In real-world applications, audio surveillance is often performed by large models that can detect many types of anomalies. However, typical approaches are based on centralized solutions characterized by significant issues related to privacy and data transport costs. In addition, the large size of these models prevented a shift to contexts with limited resources, such as edge devices computing. In this work we propose conv-SPAD, a method for convolutional SPectral audio-based Anomaly Detection that takes advantage of common tools for spectral analysis and a simple autoencoder to learn the underlying condition of normality of real scenarios. Using audio data collected from real scenarios and artificially corrupted with anomalous sound events, we test the ability of the proposed model to learn normal conditions and detect anomalous events. It shows performances in line with larger models, often outperforming them. Moreover, the model’s small size makes it usable in contexts with limited resources, such as edge devices hardware.

Cutaneous melanoma remains the most lethal form of skin cancer. Although incurable at advanced stages, if diagnosed at an early, localized stage, the five-year survival rate is remarkably high. Recent advancements in artificial intelligence have paved the way for early skin lesion diagnosis, leveraging digital imaging processes into effective solutions. Most of these, however, use Machine Learning and Deep Learning techniques compartmentalized, without combining the produced predictions. This paper introduces MultiExCam, a novel multi approach and explainable architecture for skin cancer detection that integrates both machine and deep learning. Three heterogeneous data from three different techniques are used: dermatoscopic images, features extracted from deep learning techniques, and hand-crafted statistical features. A convolutional neural network is used for both deep feature extraction and initial classification, with the extracted features being combined with handcrafted ones to train four additional machine learning models. An advanced ensemble model, implemented as a Feed Forward Neural Network with gating and attention mechanism, produces the final classification. To enhance interpretability, the architecture employs GradCAM for visualizing critical regions in input images and SHAP for evaluating the contribution of individual features to predictions. MultiExCam demonstrates robust performance across three diverse datasets (HAM10000, ISIC, MED-NODE), achieving AUC scores of 97%, 91%, and 98% respectively, with corresponding F1-scores of 92%, 87%, and 94%. Comprehensive ablation studies validate the importance of the preprocessing pipeline and ensemble integration, with the hybrid approach consistently outperforming baseline deep learning models by 1-3 percentage points. Unlike existing compartmentalized hybrid solutions, MultiExCam's adaptive ensemble architecture learns personalized decision strategies for individual lesions, mimicking expert dermatological workflows that integrate multiple evidence sources. The explainability analysis reveals clinically meaningful activation patterns corresponding to established diagnostic criteria including asymmetry, border irregularity, and color variation. MultiExCam establishes a new paradigm for AI-assisted dermatological diagnosis by demonstrating that true hybrid integration of deep learning and machine learning, combined with comprehensive explainability techniques, can achieve both superior diagnostic performance and clinical interpretability. The architecture's ability to provide accurate classifications while explaining prediction rationale addresses critical requirements for medical AI adoption, offering a promising foundation for clinical decision support systems in melanoma detection.

Recommendation systems play a crucial role in modern e-commerce and streaming services. However, the limited availability of public datasets hampers the rapid development of more efficient and accurate recommendation algorithms within the research community. This work introduces a stream-based data generator designed to generate user preferences for a set of items while accommodating progressive changes in user preferences. The underlying principle involves using user/item embeddings to derive preferences by exploring the proximity of these embeddings. Whether randomly generated or learned from a real finite data stream, these embeddings serve as the basis for generating new preferences. We investigate how this fundamental model can adapt to shifts in user behavior over time; in our framework, changes correspond to alterations in the structure of the tripartite graph, reflecting modifications in the underlying embeddings. Through an analysis of real-life data streams, we demonstrate that the proposed model is effective in capturing actual preferences and the changes that they can exhibit over time. Thus, we characterize these changes and develop a generalized method capable of simulating realistic data, thereby generating streams with similar yet controllable drift dynamics.