Smart healthcare relies on artificial intelligence (AI) functions for learning and analysis of patient data. Since large and diverse datasets for training of Machine Learning (ML) models can rarely be found in individual medical centers, classical centralized AI requires moving privacy-sensitive data from medical institutions to data centers that process the fused information. Training on data centers thus requires higher communication resource/energy demands while violating privacy. This is considered today as a significant bottleneck in pursuing scientific collaboration across trans-national clinical medical research centers. Recently, federated learning (FL) has emerged as a distributed AI approach that enables the cooperative training of ML models, without the need of sharing patient data. This paper dives into the analysis of different FL methods and proposes a real-time distributed networking framework based on the Message Queuing Telemetry Transport (MQTT) protocol. In particular, we design a number of solutions for ML over networks, based on FL tools relying on a parameter server (PS) and fully decentralized paradigms driven by consensus methods. The proposed approach is validated in the context of brain tumor segmentation, using a modified version of the popular U-NET model with representative clinical datasets obtained from the daily clinical workflow. The FL process is implemented on multiple physically separated machines located in different countries and communicating over the Internet. The real-time test-bed is used to obtain measurements of training accuracy vs. latency trade-offs, and to highlight key operational conditions that affect the performance in real deployments.

Federated Learning (FL) can be used to distribute Machine Learning (ML) tasks across edge and Internet of Things (IoT) devices with limited resources.FL provides an alternative and much more practical solution to classical Artificial Intelligence (AI), which requires moving large data volumes to energy-hungry data centers.On the other hand, sustainability of FL processes should be accurately quantified as limiting energy consumption might require sacrificing accuracy. This paper proposes a framework for real-time monitoring of energy and Green House Gas (GHG) emissions (carbon footprints) of FL systems. The framework is developed for both classical FL policies relying on the parameter server and emerging fully decentralized ones.The proposed approach considers, for the first time, the impact of ML model quantization and sparsification on the energy/carbon budget while also discussing novel Gradient Tracking (GT) FL strategies that are robust to data heterogeneity but require higher communication bandwidth.General guidelines for energy-efficient designs are discussed based on several case studies on real datasets.The paper quantifies the energy footprint of continual FL processes that implement periodic adaptation on new data as foreseen by emerging IoT industry verticals.Results show that centralized FL is advantageous when strict carbon budgets are imposed or energy-inefficient (< 50 Kbit/Joule) communication protocols are adopted. GT mechanisms are to be preferred in heterogeneous data environments and decentralized setups.Finally, using FL for continual model fine-tuning provides large energy savings (> 80%), provided the ML model compression is properly tuned.

Central Nervous System (CNS) tumors represent a significant public health concern due to their high morbidity and mortality rates. Magnetic Resonance Imaging (MRI) has emerged as a critical non-invasive modality for the detection, diagnosis, and management of brain tumors, offering high-resolution visualization of anatomical structures. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), have shown potential in augmenting MRI-based diagnostic accuracy for brain tumor detection. In this study, we evaluate the diagnostic performance of six fundamental MRI sequences in detecting tumor-involved brain slices using four distinct CNN architectures enhanced with transfer learning techniques. Our dataset comprises 1646 MRI slices from the examinations of 62 patients, encompassing both tumor-bearing and normal findings. With our approach, we achieved a classification accuracy of 98.6%, underscoring the high potential of CNN-based models in this context. Additionally, we assessed the performance of each MRI sequence across the different CNN models, identifying optimal combinations of MRI modalities and neural networks to meet radiologists’ screening requirements effectively. This study offers critical insights into the integration of deep learning with MRI for brain tumor detection, with implications for improving diagnostic workflows in clinical settings.

CERN operates and maintains the world largest cryogenic infrastructure ranging from ageing installations feeding detectors, test facilities and general services, to the state-of-the-art cryogenic system serving the flagship LHC machine complex. After several years of exploitation of a wide range of cryogenic installations and in particular following the last two years major shutdown to maintain and consolidate the LHC machine, we have analysed and reviewed the maintenance activities to implement an efficient and reliable exploitation of the installations. We report the results, statistics and lessons learned on the maintenance activities performed and in particular the required consolidations and major overhauling, the organization, management and methodologies implemented.

Functional dependencies in complex technical infrastructures can cause unexpected cascades of failures. Given the complexity and continuous evolution in time of the infrastructure architecture, the identification of functional dependencies using the classical methods of system decomposition and logic analysis is not feasible. Then, we consider the availability of databases of alarm messages and frame the identification of functional dependencies in complex technical infrastructures as an optimization problem whose objective is the maximization of a metric measuring the level of dependence among alarms. A niching-based evolutionary algorithm has been developed to sequentially evolve a population of candidate solutions (group of alarms), maintaining diversity among them thanks to the use of a mechanism of population augmentation. The proposed algorithm is applied to a synthetic database of alarms generated by a complex technical infrastructure simulation model and to a real large-scale database of alarms collected from the particle accelerator of European Organization for Nuclear Research. The proposed algorithm is shown able to identify functional dependencies and to overperform other approaches based on the use of association rule mining algorithms, in terms of capability of extracting rare rules and computational efficiency.