publications

2026

1. Development of a Measurement Procedure for Emotional States Detection Based on Single-Channel Ear-EEG: A Proof-of-Concept Study

   Marco Arnesano, Pasquale Arpaia, Simone Balatti, and 7 more authors

   Sensors, 2026

   Abs DOI HTML

   Real-time emotion monitoring is increasingly relevant in healthcare, automotive, and workplace applications, where adaptive systems can enhance user experience and well-being. This study investigates the feasibility of classifying emotions along the valence–arousal dimensions of the Circumplex Model of Affect using EEG signals acquired from a single mastoid channel positioned near the ear. Twenty-four participants viewed emotion-eliciting videos and self-reported their affective states using the Self-Assessment Manikin. EEG data were recorded with an OpenBCI Cyton board and both spectral and temporal features (including power in multiple frequency bands and entropy-based complexity measures) were extracted from the single ear-channel. A dual analytical framework was adopted: classical statistical analyses (ANOVA, Mann–Whitney U) and artificial neural networks combined with explainable AI methods (Gradient × Input, Integrated Gradients) were used to identify features associated with valence and arousal. Results confirmed the physiological validity of single-channel ear-EEG, and showed that absolute 𝛽- and 𝛾-band power, spectral ratios, and entropy-based metrics consistently contributed to emotion classification. Overall, the findings demonstrate that reliable and interpretable affective information can be extracted from minimal EEG configurations, supporting their potential for wearable, real-world emotion monitoring. Nonetheless, practical considerations—such as long-term comfort, stability, and wearability of ear-EEG devices—remain important challenges and motivate future research on sustained use in naturalistic environments.

2. IMPACTX: improving model performance by appropriately constraining the training with teacher explanations

   Andrea Apicella, Salvatore Giugliano, Francesco Isgró, and 2 more authors

   Artificial Intelligence Review, 2026

   Abs DOI HTML

   The eXplainable Artificial Intelligence (XAI) research predominantly concentrates on providing explanations about AI model decisions, especially Deep Learning (DL) models. However, there is a growing interest in using XAI techniques to automatically improve the performance of the AI systems themselves. This paper proposes IMPACTX, a novel approach that leverages XAI as a fully automated attention mechanism, without requiring external knowledge or human feedback. Experimental results show that IMPACTX improves predictive performance compared to standalone baseline models by integrating XAI-based supervision provided by teacher explanations into the training process. Furthermore, IMPACTX directly provides new feature attribution maps for the model’s decisions, without relying on external XAI methods during the inference process. Our proposal is evaluated using three widely recognized DL models (EfficientNet-B2, MobileNet, and LeNet-5) and on six publicly available benchmark datasets, comprising three image datasets (CIFAR-10, CIFAR-100, and STL-10) and three tabular datasets (Covertype, CDC Diabetes Health Indicators, and Adult). The results show that IMPACTX consistently improves the performance of all the inspected DL models across all evaluated datasets, and it directly provides appropriate explanations for its responses.

3. Comparative Study of Modal Curvature and AI-Based Approaches for Vibration-Based Damage Detection in Structural Health Monitoring Systems of Prestressed Concrete Beams

   Antonio Bilotta, Andrea Pollastro, Ivan Di Cristinzi, and 1 more author

   Infrastructures, 2026

   Abs DOI HTML

   Vibration-based damage detection methods are increasingly recognized as effective tools for monitoring the structural health of bridges. However, their reliability and applicability to various types of structural defects require further study, especially based on experimental tests, to correctly interpretate the results and compare the efficiency of different damage indexes. In the field of Structural Health Monitoring (SHM) by dynamic techniques, operational modal analysis (OMA) is of particular interest because only ambient signals are used, avoiding the service interruption of the infrastructures. However, the key issues of an efficient SHM are the possibility to have a quick alarm if an anomalous response is detected and the capability to localize the defect. Several methods can be applied for the anomaly detection considering machine learning, moving further than global modal parameters like the vibration frequency. Conversely for defect localization, local modal parameters, like modal curvature, can be efficient but also a different application of machine learning can be considered. In this paper, two approaches are compared for level 1 (detection) and 2 (localization) damage detection using acceleration measurements: the modal parameters and an Artificial Intelligence (AI)-based procedure using Variational Autoencoders (VAEs). The case study is a set of post-tensioned prestress concrete (PC) beams that represent a wide stock of existing bridges characterized by defects due to a reduction in the prestressing load, a lack of mortar in ducts, and corrosion of tendons. The results show that both methods can be effective, even if defects in PC beams are difficult to be detect with the dynamic response. Finally, the AI-based approach seems a promising solution because I allows for an earlier alarm, even with few sensors, while the modal curvature approach provides a better explanation of the identified anomaly, although it requires a greater number of sensors.

2025

1. Domain-specific text embedding model for accelerator physics

   Thorsten Hellert, João Montenegro, Marco Venturini, and 1 more author

   Phys. Rev. Accel. Beams, Apr 2025

   Abs DOI HTML

   Accelerator physics presents unique challenges for natural language processing (NLP) due to its specialized terminology and complex concepts. A key component in overcoming these challenges is the development of robust text embedding models that transform textual data into dense vector representations, facilitating efficient information retrieval and semantic understanding. In this work, we introduce AccPhysBERT, a sentence embedding model fine-tuned specifically for accelerator physics. Our model demonstrates superior performance across a range of downstream NLP tasks, surpassing existing models in capturing the domain-specific nuances of the field. We further showcase its practical applications, including semantic paper-reviewer matching and integration into retrieval-augmented generation systems, highlighting its potential to enhance information retrieval and knowledge discovery in accelerator physics.

2. SincVAE: A new semi-supervised approach to improve anomaly detection on EEG data using SincNet and variational autoencoder

   Andrea Pollastro, Francesco Isgrò, and Roberto Prevete

   Computer Methods and Programs in Biomedicine Update, 2025

   Abs DOI HTML

   Over the past few decades, electroencephalography monitoring has become a pivotal tool for diagnosing neurological disorders, particularly for detecting seizures. Epilepsy, one of the most prevalent neurological diseases worldwide, affects approximately 1 % of the population. These patients face significant risks, underscoring the need for reliable, continuous seizure monitoring in daily life. Most of the techniques discussed in the literature rely on supervised machine learning methods. However, the challenge of accurately labeling variations in epileptic electroencephalography waveforms complicates the use of these approaches. Additionally, the rarity of ictal events introduces a high imbalance within the data, which could lead to poor prediction performance in supervised learning approaches. Instead, a semi-supervised approach allows training the model only on data that does not contain seizures, thus avoiding the issues related to the data imbalance. This work introduces a semi-supervised approach for detecting epileptic seizures from electroencephalography data based on a novel deep learning-based method called SincVAE. This method integrates SincNet, designed to learn an ad-hoc array of bandpass filters, as the first layer of a variational autoencoder, potentially eliminating the preprocessing stage where informative frequency bands are identified and isolated. Experimental evaluations on the Bonn and CHB-MIT datasets indicate that SincVAE improves seizure detection in electroencephalography data, with the capability to identify early seizures during the preictal stage and monitor patients throughout the postictal stage.

2024

1. Dynamic identification methods and artificial intelligence algorithms for damage detection of masonry infills

   Alessandra De Angelis, Antonio Bilotta, Maria Rosaria Pecce, and 2 more authors

   Journal of Civil Structural Health Monitoring, 2024

   Abs HTML

   The failure of non-structural components after an earthquake is among the most expensive earthquake-incurred damage, and may also have life-threatening consequences, especially in public buildings with very crowded facilities, because exposition is high and the risk increases accordingly. The assessment of existing non-structural components is particularly complex because in-depth in situ investigation is necessary to detect the presence of deficiencies or damage. This problem concerns interior and exterior partitions made of various materials (e.g., glass and masonry), as well as equipment and facilities in construction (building, industry, and infrastructure). Defining the boundary conditions of these components is of paramount importance. Indeed, external restraints (i) affect dynamic properties and, thus, the action experienced during an earthquake, and (ii) influence the capacity to detach the component before failure from the bearing structure (e.g., an infill wall connected to the main structural frame, or equipment connected to secondary structural members such as floors). The authors, therefore, conducted environmental vibration tests of an infill wall and refined a finite element model to simulate typical damage scenarios to be implemented on the wall. Selected damage scenarios were then artificially realized on the existing infill and further ambient vibration tests were performed to measure the accelerations for each of them. Finally, the authors used these accelerations to detect the damage by means of established OMA, as well as innovative machine learning techniques. The results showed that convolutional variational autoencoders (CVAE), coupled with a one-class support vector machine (OC-SVM), identified the anomaly even when the OMA exhibited limited effectiveness. Moreover, the machine learning procedure minimizes human interaction during the damage detection process.

2. Application of deep learning methods for beam size control during user operation at the Advanced Light Source

   Thorsten Hellert, Tynan Ford, Simon C. Leemann, and 3 more authors

   Phys. Rev. Accel. Beams, Jul 2024

   Abs DOI HTML

   Past research at the Advanced Light Source (ALS) provided a proof-of-principle demonstration that deep learning methods could be effectively employed to compensate for the significant perturbations to the transverse electron beam size induced by user-controlled adjustments of the insertion devices. However, incorporating these methods into the ALS’ daily operations has faced notable challenges. The complexity of the system’s operational requirements and the significant upkeep demands has restricted their sustained application during user operation. Here, we introduce the development of a more robust neural network (NN)-based algorithm that utilizes a novel online fine-tuning approach and its systematic integration into the day-to-day machine operations. Our analysis emphasizes the process of NN model selection, demonstrates the superior performance of the NN-based method over traditional feedback methods, and examines the effectiveness and resilience of the new algorithm during user-operation scenarios.

3. PhysBERT: A text embedding model for physics scientific literature

   Thorsten Hellert, João Montenegro, and Andrea Pollastro

   APL Machine Learning, 2024

   Abs HTML

   The specialized language and complex concepts in physics pose significant challenges for information extraction through Natural Language Processing (NLP). Central to effective NLP applications is the text embedding model, which converts text into dense vector representations for efficient information retrieval and semantic analysis. In this work, we introduce PhysBERT, the first physics-specific text embedding model. Pre-trained on a curated corpus of 1.2 × 106 arXiv physics papers and fine-tuned with supervised data, PhysBERT outperforms leading general-purpose models on physics-specific tasks, including the effectiveness in fine-tuning for specific physics subdomains.

2023

1. On the effects of data normalization for domain adaptation on EEG data

   Andrea Apicella, Francesco Isgrò, Andrea Pollastro, and 1 more author

   Engineering Applications of Artificial Intelligence, 2023

   Abs DOI HTML

   In Machine Learning (ML), a well-known problem is the Dataset Shift problem where the data in the training and test sets can follow different probability distributions, leading ML systems toward poor generalization performances. This problem is intensely felt in Brain-Computer Interfaces (BCIs), where bio-signals as Electroencephalographic (EEG) are often used. Indeed, EEG signals are highly non-stationary both over time and between different subjects. To overcome this problem, several solutions are based on transfer learning approaches such as Domain Adaption (DA). In several cases, however, the actual causes of the improvements remain ambiguous. This paper focuses on the impact of data normalization strategies applied together with DA methods. In particular, using SEED, DEAP, and BCI Competition IV 2a EEG datasets, we experimentally evaluated the impact of different normalization strategies applied with and without several well-known DA methods. It results that the choice of the normalization strategy plays a key role on the classifier performances in DA scenarios, and, often, the use of only an appropriate normalization schema outperforms the DA technique. For SEED and BCI Competition IV 2a, a proper normalization strategy alone in a cross-subject context allows to reach accuracy of 81.52 ± 7.26 % and 68.52 ± 11.35 %, respectively. In a cross-session context, the accuracy of 86.56 ± 8.15 % and 67.82 ± 12.48 % for SEED and BCI Competition can be reached, respectively. For DEAP, the best cross-subject performance achieved using only normalization was 39.33 ± 14.08 %. All these results are comparable with the performance obtained by several well-known DA strategies.

2. Semi-supervised detection of structural damage using Variational Autoencoder and a One-Class Support Vector Machine

   Andrea Pollastro, Giusiana Testa, Antonio Bilotta, and 1 more author

   IEEE Access, 2023

   Abs DOI HTML

   In recent years, Artificial Neural Networks (ANNs) have been introduced in Structural Health Monitoring (SHM) systems. A semi-supervised method with a data-driven approach allows the ANN training on data acquired from an undamaged structural condition to detect structural damages. In standard approaches, after the training stage, a decision rule is manually defined to detect anomalous data. However, this process could be made automatic using machine learning methods. This paper proposes a semi-supervised method with a data-driven approach to detect structural anomalies. The methodology consists of: 1) a Variational Autoencoder (VAE) to approximate undamaged data distribution and 2) a One-Class Support Vector Machine (OC-SVM) to discriminate different health conditions using damage-sensitive features extracted from VAE’s signal reconstruction. The method is applied to a scale steel structure that was tested in nine damage scenarios by IASC-ASCE Structural Health Monitoring Task Group.

3. Adaptive filters in graph convolutional neural networks

   Andrea Apicella, Francesco Isgrò, Andrea Pollastro, and 1 more author

   Pattern Recognition, 2023

   Abs DOI HTML

   Over the last few years, the availability of an increasing data generated from non-Euclidean domains, which are usually represented as graphs with complex relationships, and Graph Neural Networks (GNN) have gained a high interest because of their potential in processing graph-structured data. In particular, there is a strong interest in performing convolution on graphs using an extension of the GNN architecture, generally referred to as Graph Convolutional Neural Networks (ConvGNN). Convolution on graphs has been achieved mainly in two forms: spectral and spatial convolutions. Due to the higher flexibility in exploring and exploiting the graph structure of data, there is recently an increasing interest in investigating the possibilities that the spatial approach can offer. The idea of finding a way to adapt the network behaviour to the inputs they process to maximize the total performances has aroused much interest in the neural networks literature over the years. This paper presents a novel method to adapt the behaviour of a ConvGNN to the input performing spatial convolution on graphs using input-specific filters, which are dynamically generated from nodes feature vectors. The experimental assessment confirms the capabilities of the proposed approach, achieving satisfying results using a low number of filters.

2022

1. EEG-based measurement system for monitoring student engagement in learning 4.0

   Andrea Apicella, Pasquale Arpaia, Mirco Frosolone, and 3 more authors

   Scientific Reports, 2022

   Abs DOI HTML

   A wearable system for the personalized EEG-based detection of engagement in learning 4.0 is proposed. In particular, the effectiveness of the proposed solution is assessed by means of the classification accuracy in predicting engagement. The system can be used to make an automated teaching platform adaptable to the user, by managing eventual drops in the cognitive and emotional engagement. The effectiveness of the learning process mainly depends on the engagement level of the learner. In case of distraction, lack of interest or superficial participation, the teaching strategy could be personalized by an automatic modulation of contents and communication strategies. The system is validated by an experimental case study on twenty-one students. The experimental task was to learn how a specific human-machine interface works. Both the cognitive and motor skills of participants were involved. De facto standard stimuli, namely (1) cognitive task (Continuous Performance Test), (2) music background (Music Emotion Recognition—MER database), and (3) social feedback (Hermans and De Houwer database), were employed to guarantee a metrologically founded reference. In within-subject approach, the proposed signal processing pipeline (Filter bank, Common Spatial Pattern, and Support Vector Machine), reaches almost 77% average accuracy, in detecting both cognitive and emotional engagement.