We present a novel AI-based ideation assistant and evaluate it in a user study with a group of innovators. The key contribution of our work is twofold: we propose a method of idea exploration in a constrained domain by means of LLM-supported semantic navigation of problem and solution spaces, and employ novel automated data input filtering to improve generations. We found that semantic exploration is preferred to the traditional prompt-output interactions, measured both in explicit survey rankings, and in terms of innovation assistant engagement, where 2.1x more generations were performed using semantic exploration. We also show that filtering input data with metrics such as relevancy, coherence and human alignment leads to improved generations in the same metrics as well as enhanced quality of experience among innovators.

Predicting user performance at the base station in telecom networks is a critical task that can significantly benefit from advanced machine learning techniques. However, labeled data for user performance are scarce and costly to collect, while unlabeled data consisting of base station metrics, are more readily accessible. Self-supervised learning provides a means to leverage this unlabeled data, and has seen remarkable success in the domains of computer vision and natural language processing, with unstructured data. Recently, these methods have been adapted to structured data as well, making them particularly relevant to the telecom domain. We apply self-supervised learning to predict user performance in telecom networks. Our results demonstrate that even with simple self-supervised approaches, the percentage of variance of the target values explained by the model, in low-labeled scenarios (e.g., only 100 labeled samples) can be improved fourfold, from 15% to 60%. Moreover, to promote reproducibility and further research in the domain, we open-source a dataset creation framework and a specific dataset created from it that captures scenarios that have been deemed to be challenging for future networks.

In light of the revolutionary requirements of the sixth generation (6G) and beyond wireless networks, reconfigurable intelligent surface (RIS) and rate-splitting multiple access (RSMA) have emerged as pivotal technologies due to their potential for improving spectral efficiency, user fairness, and interference management. This survey explores the theoretical foundations, architectural frameworks, and design strategies of RIS-assisted RSMA, emphasizing the combined adaptability of RIS's wireless propagation control and RSMA's multiuser flexibility for dynamic spectrum management. The article first discusses the fundamental concepts of RSMA and RIS technologies. Then, we investigate various enabling technologies for RIS-RSMA networks, highlighting key advancements in interference mitigation, energy efficiency, and security for future networks. Subsequently, some optimization techniques crucial for enhancing RIS-RSMA network performance are presented. Additionally, we examine advanced machine learning (ML) approaches that enable RIS configurations to dynamically adapt to changing network requirements. Techniques such as deep reinforcement learning support real-time adjustments, creating more scalable and resilient RIS-RSMA architectures. Finally, we discuss open research directions for advancing RIS-assisted RSMA in emerging 6G applications. We also consider the potential of advanced ML techniques, including quantum-based ML and large language models, to handle the complexities of large-scale network optimization. This comprehensive survey addresses critical challenges and current advancements. It offers a roadmap for future research in RIS-assisted RSMA networks, paving the way for robust, intelligent, and adaptive 6G wireless communication systems.

The ability to swiftly and efficiently assess information is essential for a successful response in the case of a natural disaster, such as wildfires. Access to accurate and timely information is critical for first responders and the public to make informed decisions within tight timelines. Although traditional sources of information such as satellite images, field surveys, and contact centers are costly and time consuming, social media content is more affordable and immediately available, making it a useful tool for real-time updates. However, extracting insights from the abundance of user-generated content on social media can be challenging. Machine learning can automatically categorize posts, filter noise and highlight crucial information for responders and the public. In this work, we introduce, WildFireCan-MMD, a new multimodal dataset tailored to Canadian wildfires, collected from user-generated posts on X during the 2022, 2023, and 2024 British Columbia and Alberta wildfires. We applied topic clustering followed by manual annotation to label this dataset into thirteen wildfire-relevant themes. We tested two Generative Large Vision-Language Models (GLVLMs), namely, gpt-4o-mini and open-source LLaVA, for the classification of WildFireCan-MMD in zero-shot setting. We also developed custom classifiers, including RoBERTa and ViT transformer combinations with early and late fusion. Our results show that if no training data is available, gpt-4o-mini might serve as a practical zero-shot classifier. However, with labeled training data, even a basic Gaussian Naive Bayes classifier outperforms gpt-4o-mini. Fine-tuned transformers outperform gpt-4o-mini by a large margin of 23%. This study underscores the value of curated, task-specific datasets and training customized classifiers despite recent advancements in zero-shot GLVLMs.

Multi-View Stereo (MVS) reconstructs detailed 3D structures from multi-view images by establishing spatial correspondences. While learning-based methods have significantly advanced the MVS task, challenges such as ambiguous matching caused by textureless surfaces and lighting variations persist. To address these issues, we propose GAP-MVSNet, a framework that leverages surface normals from a monocular normal foundation model as priors to enhance the geometric awareness of reconstruction targets. In this work, surface normal priors are seamlessly integrated into the MVS pipeline to improve depth prediction robustness and accuracy. Specifically, we introduce a structure-aware feature pyramid network that incorporates surface normal information and utilizes uncertainty-aware feature resampling to extract robust image features. Additionally, we present the spatial geometry enhanced regularization that combines sampled depth hypotheses with surface normals to generate a spatial geometric prior, guiding the cost regularization process and enforcing strong spatial coherence, particularly in textureless regions. Furthermore, we design a local consistency depth refinement module that utilizes surface normals to establish depth relationships as a local geometric prior, thereby refining classification-based depth predictions and aligning them with ground truth depth. Extensive experiments on the DTU and Tanks & Temples datasets demonstrate that our method achieves state-of-the-art performance.

Adversarial attacks on Natural Language Processing (NLP) models expose vulnerabilities by introducing subtle perturbations to input text, often leading to misclassification while maintaining human readability. Existing methods typically focus on word-level or local text segment alterations, overlooking the broader context, which results in detectable or semantically inconsistent perturbations. We propose a novel adversarial text attack scheme named Dynamic Contextual Perturbation (DCP). DCP dynamically generates context-aware perturbations across sentences, paragraphs, and documents, ensuring semantic fidelity and fluency. Leveraging the capabilities of pre-trained language models, DCP iteratively refines perturbations through an adversarial objective function that balances the dual objectives of inducing model misclassification and preserving the naturalness of the text. This comprehensive approach allows DCP to produce more sophisticated and effective adversarial examples that better mimic natural language patterns. Our experimental results, conducted on various NLP models and datasets, demonstrate the efficacy of DCP in challenging the robustness of state-of-theart NLP systems. By integrating dynamic contextual analysis, DCP significantly enhances the subtlety and impact of adversarial attacks. This study highlights the critical role of context in adversarial attacks and lays the groundwork for creating more robust NLP systems capable of withstanding sophisticated adversarial strategies.

Accurate loss estimation is fundamental to the design and maintenance of photovoltaic (PV) plants. Among the various types of losses affecting PV performance, this study focuses on mismatch losses (MsL). We introduce novel stochastic models to quantify MsL in PV strings and at the combiner box. Specifically, we leverage the statistics of the minimum of N Gaussian random variables to model the stochastic behavior of MsL under realistic PV generation scenarios. To this end, we derive both exact and approximate expressions for the expected value of this minimum. Furthermore, by adapting the proposed stochastic model for MsL in strings, we develop a new formulation to estimate MsL due to PV module degradation over time. The proposed models are numerically evaluated under different parameter configurations and compared with the deterministic approach. Additionally, the stochastic MsL model for strings is validated against the Monte Carlo (MC) method employed in PVsyst, while the MsL model for degraded PV modules is validated against real field measurements, demonstrating the applicability of the proposed approach. Notably, the results obtained from the stochastic models reveal that MsL at the string and combiner box levels depends on the number of PV modules and strings, respectively.

In order to manage electricity transmission and distribution it is now common practice for system operators to offer financial incentives that encourage large consumers to reduce energy usage during designated peak demand periods. For train operators on large rail networks it may be profitable-with selected individual journeys-to reduce energy usage during peak times and increase energy usage at other times rather than simply minimizing overall energy consumption. We will use classical methods of constrained optimization to find optimal driving strategies for a fleet of trains subject to limits on total energy consumption during specified intermediate time intervals but with no change to individual journey times. The proposed strategies can be used by a large rail organisation to reduce overall operating costs with only minimal disruption to existing schedules and with no changes to important departure and arrival times.

As minimum area SRAM bit-cells are obtained when using cell ratio and pull-up ratio of 1, we analyze the possibility of decreasing the cell ratio from the conventional values comprised between 1.5-2.5 to 1. The impact of this option on area, power, performance and stability is analyzed showing that the most affected parameter is read stability, although this impact can be overcome using some of the read assist circuits proposed in the literature. The main benefits are layout regularity enhancement, with its consequent higher tolerance to variability, cell area reduction by 25% (with respect to a cell having a cell ratio of 2), leakage current improvement by a 35%, as well as energy dissipation reduction and a soft error rate per bit improvement of around 30%.

The growing sophistication of cyberattacks has led to an urgent need for advanced techniques in malware detection and mitigation. Neural networks (NNs), with their powerful learning capabilities, have emerged as a promising tool for predicting malware propagation patterns in complex, dynamic environments. This research explores the application of neural network architectures, particularly deep learning models, to predict and analyze the spread of malware across network systems. By leveraging large datasets of network traffic, malware signatures, and attack vectors, the study employs supervised learning techniques to train NNs to identify key indicators of infection and transmission. Through experimentation with various NN models, including feedforward, recurrent, and convolutional networks, we assess their performance in predicting the speed, scope, and potential impact of malware outbreaks. The results demonstrate that neural networks can not only predict malware propagation more accurately than traditional methods but also adapt to evolving attack strategies, providing real-time insights for proactive defense mechanisms. This study highlights the potential of neural networks in enhancing cybersecurity measures and lays the foundation for developing intelligent, adaptive systems capable of combating future malware threats.

This paper presents the Modular Assistive Vertebra (MAV), a key component of the FleXo exoskeleton. FleXo is a passive, soft back-support exoskeleton designed to enhance ergonomic support by providing lifting support without compressing the user's vertebrae. This study systematically investigates the MAV's force transmission properties through theoretical and experimental methods. The results demonstrate the MAV's capability to convert linear forces exerted by a pulling cable into torque while minimizing the transmission of linear forces. Experimental setups with varying geometrical configurations validate the theoretical models, providing insights into the design and optimization of ergonomic support devices.

We present a 4 × 4 antenna array receiver designed for real-time Direction of Arrival (DoA) estimation of radio frequency sources within the 700-2600 MHz range. Our device, called RadioVision, overlays detected radio source markers onto optical images captured by a video camera. It features a heterodyne receiver with a 70 MHz intermediate frequency bandwidth, 16 ADCs operating at 140 MSPS, and an FPGA for real-time FFT and linear algebra computations. We have implemented the 2D U-ESPRIT algorithm in the frequency domain using Polyphase Filter Banks (PFBs) and pipelined FFTs, which reduces the number of needed ADCs, simplifies the calibration process, and filters out unwanted frequency bands. Experiments conducted in both rural and urban environments demonstrated the system's effectiveness. We achieved a mean squared error for DoA estimation of 1.78° at 90 m and 3.79° at 180 m in rural areas, and approximately 5.76° in urban settings at 60 m. This work represents the first on-field performance report of modern DoA algorithms, such as the 2D U-ESPRIT.

We propose a method to harden sensors against laser radiation by defocusing the employed optics by purpose, and to reconstruct the sought focused images of the scene via image reconstruction. The introduced defocus widens the laser spot incident on the sensor and greatly reduces its damage potential. We employ a coded aperture and optimize its pattern jointly with the free parameters of the image reconstruction pipeline. For the image reconstruction, we combine a state-of-theart alternating direction method of multipliers (ADMM)-based physically informed deconvolution stage with a U-Net-like neural network to remove remaining reconstruction artifacts. To evaluate the performance of our proposed approach, we conducted reconstruction experiments on simulated data, including ablation experiments and on real data and performed sensor destruction tests with and without sensor protection. Destructive experiments with increasing laser power suggest that our approach has the potential to increase the tolerable radiation threshold by about three orders of magnitudes.

It remains challenging to realize well-rounded security in IoT forensics applications as widely used resource-constrained IoT devices cannot afford many proven technologies like blockchain. Thus, security solutions are often developed in a heuristic and empirical way that leads to residual security flaws or risks. Fortunately, the concept of 'trust' provides a promising methodology. Trust is a finite set of goals which are definable, verifiable and customizable. A lightweight security solution accommodating constrained IoT devices can be derived by identifying a minimum set of trust goals and developing lightweight protocols for them. This paper exemplifies a trust model for IoT forensics applications with formal proof and detailed evaluation. Trust goals are grouped into three trust levels according to the spatial-and-temporal span, i.e. evidence-, device-and system-level trust, which then aggregate into two trust schemes, i.e. EDTrust and EDSTrust. Formal proof confirms their validity, and performance evaluation reveals their resource-efficiency compared with existing IoT forensics schemes.

The Source Code Control System (SCCS) was first introduced in 1975 [1]. It controlled computer program source code by tracking versions and recording who made changes, when, and why. The present retrospective paper assesses the strengths and weaknesses of SCCS and traces its influence on software engineering over the past fifty years.

Explainable Artificial Intelligence (XAI) emerged to reveal the internal mechanism of machine learning models and how the features affect the prediction outcome. Collinearity is one of the big issues that XAI methods face when identifying the most informative features in the model. Current XAI approaches assume the features in the models are independent and calculate the effect of each feature toward model prediction independently from the rest of the features. However, such assumption is not realistic in real life applications. We propose an Additive Effects of Collinearity (AEC) as a novel XAI method that aim to considers the collinearity issue when it models the effect of each feature in the model on the outcome. AEC is based on the idea of dividing multivariate models into several univariate models in order to examine their impact on each other and consequently on the outcome. The proposed method is implemented using simulated and real data to validate its efficiency comparing with the a state of arts XAI method. The results indicate that AEC is more robust and stable against the impact of collinearity when it explains AI models compared with the state of arts XAI method.

Gig economy platforms depend significantly on effective data processing to manage the requirements of changing environments. Nonetheless, conventional data pipelines often face obstructions and delays, which restrict overall operational efficiency. In this research, we present a novel approach aimed at enhancing data pipelines tailored for gig economy platforms. Our system integrates sophisticated design concepts with responsive algorithms to improve data flow and minimize latency. By analyzing current challenges, we intentionally distribute resources based on real-time data needs. We evaluate the effects of our optimizations through precise performance metrics that gauge data throughput and processing speed. Experimental findings show that our method greatly improves the efficiency of data operations, establishing a basis for expanding gig economy platforms while upholding service quality. This study provides important perspectives on the complex dynamics of data management within fast-evolving business environments, promoting additional investigation into effective and scalable strategies in the gig economy.

Current probes have long been used for Electromagnetic Compatibility (EMC) pre-compliance and diagnostic purposes. Dipole and transmission line-based models are often used to relate Common-Mode (CM) current levels to potential Radiated Emission (RE) levels. This relationship enables RE estimates to be performed without the use of accepted certification methodologies – reducing project costs and timelines. These models are based on a set of assumptions which cannot fully match real-world scenarios. This article expands the discussion by examining the relationship between the CM current and RE of a representative metallic structure using both measurements and a Finite Element Method (FEM) code. The relationship is shown to be frequency dependent, resulting in both over and under prediction of the RE levels. A practical recommendation is made on how to exploit the equivalence of CM current-based estimations of system RE.