The effective interpretation of medical time-series data remains a significant challenge in AI diagnostics, as temporal patterns critical for diagnosis are often obscured in conventional “black-box” models. This paper presents MedAttnAID, an explainable temporal attention framework designed for unsupervised anomaly detection in sensor-generated medical time-series data. Our approach integrates a bidirectional LSTM autoencoder with interpretable attention mechanisms that highlight diagnostically relevant temporal segments, complemented by a novel Clinical Pattern Loss (CPLoss) function that preserves clinically significant temporal patterns during reconstruction. We validate the framework on thermal time-series data from wearable breast cancer monitoring devices, achieving$85.1 \% ~\mathrm{F} 1$score and 95 % breast-level localization accuracy while producing interpretable attention heatmaps that align with clinical expectations. The framework's modular architecture addresses fundamental challenges common to sensor-based medical monitoring-limited labeled pathological data, clinical interpretability requirements, and detection of subtle anomalies in complex temporal pat-terns-making it broadly applicable to diverse sensing modalities including ECG, EEG, and continuous physiological monitoring systems.

International audience

Workload-driven approaches for partitioning and tuning traditional Parallel Database systems are well studied in the literature. Unfortunately, in the context of new generation "Big Data" warehouses, these approaches are not correctly adapted to Business Intelligence 2.0, where the analyst is at the heart of decision support systems. This "disconnect" situation strongly impacts both data partitioning and fragment allocation processes, which are essential to achieve good query performance. To overcome this problem, recent studies proposed online data partitioning and fragment allocation using AI techniques to improve query performance with adaptive behavior. Nevertheless, they have important limitations: they add significant overhead and they tend to focus on the current workload, ignoring query logs. With such motivation in mind, we first formulate the problem of optimizing database partitioning subject to feasibility constraints, based on a query workload. We then introduce a proactive partitioning approach combining offline and online processing phases, inspired by closed-loop control (used in engineering disciplines) and genetic algorithms (from AI). We present an experimental validation on a big data cluster that shows promising results on typical OLAP workloads.

This paper complements our previous results in the context of effectively and efficient designing Parallel Relational Data Warehouses (PRDW) over heterogeneous database clusters, which are represented by the proposal of a methodology called Fragmentation & Allocation (F& A). The main merit of F& A is that of combining the fragmentation and the allocation phases simultaneously, which are instead performed separately by traditional approaches. In this paper, we prove the practical impact and the reliability of F& A on a real-life parallel processing database system.

In this paper, a comprehensive methodology for designing and querying Parallel Rational Data Warehouses (PRDW) over database clusters, called Fragmentation & Allocation (F&A) is proposed. F&A assumes that cluster nodes are heterogeneous in processing power and storage capacity, contrary to traditional design approaches that assume that cluster nodes are instead homogeneous, and fragmentation and allocation phases are performed in a simultaneous manner. In classical approaches, two different cost models are used to perform fragmentation and allocation, separately, whereas F&A makes use of one cost model that considers fragmentation and allocation parameters simultaneously. Therefore, according to the F&A methodology proposed, the allocation phase/decision is done at fragmentation. At the fragmentation phase, F&A uses two well-known algorithms, namely Hill Climbing (HC) and Genetic Algorithm (GA), which the authors adapt to the main PRDW design problem over heterogeneous database clusters, as these algorithms are capable of taking into account the heterogeneous characteristics of the reference application scenario. At the allocation phase, F&A introduces an innovative matrix-based formalism capable of capturing the interactions among fragments, input queries, and cluster node characteristics, driving the data allocation task accordingly, and a related affinity-based algorithm, called F&A-ALLOC. Finally, their proposal is experimentally assessed and validated against the widely-known data warehouse benchmark APB-1 release II.