Background. Deep neural networks (DNNs) are promising for analyzing high-dimensional transcriptomic data in cancer research but are limited by data scarcity and heterogeneity.

Transfer learning (TL), which leverages large datasets to improve performance on smaller ones, has been widely applied in classification and clustering, yet its use in survival prediction remains underexplored.

Methods. We evaluated transfer learning for disease-free survival prediction using RNA-seq data from 7,509 patients across 27 tumor types from The Cancer Genome Atlas. A pan-cancer DNN was optimized via random search across 750 configurations. Tumor-specific baseline models were trained independently, while transfer learning models were built with a leave-one-out pre-training strategy followed by fine-tuning on the target tumor. Performance was assessed with the concordance index. Biological relevance was evaluated through a LIME-like interpretability approach and gene-set enrichment analysis.

Results. Transfer learning outperformed tumor-specific baselines in 24 of 27 tumor types, with the greatest improvements in smaller cohorts. Fine-tuning consistently enhanced predictions by adapting pan-cancer features to tumor-specific data. Enrichment analysis confirmed that the models captured both general oncogenic mechanisms and tumor-specific pathways.

Conclusions. Transfer learning improves survival prediction from transcriptomic data, especially in small cohorts, by exploiting molecular patterns shared across cancers. These findings support transfer learning as a robust strategy for survival modeling and a step toward foundation models in molecular oncology.

Growing awareness of environmental issues and the need for sustainable resource management increasingly underscore the importance of water as a finite and valuable resource. Despite this recognition, many water distribution systems suffer from significant inefficiencies, including excessive leakage, limited preventative measures, and a predominately reactive maintenance approach that often leaves chronic leaks unresolved. These challenges are exacerbated by rising water demand, aging infrastructure, and minimal digital oversight, all of which complicate leak detection and localization.

To address these issues, this study introduces methods for clustering system components into geographic districts based on similar leakage behaviors. This “districting” strategy aims to facilitate targeted maintenance by highlighting the most critical areas of the network. We compare our methods with established clustering techniques - such as K-Means, Spectral Clustering, and graph-based algorithms - and explore the impact of the different hyperparameters. Experiments conducted on a modeled urban dataset from the BattLeDIM project demonstrate the effectiveness of our proposed approach in enhancing network segmentation and optimizing leak management. The presented results offer valuable insights for improving the overall performance of water distribution systems.

Clathrate hydrates enable the storage of hydrocarbons in solid form with considerably lower energy requirements than conventional methods. In this study, we use Machine Learning (ML) techniques to develop and validate a model that describes clathrate formation, drawing on laboratory data from propane, ethane, methane, and carbon dioxide.

The ML-based framework augments traditional analytical and theoretical approaches by predicting equilibrium points of gas compounds and guiding the process parameters to meet scientific and engineering needs. In addition, we address critical challenges in framing the problem for ML modeling, including selecting pertinent input and output variables and incorporating ML outputs into the physical-chemical equations, thereby providing a physics-informed estimate of the stored gas volume.

Modern power grids are becoming increasingly complex with the integration of heterogeneous distributed energy resources, underscoring the need for accurate and efficient Power Flow Analysis to ensure stability, reliability, and market operations. Existing methods generally rely on iterative numerical techniques (INT) or machine learning (ML). While INT is physically consistent and highly accurate, it can be computationally expensive and vulnerable to slow or non-convergence. ML methods offer faster solutions but often require extensive data, suffer from limited extrapolation capabilities, and lack physical consistency. Physics-informed ML (PIML) bridges these gaps by embedding domain knowledge before, during, and after training. However, current PIML approaches typically do not leverage this full range of opportunities. In this paper, we propose a novel PIML framework for Power Flow Analysis that integrates physical insights at all three stages (pre, in, and post-processing) to achieve superior accuracy and efficiency. Notably, we introduce a new post-processing technique that partitions the power network into its mesh and radial components: the mesh portion is handled via PIML, while the radial portion is efficiently solved with a convex optimization approach informed by the PIML outputs.

This approach is efficient with radial topologies, especially in power distribution networks where the radial part is predominant. Experiments on realistic power networks demonstrate that our method outperforms state-of-the-art approaches in both accuracy and computational performance.

In contemporary industrial applications, predictive models have been pivotal in bolstering production efficiency, product quality, scalability, and cost-effectiveness while promoting sustainability. These predictive models can be constructed solely based on domain-specific knowledge, exclusively on observational data, or by amalgamating both approaches. They are commonly referred to as Full-, Zero-, or Partial-knowledge-based predictive models, respectively. Full-knowledge based models are highly explainable, point-wise accurate, data efficient, computationally demanding in the prediction phase to achieve high accuracy. Zero-knowledge based models are poorly explainable, data hungry, highly accurate on average, computationally demanding in the construction phase but inexpensive the prediction phase. Partial-knowledge based models focus on taking the best of the two worlds. To maintain a focused scope and avoid redundancy with existing literature, our review will primarily delve into the key industrial applications within a narrow context, encompassing sectors such as extraction, chemical processes, manufacturing, transportation, energy, and construction. Our approach entails several steps. Initially, we conducted a meta-review to pinpoint gaps in previously published surveys on Full-, Zero-, and Partial-knowledge-based predictive models for industrial applications. Subsequently, we present a formal analysis of the subject matter, supplemented with illustrative examples to offer valuable insights. The core of our work comprises a review of existing research categorized by specific industrial applications.

Finally, we outline the unresolved challenges and future prospects in this burgeoning field of research. We contend that our work serves as a valuable resource, catering to the needs of both young researchers seeking a solid foundation for their studies, industrial practitioners aiming to grasp core concepts and applications, and senior researchers seeking potential real-world applications for their findings.

With the rapid growth of online space and the rising number of interconnected devices, security threats related to both personal and corporate data have increased considerably. Phishing attacks are commonly used to target corporate networks and gain initial access into security perimeters. The campaigns associated to these attacks span different propagation media and, in the case of web pages, attackers mimic real pages to trick users into downloading malicious software or sharing their credentials. In this paper we propose PhishVision, a framework for visually detecting phishing websites by identifying the main logo that characterizes them and comparing it with a set of logos which PhishVision protects. In case of presence of multiple logos, the framework is able to reconstruct which logo identifies the page while ignoring the others. The framework has been designed to have a lower false positive rate, fast detection times, and works in near-real-time fashion to provide a phishing detection service to Security Operation Centers. Its operators can use it make informed decisions about potential phishing activities by offering a comprehensible grey box explanation about how the framework has reached its conclusions. PhishVision achieves 0.997 ROC AUC on a test set of 404 screenshots, including both benign and malicious samples.

Flooding is one of the most frequent natural disasters worldwide, resulting in substantial socioeconomic losses and public health threats. The EASTERN project proposes an innovative approach exploiting Artificial Intelligence (AI) techniques to combine data collected from Synthetic Aperture Radar (SAR) imaging and ground measurements for real-time flood risk assessment. Specifically, the goal is to focus on both immediate dangers associated with landslides and secondary risks resulting from the increased prevalence of disease vectors in affected regions. For the first use case, data from GNSS (Global Navigation Satellite System) technologies and interferometric analysis of synthetic aperture radar images (InSAR) will be combined to offer complementary insights into Earth’s surface deformation and identify susceptible locations prone to landslides. For the second use case, high resolution SAR data will be exploited to predict whether a flooded area may become an ecological niche for arbovirosis vectors.

The application of advanced AI technologies for these Earth Observation tasks will allow for a prompt response to flooding events and will become a valuable supportinthedecision-makingprocessofpreventingand mitigating the consequences of extreme weather events