Search

In recent years, the field of neural machine translation (NMT) for SPARQL query generation has witnessed significant growth. Incorporating the copy mechanism with traditional encoder-decoder architectures and using pre-trained encoder-decoder and large language models have set new performance benchmarks. This paper presents various experiments that replicate and expand upon recent NMT-based SPARQL generation studies, comparing pre-trained language models (PLMs), non-pre-trained language models (NPLMs), and large language models (LLMs), highlighting the impact of question annotation and the copy mechanism and testing various fine-tuning methods using LLMs. In particular, we provide a systematic error analysis of the models and test their generalization ability. Our study demonstrates that the copy mechanism yields significant performance enhancements for most PLMs and NPLMs. Annotating the data is pivotal to generating correct URIs, with the “tag-within” strategy emerging as the most effective approach. Additionally, our findings reveal that the primary source of errors stems from incorrect URIs in SPARQL queries that are sometimes replaced with hallucinated URIs when using base models. This does not happen using the copy mechanism, but it sometimes leads to selecting wrong URIs among candidates. Finally, the performance of the tested LLMs fell short of achieving the desired outcomes.

Reaching the European climate target is a complex and multifaceted challenge that involves different sectors and requires coordinated efforts at various levels. Increasing the share of renewable energy sources in the energy mix and the electrification of final energy uses in all sectors represent viable solutions for the energy transition. Distribution networks are expected to be strongly influenced by and influence such transformation while withstanding progressively increasing climate impacts, and this will require a revolution at the Power System level, starting from the planning phase. The future roadmap of a power distribution system shall include not exclusively network upgrades but also non-network solutions focusing on operation strategies exploiting the flexibility gathered from distributed energy resources. To design new transition-ready planning tools for distribution systems able to consider these aspects, the role of flexibility has been analyzed on a real-world, large-scale test case characterized by a high number of connection requests and an expected high yearly electrification rate. Stress has been put on correctly assessing the value of flexibility in planning the distribution system development. One of the most important findings is that flexibility can be a valid option in helping grid management but, most importantly, an opportunity to reconsider planning by applying a new revolutionary risk-oriented approach that may lead to modify the way distribution grids have been planned and operated so far. This emerges as the disruptive value of local flexibility utilization, besides the engagement of all players, that is necessary for the completion of the energy transition.

Machine learning algorithms present a robust alternative for building Intrusion Detection Systems due to their ability to recognize attacks in computer network traffic by recognizing patterns in large amounts of data. Typically, classifiers are trained for this task. Together, ensemble learning algorithms have increased the performance of these detectors, reducing classification errors and allowing computer networks to be more protected. This research presents a comprehensive Systematic Review of the Literature where works related to intrusion detection with ensemble learning were obtained from the most relevant scientific bases. We offer 188 works, several compilations of datasets, classifiers, and ensemble algorithms, and document the experiments that stood out in their performance. A characteristic of this research is its originality. We found two surveys in the literature specifically focusing on the relationship between ensemble techniques and intrusion detection. We present for the last eight years covered by this survey a timeline-based view of the works studied to highlight evolutions and trends. The results obtained by our survey show a growing area, with excellent results in detecting attacks but with needs for improvement in pruning for choosing classifiers, which makes this work unprecedented for this context.

We report on the development and characterization of superconducting damascene tantalum nitride (TaN) nanowires, 100 nm–3 μm wide, with TaN thicknesses varying from 5 to 35 nm, using 193-nm optical lithography and chemical mechanical planarization among other 300-mm wafer-scale processes. The TaN film composition chosen for nanowire fabrication was informed by a detailed study of unpatterned TaN films with varying nitrogen to tantalum ratios, formed by reactive sputtering. We also discuss the influence of encapsulation by copper and disordered atomic layer deposited TaN on the critical current of superconducting nanowires. Superconducting critical current density (measured at 12 mK) ranges from 0.12 to 0.85 MA/cm2 depending on nanowire width and film thickness. The potential of ultrathin TaN nanowires at 300-mm scale is discussed in the context of applications such as on-chip integration for readout of superconducting qubits, in single-photon detection for quantum computing, as well as in large single-photon detecting focal plane arrays for cosmology in a broader range of wavelengths.

The monocular depth estimation (MDE) is the task of estimating depth from a single frame. This information is an essential knowledge in many computer vision tasks such as scene understanding and visual odometry, which are key components in autonomous and robotic systems. Approaches based on the state of the art vision transformer architectures are extremely deep and complex not suitable for real-time inference operations on edge and autonomous systems equipped with low resources (i.e. robot indoor navigation and surveillance). This paper presents SPEED, a Separable Pyramidal pooling EncodEr-Decoder architecture designed to achieve real-time frequency performances on multiple hardware platforms. The proposed model is a fast-throughput deep architecture for MDE able to obtain depth estimations with high accuracy from low resolution images using minimum hardware resources (i.e. edge devices). Our encoder-decoder model exploits two depthwise separable pyramidal pooling layers, which allow to increase the inference frequency while reducing the overall computational complexity. The proposed method performs better than other fast-throughput architectures in terms of both accuracy and frame rates, achieving real-time performances over cloud CPU, TPU and the NVIDIA Jetson TX1 on two indoor benchmarks: the NYU Depth v2 and the DIML Kinect v2 datasets.

Electroencephalogram signals (EEG) have provided biometric identification systems with great capabilities. Several studies have shown that EEG introduces unique and universal features besides specific strength against spoofing attacks. Essentially, EEG is a graphic recording of the brain’s electrical activity calculated by sensors (electrodes) on the scalp at different spots, but their best locations are uncertain. In this paper, the EEG channel selection problem is formulated as a binary optimization problem, where a binary version of the Grey Wolf Optimizer (BGWO) is used to find an optimal solution for such an NP-hard optimization problem. Further, a Support Vector Machine classifier with a Radial Basis Function kernel (SVM-RBF) is then considered for EEG-based biometric person identification. For feature extraction purposes, we examine three different auto-regressive coefficients. A standard EEG motor imagery dataset is employed to evaluate the proposed method, including four criteria: (i) Accuracy, (ii) F-Score, (iii) Recall, and (v) Specificity. In the experimental results, the proposed method (named BGWO-SVM) obtained 94.13% accuracy using only 23 sensors with 5 auto-regressive coefficients. Besides, BGWO-SVM finds electrodes not too close to each other to capture relevant information all over the head. As concluding remarks, BGWO-SVM achieved the best results concerning the number of selected channels and competitive classification accuracies against other meta-heuristics algorithms.

Motivated by the increased use of UAS in commercial applications, in this paper we tackle the problem of path planning when requests are submitted by UAS managed by different operators. We propose the new problem of generalized path planning for UAS Traffic Management, where the UAS path is described by operators with a sequence of waypoint groups and a solution trajectory must pass through a waypoint in each group. This problem is typical for applications where multiple charging stations and pickup/drop-off locations are distributed in a flight area. Our solution builds upon prior work on discretized space-time graph path planning and proposes a novel multi-source/multi-destination graph search algorithm that generates collision-free trajectories for pre-flight CDR. Our efficient algorithm has runtime proportional to the number of groups and avoids combinatorial explosion. We apply our mechanism to the energy-constrained UAS package delivery problem with multiple warehouses and battery charging stations. Simulation results show that our algorithm is efficient and scalable with the number of requests and graph size. The addition of charging stations and the option for multiple warehouses increases the request admission ratio and reduces the overall trajectory duration, effectively improving both the planner’s quality of service and the efficiency of airspace usage.

People with special nutritional needs must monitor their food intake as accurately as possible. One of the problems with food-intake monitoring is a correct estimation of portion sizes. This paper presents an open-source solution to the problem of portion-size estimation consisting of a pocket-sized kitchen scale and a demo mobile application. It is implemented as a soft real-time, low-power consuming embedded system with a Bluetooth connection to the smartphone. Releasing both the hardware and software of the system as open source paves the way towards accurate nutrition tracking and personalized nutrition. This can lead to important advances in the field of nutrition research, as well as practice. This is demonstrated with a proof-of-concept integration of our solution with a commercial mobile application that helps users to keep track of their health and lifestyle goals.

Concept drift techniques aim at learning patterns from data streams that may change over time. Although such behavior is not usually expected in controlled environments, real-world scenarios can face changes in the data, such as new classes, clusters, and features. Traditional classifiers can be easily fooled in such situations, resulting in poor performances. Common concept drift domains include recommendation systems, energy consumption, artificial intelligence systems with dynamic environment interaction, and biomedical signal analysis (e.g., neurogenerative diseases). In this paper, we surveyed several works that deal with concept drift, as well as we presented a comprehensive study of public synthetic and real datasets that can be used to cope with such a problem. In addition, we considered a review of different types of drifts and approaches to handling such changes in the data. We considered different learners employed in classification tasks and the use of drift detection mechanisms, among other characteristics.

In the modern life, the authentication technique for any system is considered as one of the most important and challenging tasks. Therefore, many researchers have developed traditional authentication systems to deal with our digital society. Recently, several studies showed that the brain electrical activity or electroencephalogram (EEG) signals could provide robust and unique features that can be considered as a new biometric authentication technique, given that accurate methods to decompose the signals must also be considered. This paper proposes a novel method for extracting EEG features using multi-objective flower pollination algorithm and the wavelet transform. The proposed method was applied in two scenarios for EEG signal decomposition to extract unique features from the original signals. Moreover, the proposed method is compared with the state-of-the-art techniques using different criteria with promising results.