Search

Human Activity Recognition (HAR) has seen remarkable advances in recent years, driven by the widespread use of wearable devices and the increasing demand for personalized healthcare and activity tracking. Federated Learning (FL) is a promising paradigm for HAR that enables the collaborative training of machine learning models on decentralized devices while preserving data privacy. It improves not only data privacy but also training efficiency as it utilizes the computing power and data of potentially millions of smart devices for parallel training. In addition, it helps end-user devices avoid sending users’ private data to the cloud, eliminates the need for a network connection, and saves the latency of back-and-forth communication. FL also offers significant advantages for communication by reducing the amount of data transmitted over the network, alleviating network congestion and reducing communication costs. By distributing the training process across devices, FL minimizes the need for centralized data storage and processing, leading to more scalable and resilient systems. This paper provides a comprehensive survey of the integration of FL into HAR applications. Unlike existing reviews, this paper uniquely focuses on the intersection of FL and HAR, providing an in-depth analysis of recent advances and their practical implications. We explore key advances in FL-based HAR methodologies, including model architectures, optimization techniques, and different applications. Furthermore, we highlight the major challenges and future research questions in this domain, such as model personalization and robustness, privacy concerns, concept drift, and the limited capacity of edge devices.

An innovative spectroscopic method that allows to chemically and structurally characterize viruses directly in suspension within few minutes was developed. A library of five different plant viruses was obtained combining dielectrophoresis (DEP), performed with a device specifically designed to capture and agglomerate virus particles, and Raman spectroscopy to provide a chemical fingerprint of virions. The tested viruses, purified from infected plants, were chosen for their economic impact on horticultural crops and for their different morphological and structural features. Using the Raman-DEP device, specific profiles for each virus were successfully obtained, relying on chemical differences occurring even with genetically similar viruses belonging to the same taxonomic species and morphologically indiscernible by transmission electron microscopy (TEM). Moreover, we investigated the potentiality of Raman-DEP to follow dynamic changes occurring upon heat treatment of tobacco mosaic virus (TMV) particles. Raman peak deviations linked to TMV coat protein conformation were observed upon treatment at temperatures equal or higher than 85°C, substantiating the rod-to-spherical shape transitions observed by TEM and the concomitant drastic loss of infectivity following plant inoculation. Overall, the Raman-DEP method can be useful for the characterization of virus (nano)particles, setting the basis to create a database suitable for the study of viruses or virus derived-nanoparticles relevant for the agricultural, medical, or biotechnological fields.

The unceasing development of Artificial Intelligence (AI) and Machine Learning (ML) techniques is growing with privacy problems related to the training data. A relatively recent approach to partially cope with such concerns is Federated Learning (FL), a technique in which only the parameters of the trained neural network models are transferred rather than data. Despite the benefits that FL may provide, such an approach can lead to synchronization issues (especially when applied in the context of numerous IoT devices), the network and the server may turn into bottlenecks, and the load may become unsustainable for some nodes. To solve this issue and reduce the traffic on the network, in this paper, we propose P4FL, a novel FL architecture that uses the paradigm of network programmability to program P4 switches to compute intermediate aggregations. In particular, we defined a custom in-band protocol based on MPLS to carry the model parameters and adapted the P4 switch behavior to aggregate model gradients. We then evaluated P4FL in Mininet and verified that using network nodes for in-network model caching and gradient aggregating has two advantages: first, it alleviates the bottleneck effect of the central FL server; second, it further accelerates the entire training progress.

In this work, biomolecules, such as membrane proteins, lipids, and DNA, were identified and their spatial distribution was mapped within a single Escherichia coli cell by Raman hyperspectral imaging. Raman spectroscopy allows direct, nondestructive, rapid, and cost-effective analysis of biological samples, minimizing the sample preparation and without the need of chemical label or immunological staining. Firstly, a comparison between an air-dried and a freeze-dried cell was made, and the principal vibrational modes associated to the membrane and nucleic acids were identified by the bacterium’s Raman chemical fingerprint. Then, analyzing the Raman hyperspectral images by multivariate statistical analysis, the bacterium biological status was investigated at a subcellular level. Principal components analysis (PCA) was applied for dimensionality reduction of the spectral data, then spectral unmixing was performed by multivariate curve resolution–alternating least squares (MCR-ALS). Thanks to multivariate data analysis, the DNA segregation and the Z-ring formation of a replicating bacterial cell were detected at a sub-micrometer level, opening the way to real-time molecular analysis that could be easily applied on in vivo or ex vivo biological samples, avoiding long preparation and analysis process.