for discrete pose verification, and 0.4 mm and 14.5 • for continuous pose control, respectively. The VALG system enables smooth, precise, fast, and adaptive pose control of floating objects, establishing it as a universal and versatile platform for noncontact manipulation at the air-liquid interface.

Congestion due to morning and evening traffic peaks has caused economic losses amounting to billions of dollars annually. Thus, accurately predicting the departure time of commuters is of interest to transportation planners, engineers, and elected officials alike. We develop a statistically-informed deep learning approach to improve commuter departure time prediction models. Specifically, we leverage elements of the proportional hazards model, a class of time-to-event prediction approaches, to augment vanilla deep neural network (DNN) architectures. The proposed approach also employs collaborative filtering to segment the commuter population into distinct behavioral classes, allowing tailored predictions for specific commuter profiles. We find that our class of survival analysis-enhanced DNNs outperforms conventional neural networks in predicting trip departure times, while also offering more interpretability through the hazard coefficients.

Traditional feature selection methods focus on reducing dimensionality, often at the cost of discarding valuable but hidden information. In contrast, this paper introduces Feature Expansion, a novel approach that dynamically identifies and reconstructs missing features, enhancing predictive performance without requiring additional data collection. We propose the Semi-Dynamic Feature Set (SDFS) Network, a unique machine learning framework that learns both feature values and model parameters, thereby improving robustness in complex datasets. To enhance interpretability, we integrate Large Language Models (LLMs) to assign meaningful labels to the newly discovered features, transforming them from black-box variables into actionable insights. Our proposed feature expansion method consistently improved classification performance across multiple datasets. In particular, the accuracy increased from 89% to 95% for Parkinson's Disease Detection and from 82% to 85% for Wine Quality Prediction, demonstrating its effectiveness in discovering missing features. By combining feature expansion with LLM-driven explainability, this method not only enhances model accuracy but also unveils hidden data patterns in critical domains such as healthcare, finance, and scientific research. The full implementation is available on GitHub [GitHub link], and the model can be installed via PyPI under the package 'sdfs' [PyPI link].

Sparse arrays offer a promising solution for reducing the data volume required to reconstruct ultrasound images, making them well-suited for portable and wireless devices. However, the quality of images beamformed from a limited number of transducer elements is significantly degraded. This study proposes a deep learning-based inpainting technique to estimate complete radio frequency (RF) signals (before beamforming) from downsampled RF signals obtained using a subset of transducer elements. This approach allows for quality enhancement without the need for beamforming images, offering flexibility particularly beneficial for applications such as speed-of-sound estimation algorithms. We introduce a model-based loss function that combines the signal and image domains by incorporating a measurement model associated with image reconstruction. We then compare this loss with another that accounts solely for the RF signal domain. Additionally, we train our network exclusively in the RF image domain, mapping images beamformed from downsampled RF signals to those from complete signals. We compare these approaches qualitatively and quantitatively, with all enhancing image quality. The proposed method with the modelbased loss achieves superior detail and quality metrics. Although trained on downsampled RF signals simulating sparse arrays in reception, all methods-especially our inpainting approach with the model-based loss-demonstrate strong adaptability to ultrafast acquisitions with reduced transducer elements in both transmission and reception. This highlights their potential to reduce the number of transducer elements in ultrasound probes. Furthermore, the proposed method exhibits superior generalization performance when evaluated on a different probe than the one used to acquire the training dataset.

In recent years, several datasets containing trajectories of road users have been published, providing valuable insights for the analysis and modeling of traffic participant behavior. However, road user trajectories from highway datasets are often limited to lengths of less than 2.5 km, restricting the analysis of consecutive traffic scenarios, such as multiple lane changes. To address this gap, we introduce the DLR Highway Traffic dataset, the longest road user trajectory dataset from a German highway. This dataset contains 38,209 trajectories, along with local weather and road condition data collected over a period of 10 h at the Testbed Lower Saxony. A comparison with other publicly available datasets reveals that our dataset, with trajectories reaching up to 6,428 m in length, contains the longest trajectories from German highways, enabling the analysis of long-duration traffic scenarios. With a total of 143,371 km, our dataset is approximately three times larger than the largest existing German highway dataset, the highD dataset, which covers 44,500 km. However, it is 28 times smaller than the largest highway dataset, the I-24 MOTION dataset, which covers approximately 4,050,000 km. In contrast, our dataset stands out by including additional raw data beyond just trajectories, such as locally recorded weather data and road condition data. Furthermore, the traffic volume data, derived from the trajectory data, provide valuable insights into traffic flow. Additionally, the trajectory data are available in OpenSCENARIO format, facilitating the visualization and simulation of traffic scenarios. Overall, the dataset provides valuable resources for researchers seeking to conduct datadriven behavior modeling. It is available for non-commercial use and can be directly downloaded from https://doi.org/ 10.5281/zenodo.14811064.

The search for a deeper understanding of Dark Matter and Baryonic Matter interactions, has led to the experimental observations of Dark Matter and Baryonic Matter relationships, characteristics, and parameters using Quantum-Wave Computation. With the repeated observation of these interactions, we sought to gain a deeper understanding of the visual underpinning of Dark Matter, to attempt novel approaches at conceptualizing Dark Matter, as well as possible interactions with Baryonic Matter for the human-eye. This effort led to unique experiments covering various attempts, failures, and success, exploring cutting-edge methodologies towards uncovering imagery of dark matter. The photography of the quantum-level interaction, is a repeatable dimensionality of reality, in all its complexity and simplicity, as we expose its ways in the following experimental results.

Automatic detection of early-stage electrical faults in motors is challenging because physical signals, such as temperature or vibrations, are often linked to regular machine operation. Inter-turn short-circuit (ITSC) faults in the stator winding are a leading cause of irreversible damage. Given industrial demands, there is an ongoing need for innovative approaches that reduce dimensionality and expedite feature extraction from three-phase current signals. This work presents an ITSC multi-fault classification algorithm by modeling the behavior of three-phase stator currents in a complex space. To achieve this, two geometric-based and two optimization-based approaches are employed to parameterize the shape of the complex signal. Rather than directly extracting features from the raw signals, these parameters capture the similarities among incipient failures. The algorithm is evaluated using machine learning classifiers trained on a dataset generated from an experimental test bench and a publicly available dataset. The proposed method achieved an accuracy of 95.30% across 13 categories, demonstrating its robustness and reliability and positioning it as a highly competitive alternative to state-of-the-art techniques.

This article expands upon previous peer-reviewed research, presenting novel evidence of Particle 11's stabilization at room temperature—a groundbreaking discovery in particle physics. By leveraging sophisticated computational models and comprehensive signal analysis, this article details the experimental setups, methodologies, and the particle's potential in energy harvesting, alongside broader implications for quantum computing and global energy solutions.

Non-terrestrial networks (NTNs) are crucial for 6G communications but face challenges like classical communication bottlenecks. Semantic communication (SemCom) offers a solution, but efficient resource allocation remains unresolved. This paper proposes a joint approach for resource allocation and semantic information selection in low-earth orbit (LEO) satellitebased image transmission. A novel metric, Quality of Semantic (QoSem), is introduced to measure the SemCom performance. An optimization problem is formulated to maximize QoSem by allocating computing, power, and bandwidth resources and selecting optimal semantic information. Furthermore, we propose a novel deep reinforcement learning (DRL) algorithm, namely Mixed-Action Decision Intelligent Semantic-aware Soft-Actor Critic (MADIS-SAC), to handle the problem's nonconvexity and NP-hard complexity. Results show 82.5% of QoSem improvement compared to baselines under resource constraints.

Recent advancements in log classification have been significantly influenced by the growing adoption of artificial intelligence (AI) and machine learning (ML) techniques. The integration of large language models (LLMs) has further enhanced the efficiency and accuracy of log classification across various domains. This paper explores the strategies and aggregates a new frontier in log classification: Adaptive Log Classification [1], which builds on these innovations by incorporating dynamic, AI/ML-driven methods to continuously optimize log classification processes. Unlike traditional static approaches, Adaptive Log Classification adjusts to the evolving requirements of systems and environments, making it particularly well-suited for diverse use cases such as system, software, and network logs. This adaptive methodology is context-aware, tailoring the classification process to the specific needs of each environment. The dynamic nature of this approach is especially beneficial in industries where data landscapes are rapidly changing, such as cybersecurity, where log data may shift in response to new threats or attack patterns. The paper explores the application of Adaptive Log Classification in the domains of Edge AI and Internet of Things (IoT) log management, highlighting the role of autonomous agents in facilitating real-time adaptation. By leveraging AI-powered agents, this approach aims to enhance the efficiency, accuracy, and responsiveness of log classification, ultimately improving system performance and security in complex, evolving environments.

In this paper we continue our previous investigation about the use of stress function in the flow of generalized Newtonian fluids through conduits of circular and noncircular (or/and multiply connected) cross sections where we inspect the flow of Ellis fluids in tubes of elliptical cross sections. We derive analytical expressions for the flow velocity profile and for the volumetric flow rate. The obtained analytical expressions were tested against the available analytical expressions for the special cases of Newtonian flow in circular tubes, Newtonian flow in elliptical tubes and Ellis flow in circular tubes and the results were identical. The obtained analytical expressions were also tested for sensible trends, tendencies and correlations and they passed all these tests.

Tremor is the most common movement disorder and a prevalent symptom of neurodegenerative conditions such as Parkinson's disease (PD). Given the limitations of medication, which may not effectively treat tremor, and the limited availability of surgical treatments such as deep brain stimulation, there is a pressing clinical need for non-invasive therapeutic alternatives, including peripheral electrical stimulation. The high variability of PD tremor poses a challenge to such therapies and calls for patient-specific stimulation parameters. This study presents the design and validation of an adaptable algorithm for online tracking of Parkinsonian rest tremor phase, integrated into a wrist-worn system for use in triggering phase-specific peripheral electrical stimulation, delivered non-invasively to a nerve in the wrist. The system dynamically adapts to tremor variability, including shifts in the axis of maximum excursion and changes in center frequency. Validation was first performed offline, followed by online tests with a vibration plate providing stable frequencies, trials involving a healthy experimenter emulating tremor in different orientations while wearing the device, and evaluations with three individuals with PD. Results show accurate and robust phase estimation accounting for changes in tremor dominant axis and center frequency, underscoring the system's adaptability. This work provides a robust platform for research and a foundation for developing personalized, non-invasive tremor management strategies.