Abstract
The main challenges limiting the adoption of deep learning-based solutions in medical workflows are the availability of annotated data and the lack of interpretability of such systems. Concept Bottleneck Models (CBMs) tackle the latter by constraining the model output on a set of predefined and human-interpretable concepts. However, the increased interpretability achieved through these concept-based explanations implies a higher annotation burden. Moreover, if a new concept needs to be added, the whole system needs to be retrained. Inspired by the remarkable performance shown by Large Vision-Language Models (LVLMs) in few-shot settings, we propose a simple, yet effective, methodology, CBVLM, which tackles both of the aforementioned challenges. First, for each concept, we prompt the LVLM to answer if the concept is present in the input image. Then, we ask the LVLM to classify the image based on the previous concept predictions. Moreover, in both stages, we incorporate a retrieval module responsible for selecting the best examples for in-context learning. By grounding the final diagnosis on the predicted concepts, we ensure explainability, and by leveraging the few-shot capabilities of LVLMs, we drastically lower the annotation cost. We validate our approach with extensive experiments across four medical datasets and twelve LVLMs (both generic and medical) and show that CBVLM consistently outperforms CBMs and task-specific supervised methods without requiring any training and using just a few annotated examples. More information on our project page: https://cristianopatricio.github.io/CBVLM/. © 2025 Elsevier B.V., All rights reserved.

Abstract

Abstract
Planning Energy communities (ECs) requires engaging members, designing business models and governance rules, and sizing distributed energy resources (DERs) for a cost-effective investment. Meanwhile, the growing share of non-dispatchable renewable generation demands more flexible energy systems. Local flexibility markets (LFMs) are emerging as effective mechanisms to procure this flexibility, granting ECs a new revenue stream. Since sizing with flexibility becomes a highly complex problem, we propose a 2-stage methodology for estimating DERs size in an EC with collective self-consumption, flexibility provision and cross-sector (CS) assets such as thermal loads and electric vehicles (EVs). The first stage computes the optimal DER capacities to be installed for each member without flexibility provision. The second stage departs from the first stage capacities to assess how to modify the initial capacities to profit from providing flexibility. The impact of data clustering and flexibility provision are assessed through a case study.

Abstract
Underwater wireless communications face significant challenges due to propagation constraints, limiting the effectiveness of traditional radio and optical technologies. Long-range acoustic communications support distances up to a few kilometers, but suffer from low bandwidth, high error ratios, and multipath interference. Semantic communications, which focus on transmitting extracted semantic features rather than raw data, present a promising solution by significantly reducing the volume of data transmitted over the wireless link. This paper evaluates the resilience of SAGE, a semantic-oriented communications framework that combines semantic processing with Generative Artificial Intelligence (GenAI) to compress and transmit image data as textual descriptions over acoustic links. To assess robustness, we use a custom-tailored simulator that introduces character errors observed in underwater acoustic channels. Evaluation results show that SAGE can successfully reconstruct meaningful image content even under varying error conditions, highlighting its potential for robust and efficient underwater wireless communication in harsh environments.

Abstract

Abstract
Human Activity Recognition (HAR) is the identification and classification of static and dynamic human activities, which find applicability in domains like healthcare, entertainment, security, and cyber-physical systems. Traditional HAR approaches rely on wearable sensors, vision-based systems, or ambient sensing, each with inherent limitations such as privacy concerns or restricted sensing conditions. Instead, Radio Frequency (RF)-based HAR relies on the interaction of RF signals with people to infer activities. Reconfigurable Intelligent Surfaces (RISs) are significant for this use-case by allowing dynamic control over the wireless environment, enhancing the information extracted from RF signals. We present an Hand Gesture Recognition (HGR) approach using our own 6.5GHz RIS design, which we use to gather a dataset for HGR classification for three different hand gestures. By employing two Convolutional Neural Networks (CNNs) models trained on data gathered under random and optimized RIS configuration sequences, we achieved classification accuracies exceeding 90%.