Marianna La Rocca

Traumatic brain injury (TBI) is one of the leading causes of acquired epilepsy, with a significant proportion of patients developing post-traumatic epilepsy (PTE) even months or years after the initial injury. The identification of reliable imaging biomarkers able to predict epileptogenesis remains a major clinical challenge. In recent years, diffusion-weighted imaging (DWI) and structural connectome analysis have emerged as promising tools to investigate brain network alterations associated with late seizure development. Machine learning approaches may further support the detection of predictive patterns in complex neuroimaging data. The goal of this study is to perform a binary classification between seizure-free and late seizure-affected patients following TBI, with a specific focus on the identification of the anatomical regions potentially connected with late seizure development. A dataset of 59 diffusion weighted images (DWI) scans from the EpiBioS4Rx project, including 42 seizure-free and 17 late seizure-affected TBI patients, was analyzed. A Random Forest classification algorithm was applied, incorporating network feature importance based on the Gini index to investigate model’s decisions and allow a clinical interpretation. The model reported a 69% ± 0.03 accuracy for discrimination and a 73% AUC ± 0.05. Despite the limited and imbalanced nature of the dataset, and the fact that the performance does not significantly exceed chance once all data-dependent steps are taken into account, our approach allows us to achieve accurate classification results compared to the literature and to identify brain regions potentially associated with epileptogenesis.

Quantum-enhanced machine learning, encompassing both quantum algorithms and quantum-inspired classical methods such as tensor networks, offers promising tools for extracting structure from complex, high-dimensional data. In this work, we study the training dynamics of Matrix Product State (MPS) classifiers applied to three-class problems, using both fashion MNIST and hyperspectral satellite imagery as representative datasets. We investigate the phenomenon of grokking, where generalization emerges suddenly after memorization, by tracking entanglement entropy, local magnetization, and model performance across training sweeps. Additionally, we employ information-theory tools to gain deeper insights: transfer entropy is used to reveal causal dependencies between label-specific quantum masks, while O-information captures the shift from synergistic to redundant correlations among class outputs. Our results show that grokking in the fashion MNIST task coincides with a sharp entanglement transition and a peak in redundant information, whereas the overfitted hyperspectral model retains synergistic, disordered behavior. These findings highlight the relevance of high-order information dynamics in quantum-inspired learning and emphasize the distinct learning behaviors that emerge in multi-class classification, offering a principled framework to interpret generalization in quantum machine learning architectures.

Echocardiography is a fundamental tool in cardiovascular diagnostics, providing radiation-free real-time assessments of cardiac function. However, its accuracy strongly depends on operator expertise, resulting in inter-operator variability that affects diagnostic consistency. Recent advances in artificial intelligence have enabled new applications for real-time image classification and probe guidance, but these typically rely on large datasets and specialized hardware such as GPU-based or embedded accelerators, limiting their clinical adoption. Here, we address this challenge by developing a cognitive electronic unit that integrates convolutional neural network (CNN) models and an inertial sensor for assisted echocardiography. We show that our system—powered by an NVIDIA Jetson Orin Nano—can effectively classify standard cardiac views and differentiate good-quality from poor-quality ultrasound images in real time even when trained on relatively small datasets. Preliminary results indicate that the combined use of CNN-based classification and inertial sensor-based feedback can reduce inter-operator variability and may also enhance diagnostic precision. By lowering barriers to data acquisition and providing real-time guidance, this system has the potential to benefit both novice and experienced sonographers, helping to standardize echocardiographic exams and improve patient outcomes. Further data collection and model refinements are ongoing, progressing the way for a more robust and widely applicable clinical solution.

Grokking is a intriguing phenomenon in machine learning where a neural network, after many training iterations with negligible improvement in generalization, suddenly achieves high accuracy on unseen data. By working in the quantum-inspired machine learning framework based on tensor networks, we numerically prove that grokking phenomenon can be related to an entanglement dynamical transition in the underlying quantum many-body systems, consisting in a one-dimensional lattice with each site hosting a qubit. Two datasets are considered as use case scenarios, namely fashion MNIST and gene expression communities of hepatocellular carcinoma. In both cases, we train Matrix Product State (MPS) to perform binary classification tasks, and we analyse the learning dynamics. We exploit measurement of qubits magnetization and correlation functions in the MPS network as a tool to identify meaningful and relevant gene subcommunities, verified by means of enrichment procedures.

The development of early warning systems for systemic crises has recently received growing interests. Recent studies have proposed possible solutions to address this challenging topic, in particular by means of cutting-edge artificial intelligence (AI) approaches. Financial data are fundamentally characterized by intrinsic temporal dynamics and the presence of both short-/long-term interactions. Hence, it is of paramount importance, when validating the proposed solutions to adopt validation strategies which consider this aspect. To this aim, we show here how Temporal Cross Validation (TCV) deeply affects the models’ learning. Moreover, to take into account the data imbalance often characterizing these models, we combine the TCV with a popular solution, which is the SMOTE (Synthetic Minority Oversampling TEchnique) algorithm.

Neuroaesthetics is a relatively young field that connects neuroscience with empirical aesthetics and originates in the neurological theory of aesthetic experience. It investigates brain structures and activity during the phenomena of artistic perception and production and, at the same time, attempts to understand the influence of neurological pathologies on these mechanisms. For each participant (six subjects with mild cognitive decline and ten controls), electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) data were acquired thanks to a wearable EEG–fNIRS system during the execution of a P300 task.

In the field of neuroscience, brain–computer interfaces (BCIs) are used to connect the human brain with external devices, providing insights into the neural mechanisms underlying cognitive processes, including aesthetic perception. Non-invasive BCIs, such as EEG and fNIRS, are critical for studying central nervous system activity and understanding how individuals with cognitive deficits process and respond to aesthetic stimuli. This study assessed twenty participants who were divided into control and impaired aging (AI) groups based on MMSE scores. EEG and fNIRS were used to measure their neurophysiological responses to aesthetic stimuli that varied in pleasantness and dynamism. Significant differences were identified between the groups in P300 amplitude and late positive potential (LPP), with controls showing greater reactivity. AI subjects showed an increase in oxyhemoglobin in response to pleasurable stimuli, suggesting hemodynamic compensation. This study highlights the effectiveness of multimodal BCIs in identifying the neural basis of aesthetic appreciation and impaired aging. Despite its limitations, such as sample size and the subjective nature of aesthetic appreciation, this research lays the groundwork for cognitive rehabilitation tailored to aesthetic perception, improving the comprehension of cognitive disorders through integrated BCI methodologies.