Publications

This preprint studies how empirical Jensen-Shannon divergence estimates vary across estimator families, sampling protocols, calibration, dimensionality, and class balance.

This preprint examines the limits and trade-offs of synthetic image generation when data are scarce, with implications for reliable model evaluation and augmentation.

This preprint studies patient-level causal survival analysis under censoring, using deep learning methods designed for medical data and treatment-effect reasoning.

This study analyzes privacy risks in federated generative settings through a two-stage membership inference perspective, connecting geometric structure with attack behavior.

This work applies interpretable machine learning to in situ impedance spectra for real-time diagnostics in solid-state batteries, supporting transparent monitoring of electrochemical behavior.

Federated learning is an Artificial Intelligence framework that allows to train machine learning models in a distributed way, avoiding the sharing of sensitive data. This method is crucial, especially in healthcare applications, where patient privacy is an enormous concern. This study aims to demonstrate the effectiveness of federated learning in addressing real use-cases. In addition, the platform developed by the GenoMed4All consortium was tested against computational simulations. Two real use-cases were considered: a survival estimation task on a Myelodysplastic Syndrome cohort, and a classification task on a scarce Sickle Cell Disease dataset. Both cohorts were distributed in a federated learning scenario with 3 clients and a common test set. Federated learning came out to be crucial in improving the performance of local survival models, especially for nodes with the lower number of samples, which most benefit from federated aggregation. Despite the limited number of patients for the classification task, federated learning consistently improved model performance across multiple metrics beyond the F1 score, including comparisons between different sample distributions among the three clients. These results confirmed the effectiveness of federated learning in healthcare applications, especially for scarce datasets, for which this technique can represent a viable solution. Moreover, the results of the GenoMed4All platform are completely in agreement with the computer simulations, proving the reliability of the developed platform.

Causal generative models provide a principled framework for answering observational, interventional, and counterfactual queries from observational data. However, many deep causal models rely on highly expressive architectures with opaque mechanisms, limiting auditability in high-stakes domains. We propose KaCGM, a causal generative model for mixed-type tabular data where each structural equation is parameterized by a Kolmogorov–Arnold Network (KAN). This decomposition enables direct inspection of learned causal mechanisms, including symbolic approximations and visualization of parent–child relationships, while preserving query-agnostic generative semantics. We introduce a validation pipeline based on distributional matching and independence diagnostics of inferred exogenous variables, allowing assessment using observational data alone. Experiments on synthetic and semi-synthetic benchmarks show competitive performance against state-of-the-art methods. A real-world cardiovascular case study further demonstrates the extraction of simplified structural equations and interpretable causal effects. These results suggest that expressive causal generative modeling and functional transparency can be achieved jointly, supporting trustworthy deployment in tabular decision-making settings. Code: https://github.com/aalmodovares/kacgm

Integrating Artificial Intelligence (AI) into Survival Analysis (SA) has advanced predictive modeling in healthcare, enabling precise and personalized predictions of time-to-event outcomes, such as patient survival. However, real-world SA datasets often suffer from data scarcity, heterogeneity, and privacy constraints, which limit the applicability of traditional and modern AI methods. To address these challenges, we propose the Federated Synthetic Data Sharing (FedSDS) framework, which integrates synthetic data generation with Federated Learning (FL). For SA, we leverage SAVAE, a state-of-the-art model for complex datasets. Using the Variational Autoencoder-Bayesian Gaussian Mixture model enhanced with artificial inductive bias, FedSDS generates high-quality synthetic data locally and shares them among nodes, enabling collaborative model training without direct data sharing. FedSDS introduces a biased aggregation strategy that aligns synthetic data with local distributions, outperforming traditional FL methods, such as Federated Average. Validated under independent and identically distributed (IID) and non-IID scenarios, FedSDS mitigates data imbalances and heterogeneity, showing significant performance improvements in scarce and heterogeneous data. The proposed framework offers a scalable and privacy-preserving solution for SA in decentralized environments. By enhancing model generalizability and robustness, FedSDS provides a promising path forward for collaborative analytics in healthcare, paving the way for improved patient outcomes and greater adoption of federated techniques in real-world applications.

The scarcity of medical data, particularly in Survival Analysis (SA) for cancer-related diseases, challenges data-driven healthcare research. While Synthetic Tabular Data Generation (STDG) models have been proposed to address this issue, most rely on datasets with abundant samples, not reflecting real-world limitations. We suggest using an STDG approach that leverages transfer learning and meta-learning techniques to create an artificial inductive bias, guiding generative models trained on limited samples. Initial experiments were conducted on larger classification datasets, allowing us to asses the methodology across varying sample sizes and abundant versus scarce data scenarios. We primarily employed clinical utility validation for cancer-related SA data, as divergence-based similarity validation was not feasible. The methodology improved STDG under constrained data conditions, with divergence-based similarity validation proving to be a robust measure of data quality. Conversely, clinical utility validation yielded similar results regardless of sample size, indicating its limitations in statistically confirming effective STDG. In SA experiments, we observed that altering the task can reveal if relationships among variables are accurately generated, with most cases benefiting from the proposed methodology. Our study underscores the efficacy of the approach in tackling medical data scarcity by effectively generating high-quality synthetic data under constrained conditions. While divergence-based similarity validation is essential when sufficient data are available, clinical utility validation alone is insufficient and should be complemented by similarity validation. These findings underscore the potential and limitations of STDG methodologies in addressing medical data scarcity.

Federated Learning (FL) is a machine learning framework that enables privacy-preserving collaboration between institutions. Instead of transferring sensitive data to a central server, FL allows individual entities to learn locally and shares only AI model parameters. FL has enormous potential in haematology, where scarce and sensitive data and diverse datasets are analysed. A platform to perform FL experiments has been developed by the EU project GenoMed4All, whose reliability and robustness must be tested. FL was implemented for Myelodysplastic Syndrome (MDS) and Sickle Cell Disease (SCD) datasets, of 4427 and 65 patients, respectively. For MDS, a DeepSurv model predicted overall survival, while for SCD, a Logistic Regression predicted the emergence of silent cerebral infarction. Models’ performances were assessed through C-index and F1-score, respectively. FL tests were conducted exploiting both simulations using Flower library and experiments utilizing the GenoMed4All platform. The MDS and SCD datasets were distributed across 3 clients, each one with a different number of samples. Simulations were performed on a single computer, that acts as central server and generates virtual clients and then compared to the FL platform results. FL aggregation allowed to improve models’ performance for both MDS and SCD use case. In particular, clients with less samples benefits from FL training. In general, federated training performed better than isolated training and allowed to achieve performances close to centralized training, where all datasets are merged. Numerical results from simulations and runs on FL platform were comparable and validated the platform robustness. Experiments confirmed the benefits of FL for MDS and SCD use cases and assessed the reliability of the GenoMed4All platform. Results’ improvements could be also achieved by different strategies of model parameters’ aggregation during the FL training. These findings open the possibility of implementing FL for other haematological tasks.

Background: Access to large, diverse, and high-quality datasets is essential for advancing research in rare hematological diseases (RHDs). However, privacy regulations and institutional silos often limit data sharing across centers. Synthetic data (SD) generation using deep learning offers a promising solution to augment clinical datasets while preserving patients confidentiality. Merging generative Artificial Intelligence (AI) training with federated learning (FL), which enables decentralized model training across institutions without data sharing, is particularly suited for healthcare applications. SD offer a series of transformative advantages for RHDs: by mimicking clinical and genomic characteristics of real patients without reproducing identifiable information, SD enable secure and scalable data sharing that respects privacy constraints. Furthermore, SD allow for data augmentation, missing data imputation and cohort balancement. SD have demonstrated high fidelity in replicating survival outcomes, molecular profiles, and complex gene-gene interactions observed in real populations. They have been successfully used to anticipate molecular classifications and prognostic scoring systems, demonstrating their potential to accelerate translational research. Furthermore, SD can support clinical trial innovation as synthetic control arms, reducing the need for placebo groups and streamlining trial design. Together, the integration of SD generation and FL offers a privacy-preserving, high-utility framework for advancing precision medicine and collaborative research in RHDs.

Aims: This project was conducted within the Synthema and Synthia Consortia. Specifically, the project aimed to 1) Implement and compare generative models within a FL framework to synthesize high-fidelity patient data while preserving privacy; 2) Validate the statistical, clinical, and privacy performance of generated SD using Synthetic vAlidation FramEwork (SAFE); 3) Assess the effectiveness of federated training strategies against centralized and isolated training scenarios to benchmark performance and ensure scalability in real-world clinical research networks.

Methods: A multi-institutional simulation was conducted using a myelodysplastic syndromes (MDS) dataset of 4427 patients distributed across three federated nodes. Several generative models (CTGAN, Bayesian Networks, and VAE-BGM) were trained under four FL strategies, including Federated Averaging and SD sharing. Data quality was assessed using the SAFE framework, which evaluates statistical fidelity, clinical relevance (including genomic associations and survival analysis), and privacy risk (via Nearest Neighbor Distance Ratio, NNDR). Model performance was compared against centralized and isolated training settings.

Results: Across the FL training rounds, SD demonstrated strong alignment with real-world data. Models trained via FL achieved high Clinical Synthetic Fidelity (CSF) and Genomic Synthetic Fidelity (GSF) scores, comparable to the centralized (upper-bound) setting and clearly superior to isolated node training. By training round 5, CSF and GSF reached 0.942 and 0.902 respectively, indicating a high degree of statistical and clinical similarity between real and SD. Privacy metrics, including the Nearest Neighbor Distance Ratio (NNDR), confirmed that the SD maintained strong privacy safeguards. Performance improvements were observed across rounds, as shown by the increasing fidelity scores, illustrating the learning benefits of model collaboration without data sharing. Furthermore, survival analysis and gene mutation frequency comparisons confirmed clinical utility of SD. SD preserved key genomic patterns and patient outcome distributions, validating their use in applications such as risk stratification and biomarker discovery.

Conclusions: This study demonstrates that FL can successfully generate high-fidelity SD while preserving patient privacy and data integrity. Generative AI trained with FL not only retain clinically relevant information but also achieve performance metrics comparable to centralized training, without requiring direct data sharing. These findings support the use of federated SD generation as a scalable and privacy-preserving solution for enabling secure multi-institutional research collaborations, particularly in settings where access to comprehensive medical data is limited, and for the development of personalized precision medicine AI models.

While synthetic tabular data generation using Deep Generative Models (DGMs) offers a compelling solution to data scarcity and privacy concerns, their effectiveness relies on the availability of substantial training data, often lacking in real-world scenarios. To overcome this limitation, we propose a novel methodology that explicitly integrates artificial inductive biases into the generative process to improve data quality in low-data regimes. Our framework leverages transfer learning and meta-learning techniques to construct and inject informative inductive biases into DGMs. We evaluate four approaches (pre-training, model averaging, Model-Agnostic Meta-Learning (MAML), and Domain Randomized Search (DRS)) and analyze their impact on the quality of the generated text. Experimental results show that incorporating inductive bias substantially improves performance, with transfer learning methods outperforming meta-learning, achieving up to 60 % gains in Jensen-Shannon divergence. The methodology is model-agnostic and especially relevant in domains such as healthcare and finance, where high-quality synthetic data are essential, and data availability is often limited.

Accurate survival prediction is critical in oncology for prognosis and treatment planning. Traditional approaches often rely on a single data modality, limiting their ability to capture the complexity of tumor biology. To address this challenge, we introduce a multimodal deep learning framework for survival analysis capable of modeling both single and competing risks scenarios, evaluating the impact of integrating multiple medical data sources on survival predictions. We propose SAMVAE (Survival Analysis Multimodal Variational Autoencoder), a novel deep learning architecture designed for survival prediction that integrates six data modalities: clinical variables, four molecular profiles, and histopathological images. SAMVAE leverages modality specific encoders to project inputs into a shared latent space, enabling robust survival prediction while preserving modality specific information. Its parametric formulation enables the derivation of clinically meaningful statistics from the output distributions, providing patient-specific insights through interactive multimedia that contribute to more informed clinical decision-making and establish a foundation for interpretable, data-driven survival analysis in oncology. We evaluate SAMVAE on two cancer cohorts breast cancer and lower grade glioma applying tailored preprocessing, dimensionality reduction, and hyperparameter optimization. The results demonstrate the successful integration of multimodal data for both standard survival analysis and competing risks scenarios across different datasets. Our model achieves competitive performance compared to state-of-the-art multimodal survival models. Notably, this is the first parametric multimodal deep learning architecture to incorporate competing risks while modeling continuous time to a specific event, using both tabular and image data.

Deep neural networks achieve state-of-the-art performance in estimating heterogeneous treatment effects, but their opacity limits trust and adoption in sensitive domains such as medicine, economics, and public policy. Building on well-established and high-performing causal neural architectures, we propose causalKANs, a framework that transforms neural estimators of conditional average treatment effects (CATEs) into Kolmogorov–Arnold Networks (KANs). By incorporating pruning and symbolic simplification, causalKANs yields interpretable closed-form formulas while preserving predictive accuracy. Experiments on benchmark datasets demonstrate that causalKANs perform on par with neural baselines in CATE error metrics, and that even simple KAN variants achieve competitive performance, offering a favorable accuracy–interpretability trade-off. By combining reliability with analytic accessibility, causalKANs provide auditable estimators supported by closed-form expressions and interpretable plots, enabling trustworthy individualized decision-making in high-stakes settings.

Why should a clinician trust an Artificial Intelligence (AI) prediction? Despite the increasing accuracy of machine learning methods in medicine, the lack of transparency continues to hinder their adoption in clinical practice. In this work, we explore Kolmogorov-Arnold Networks (KANs) for clinical classification tasks on tabular data. In contrast to traditional neural networks, KANs are function-based architectures that offer intrinsic interpretability through transparent, symbolic representations. We introduce \emph{Logistic-KAN}, a flexible generalization of logistic regression, and \emph{Kolmogorov-Arnold Additive Model (KAAM)}, a simplified additive variant that delivers transparent, symbolic formulas. Unlike ``black-box’’ models that require post-hoc explainability tools, our models support built-in patient-level insights, intuitive visualizations, and nearest-patient retrieval. Across multiple health datasets, our models match or outperform standard baselines, while remaining fully interpretable. These results position KANs as a promising step toward trustworthy AI that clinicians can understand, audit, and act upon. We release the code for reproducibility in \codeurl.

The integration of Artificial Intelligence (AI) in healthcare holds significant promise, yet remains constrained by data scarcity, privacy and ethical concerns, and the inherent complexity of medical tasks. This review explores the intersection of deep generative modeling and federated learning, two complementary paradigms that enable collaborative and privacy-preserving innovation. We survey recent trends in deep generative models applied to healthcare, with a particular focus on federated settings. The review is structured around model architectures, providing a taxonomy-driven analysis that emphasizes solutions tailored to the heterogeneity of healthcare data. Although this review initially intended to cover multimodal applications, the current literature is largely limited to single-modality models. Among deep generative approaches, Generative Adversarial Networks dominate the field, particularly in domains such as medical image synthesis and data augmentation. In contrast, alternative architectures, including variational autoencoders, diffusion models, and autoregressive models, remain barely explored in federated healthcare scenarios. Notably, the number of related publications has increased substantially in recent years, rising from 1 in 2020 to 16 in 2024, reflecting the growing interest and momentum in this area. This paper highlights key challenges, open research questions, and future directions for the development of trustworthy, distributed generative AI in healthcare.

Background. Rare haematological diseases (RHD) pose significant clinical challenges due to their heterogeneity, limited patient populations, and fragmented datasets. To overcome these limitations, improve access to, and use of real-world multimodal data for scientific and clinical purposes, the GenoMed4All Consortium developed an open-source Federated Learning (FL) platform. This platform enables collaborative, privacy-preserving AI model training without the need to centralize sensitive patient information. Methods. The FL platform was deployed within EuroBloodNet, the European Reference Network for RHD, across multiple use cases, including myelodysplastic syndromes (MDS), acute myeloid leukemia (AML), chronic myelomonocytic leukemia (CMML), and multiple myeloma (MM). Multimodal datasets (including clinical, genomic information together with histopathological and radiological extracted features) were utilized. Predictive models (DeepSurv and SAVAE) and generative Artificial intelligence (AI) algorithms (CTGAN, Bayesian Networks, and VAE-BGM) were trained using a federated approach. A dedicated data harmonization pipeline based on the FHIR standard ensured consistency across participating centers. Findings. Federated models achieved performance comparable to centralized approaches, with highest benefit for institutions with smaller datasets. The platform enabled integration of multimodal data demonstrating flexibility across diverse data types and clinical endpoints. The inclusion of multimodal information improved predictive accuracy over currently available prognostic schemes. Generative models successfully created synthetic datasets that preserved both clinical and statistical fidelity while ensuring patient privacy; this allows extraction of insights from real-world data that can be used beyond the boundaries of FL, as a source for accelerating the conduction of clinical trials. A preliminary implementation within the EuroBloodNet clinical network demonstrated feasibility for broader scale-up. Interpretation. This study validates FL as a robust, privacy-compliant approach to enable AI-driven precision medicine in RHD. The platform facilitates real-world data integration and model scalability, providing a foundation for multicenter collaboration, regulatory-grade evidence generation, and innovative trial designs in rare diseases. Funding. European Union’s Horizon 2020 research and innovation programme.

Membership attacks pose a major issue in terms of secure machine learning, especially in cases in which real data are sensitive. Models tend to be overconfident in predicting labels from the training set. Nevertheless, its application has traditionally been limited to supervised models, while in the case of generative models we have found that there is a lack of theoretical foundations to bring this concept into the scene. Hence, this article provides the theoretical background in the context of membership inference attacks and their relationship to generative models, including the derivation of an evaluation metric. In addition, the link between these types of attack and differential privacy is shown to be a particular case. Lastly, we empirically show through simulations the intuition and application of the concepts derived.

Federated Learning is an emerging, powerful approach that allows training an artificial intelligence model in distributed setting. Two survival models, Cox and DeepSurv, have been trained in a federated setting, exploiting both code simulations and real experiments on the new platform, developed by the GenoMed4All consortium. Different scenarios have been tested by splitting a Myelodysplastic Syndrome dataset into three nodes and performing feature removal. A significant gain in model performance has been observed due to federated aggregation.

BACKGROUND. Most oncological and non-oncological hematological conditions fall under the category of rare diseases. Rare diseases present unique challenges due to the limited availability of data, which impacts diagnostic rates and the generation of clinical evidence. Overall, they constitute a public health concern, highlighting the urgent need to develop new methods for improving data accessibility. In this context, Federated Learning (FL) is a Machine Learning approach that allows multiple centers to collaborate on complex research questions without the need to centralize or share data. This project was conducted by the Genomed4all and Synthema consortia with the goal of developing an innovative FL platform for rare hematological diseases. This platform enables the development of novel Artificial Intelligence (AI) models for personalized medicine without data sharing, to be implemented in the referral centers of EuroBloodNET, the European Reference Network for rare hematological diseases. The aims of the project were: 1) to develop robust federated models for personalized prediction using multicentric, real-world datasets; 2) to protect patients’ privacy; and 3) to enhance collaboration between institutions while avoiding the creation of centralized data repositories. METHODS. The FL platform includes a manager node (MN) and multiple worker nodes (WN). Users upload their model to the MN, which distributes it to WNs for local training. Trained weights are returned to the MN for aggregation, repeating until training is complete. The MN is hosted at the Humanitas Research Hospital, and the 3 WNs are located at Humanitas, Universität Leipzig, and the University of Bologna. The testing approach is based on MOSAIC, an AI-based framework for multimodal analysis in rare cancers (PMID 38875514). We focused on myelodysplastic syndromes as a use case, since they are a rare hematological disease with high clinical and genomic heterogeneity (which represents a challenging scenario). The dataset includes comprehensive clinical and genomic data, with information on treatments and clinical outcomes. Each of the 3 WNs has a different sample size of patient data, with 2656, 1328, and 443 samples. Our goal was to build an AI-based model for prediction of survival probability, combining molecular information with clinical data. We planned 3 scenarios for testing: a best-case scenario using all available features, a worst-case scenario excluding up to 50% of genomic information, and a random scenario where an incremental percentage of the genomic variables were randomly removed. For each setting, a DeepSurv model is trained using 80% of local data for training and 20% for testing. RESULTS. In the best-case scenario, the model achieved a concordance index (c-index) of 0.75. The node with the smallest dataset initially registered the lowest c-index value (0.4); however, after the first training step, its performance began to follow the trend at levels comparable to the nodes with larger samples (0.54). In the random scenarios, the model still achieved a c-index of 0.75 but required more training rounds to converge. The same pattern was also observed in the worst-case scenario that was able to achieve a c-index of 0.74. The most relevant difference between the tested scenarios is the increased training time required to reach a good and stable performance when missing information is present. The developed technology is fully compliant with European GDPR regulations. In the upcoming months, the FL platform will be implemented in the referral centers of the EuroBloodNET clinical network. In addition to tabular data, we plan to address the federation of features extracted from medical images, including both histological slides and radiological scans (CT, MRI and PET-CT). CONCLUSION. The FL platform for rare hematological diseases allows multicentric training of AI models without sharing sensitive patient data. This approach ensures data privacy and security, addressing specific challenges associated with rare diseases. Moreover, this platform manages scenarios with missing data or variables, maintaining robustness and accuracy of predictive models. Overall, FL approach is expected to facilitate the development of more advanced and reliable healthcare solutions, paving the way toward personalized medicine in hematology.

Competing risks in survival analysis pose a significant challenge in healthcare, but few methods effectively address this problem. Moreover, the use of deep learning techniques remains limited, with the most prevalent approach, DeepHit, being non-parametric. This often leads to limitations in inter-pretability and statistical inference, which are crucial aspects. We propose a new alternative that harnesses the power of variational autoencoders and deep learning to tackle survival analysis with competing risks within a parametric framework. Our model, CR-SAVAE, allows direct interpretation of covariate effects on survival outcomes and enables researchers to perform robust statistical analysis compared to non-parametric approaches, essential for understanding risk mechanisms and making informed clinical decisions. It provides personalized medicine insight by accurately estimating the cumulative incidence function and avoiding the need for proportional hazards assumptions inherent in other models. Through comprehensive experiments on datasets with varying degrees of censoring and competing risks, we demonstrate the potential of our approach to achieve performance comparable to that of DeepHit based on the concordance index and integrated Brier score. This study highlights the potential of CR-SAVAE to advance survival analysis, improve interpretability, and enable more accurate and personalized clinical decision making in healthcare settings.

In recent years, passive vehicle safety has become one of the major concerns for the automotive industry due to the considerable increase in the use of cars as a means of daily transport. Since real crash testing has a high financial cost, finite element simulations are generally used, which entail high computational cost and long simulation times. In this paper, we make use of the recent advances in the deep learning field to propose an affordable method to provide reliable approximations of the finite element simulator model that significantly reduce the computational load and time required. We compare the prediction performance in crash tests of different models, namely feed-forward neural networks and bayesian neural networks, as well as two multi-output regression methods. Our results show promising results, as deep learning models are able to drastically reduce the engineering costs while providing a feasible first approximation to the passenger’s injuries in a crash event, thus being a potential game changer in the vehicle safety design process.

Rare cancers constitute over 20% of human neoplasms, often affecting patients with unmet medical needs. The development of effective classification and prognostication systems is crucial to improve the decision-making process and drive innovative treatment strategies. We have created and implemented MOSAIC, an artificial intelligence (AI)–based framework designed for multimodal analysis, classification, and personalized prognostic assessment in rare cancers. Clinical validation was performed on myelodysplastic syndrome (MDS), a rare hematologic cancer with clinical and genomic heterogeneities. We analyzed 4,427 patients with MDS divided into training and validation cohorts. Deep learning methods were applied to integrate and impute clinical/genomic features. Clustering was performed by combining Uniform Manifold Approximation and Projection for Dimension Reduction + Hierarchical Density-Based Spatial Clustering of Applications with Noise (UMAP + HDBSCAN) methods, compared with the conventional Hierarchical Dirichlet Process (HDP). Linear and AI-based nonlinear approaches were compared for survival prediction. Explainable AI (Shapley Additive Explanations approach [SHAP]) and federated learning were used to improve the interpretation and the performance of the clinical models, integrating them into distributed infrastructure. UMAP + HDBSCAN clustering obtained a more granular patient stratification, achieving a higher average silhouette coefficient (0.16) with respect to HDP (0.01) and higher balanced accuracy in cluster classification by Random Forest (92.7% ± 1.3% and 85.8% ± 0.8%). AI methods for survival prediction outperform conventional statistical techniques and the reference prognostic tool for MDS. Nonlinear Gradient Boosting Survival stands in the internal (Concordance-Index [C-Index], 0.77; SD, 0.01) and external validation (C-Index, 0.74; SD, 0.02). SHAP analysis revealed that similar features drove patients’ subgroups and outcomes in both training and validation cohorts. Federated implementation improved the accuracy of developed models. MOSAIC provides an explainable and robust framework to optimize classification and prognostic assessment of rare cancers. AI-based approaches demonstrated superior accuracy in capturing genomic similarities and providing individual prognostic information compared with conventional statistical methods. Its federated implementation ensures broad clinical application, guaranteeing high performance and data protection.

The recent advances in Deep Learning have a significant impact on the security of wireless networks, such as intelligent attackers which are able to successfully exploit a possibly unknown defense mechanism simply by interacting with it. Their capacity of adapting to standard defense mechanisms, such as statistical tests, makes them a significant threat. In this article, we develop two intelligent defense mechanisms using inverse reinforcement learning tools, that can be used to enhance the capabilities of current defense mechanisms. We test our proposal on a backoff attack setup against an intelligent attacker, obtaining very significant gains in the defense performance.

Discounted repeated games are currently being used to model the conflicts that arise between the nodes in a wireless network, such as distributed resource allocation, interference management or defending the network against attacks. In current literature, it is frequent that authors devise a specific strategy that performs well only for their concrete problem, thus, it would be desirable to have a generic algorithm that allows learning strategies for such games. However, current learning algorithms focus on average payoff repeated games, and we show analytically that there are important differences that prevent us from using such algorithms for discounted repeated games. In this work, we aim to fill this gap and we propose LEWIS, a lightweight, online learning algorithm specifically designed for these games, that deals with imperfect and incomplete information and is able to return a good payoff. We test LEWIS on two settings based on current literature problems to show that it has a good performance, hence, being a promising method to learn how to play a discounted repeated game in wireless networks.

We use the recent advances in Deep Learning to solve an underwater motion planning problem by making use of optimal control tools—namely, we propose using the Deep Galerkin Method (DGM) to approximate the Hamilton–Jacobi–Bellman PDE that can be used to solve continuous time and state optimal control problems. In order to make our approach more realistic, we consider that there are disturbances in the underwater medium that affect the trajectory of the autonomous vehicle. After adapting DGM by making use of a surrogate approach, our results show that our method is able to efficiently solve the proposed problem, providing large improvements over a baseline control in terms of costs, especially in the case in which the disturbances effects are more significant.