Conference Papers
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/123456789/28506
Browse
7 results
Search Results
Item FedPruNet: Federated Learning Using Pruning Neural Network(Institute of Electrical and Electronics Engineers Inc., 2022) Gowtham, L.; Annappa, A.; Sachin, D.N.Federated Learning (FL) is a distributed form of training the machine learning and deep learning models on the data spread over heterogeneous edge devices. The global model at the server learns by aggregating local models sent by the edge devices, maintaining data privacy, and lowering communication costs by just communicating model updates. The edge devices on which the model gets trained usually will have limitations towards power resource, storage, computations to train the model. This paper address the computation overhead issue on the edge devices by presenting a new method named FedPruNet, which trains the model in edge devices using the neural network model pruning method. The proposed method successfully reduced the computation overhead on edge devices by pruning the model. Experimental results show that for the fixed number of communication rounds, the model parameters are pruned up to 41.35% and 65% on MNIST and CIFAR-10 datasets, respectively, without compromising the accuracy compared to training FL edge devices without pruning. © 2022 IEEE.Item Federated Learning for Wearable Sensor-Based Human Activity Recognition(Springer Science and Business Media Deutschland GmbH, 2023) Sachin, D.N.; Annappa, B.; Ambesenge, S.Computing devices that can identify various human behaviors or motions may be used to aid individuals in multiple contexts, including sports, healthcare, and interactions between humans and robots. Data readily accessible for this purpose may be collected from smartphones and wearable devices used daily. Therefore, efforts are made to classify real-time activity data effectively utilizing various machine learning models. However, current methods for human activity recognition do not sufficiently consider user data privacy. To mitigate privacy issues, federated learning can be employed to build generic activity classification model by aggregating a locally trained model at a user-edge device. This paper adopted a deep-learning neural network model called the transformer for motion signal time-series analysis. It uses the attention mechanism to provide context for each point in the time series. It also compares federated learning’s performance to centralized learning. Experimental results show that federated learning outperformed centralized training without com- promising user data privacy, with an accuracy of 96.87%. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item FedRH: Federated Learning Based Remote Healthcare(Institute of Electrical and Electronics Engineers Inc., 2023) Sachin, D.N.; Annappa, B.; Ambesenge, S.More and more people are developing chronic illnesses, which require regular visits to the hospital for treatment and monitoring. Due to the advancements in wearable computing technology, there is a growing interest in remote health monitoring worldwide. The Internet of Medical Things (IoMT) devices allow users to access their health data. Recent developments in smart healthcare have shown that machine learning model training using vast amounts of healthcare data has been remarkably successful. However, remote healthcare is facing a significant challenge as user data is stored in isolated silos, thus making data aggregation impractical while ensuring data privacy and security. To overcome this challenge, this paper presents FedRH, a federated learning framework for remote healthcare. FedRH solves the issue of data isolation through a cloud-edge-based computing architecture and FL, providing personalized healthcare without sacrificing privacy and security. Experiments show that FedRH achieves a greater accuracy of 5.6% compared to conventional approaches. FedRH is a flexible and versatile tool that can be applied to various healthcare applications, making it an ideal choice for many use cases. © 2023 IEEE.Item FedLSF: Federated Local Graph Learning via Specformers(Institute of Electrical and Electronics Engineers Inc., 2024) Ram Samarth, B.B.; Annappa, B.; Sachin, D.N.The abundance of graphical data and associated privacy concerns in real-world scenarios highlight the need for a secure and distributed methodology utilizing Federated Learning for Graph Neural Networks(GNNs). While spatial GNNs have been explored in FL, spectral GNNs, which capture rich spectral information, remain relatively unexplored. Despite enhancing GNNs' expressiveness through attention-based mechanisms, challenges persist in the spatial approach for FL due to cross-client edges. This work introduces two information capture methods for spectral GNNs in FL settings, Global Information Capture and Local Information Capture, which address cross-client edges. Federated Local Specformer (FedLSF) is proposed as a novel methodology that combines local information capture with state-of-the-art(SOTA) Specformer, enabling local graph learning on clients. FedLSF leverages Specformers involving spectral and attention approaches by integrating Eigen Encoding, Transformer architecture, and graph convolution. This enables capturing rich information from eigen spectra and addresses concerns related to cross-client edges through fully connected eigen-spaces. Experimental results demonstrate FedLSF's efficacy in both homophily and heterophily datasets, showing significant accuracy improvements (2-50)% in highly non-independent and identically distributed (Non-IID) scenarios compared to the present SOTA. This research advances attention-based spectral mechanisms in FL for GNNs, providing a promising solution for preserving privacy in non-IID graph data environments. Implementation can be found at https://github.com/achiverram28/FedLSF-DCOSS. © 2024 IEEE.Item Abdominal Multi-Organ Segmentation Using Federated Learning(Institute of Electrical and Electronics Engineers Inc., 2024) Yadav, G.; Annappa, B.; Sachin, D.N.Multi-organ segmentation refers to precisely de-lineating and identifying multiple organs or structures within medical images, such as Computed Tomography (CT) scans or Magnetic Resonance Imaging (MRI), to outline boundaries and regions for each organ accurately. Medical imaging is crucial to comprehending and diagnosing a wide range of illnesses for which accurate multi-organ image segmentation is often required for successful analysis. Due to the delicate nature of medical data, traditional methods for multi-organ segmentation include centralizing data, which presents serious privacy problems. This centralized training strategy impedes innovation and collaborative efforts in healthcare by raising worries about patient confidentiality, data security, and reg-ulatory compliance. The development of deep learning-based image segmentation algorithms has been hindered by the lack of fully annotated datasets, and this issue is exacerbated in multi-organ segmentation. Federated Learning (FL) addresses privacy concerns in multi-organ segmentation by enabling model training across decentralized institutions without sharing raw data. Our proposed FL-based model for CT scans ensures data privacy while achieving accurate multi-organ segmentation. By leveraging FL techniques, this paper collaboratively trains segmentation models on local datasets held by distinct medical institutions. The expected outcomes encompass achieving high Dice Similarity Coefficient (DSC) metrics and validating the efficacy of the proposed FL approach in attaining precise and accurate segmentation across diverse medical imaging datasets. © 2024 IEEE.Item Enhancing Healthcare AI with Cross-Silo Personalized Federated Learning on Naturally Split Heterogeneous Data(Institute of Electrical and Electronics Engineers Inc., 2024) Mukeshbhai, A.N.; Annappa, B.; Sachin, D.N.The potential of Artificial Intelligence (AI) in health-care is unavoidable. However, its success depends on the availabil-ity of large, high-quality datasets. Because of data heterogeneity across institutions and privacy concerns, traditional centralized Machine Learning (ML) approaches often face difficulties in this field. Federated Learning (FL) allows collaborative model training without requiring the transfer of sensitive patient data from the original institution. Recent research in FL within the healthcare domain has predominantly relied on centralized datasets, which do not represent real-time data heterogeneity and made assumptions by random data splitting to different medical client institutions. Additionally, it may be challenging for a single global model to encompass the diverse characteristics of various healthcare settings accurately. This paper examines the application of Personalized Federated Learning (PFL) in realistic cross-silo healthcare scenarios with federated natural split datasets in different medical client institutions. This paper discusses the experiments conducted on brain segmentation, survival prediction, melanoma classification, and heart disease di-agnosis. Our experiments show that the proposed PFL techniques consistently improve local model performance over standard FL strategies by up to 10% in different medical use cases. © 2024 IEEE.Item Federated Split Learning with HyperNetworks for Medical Image Classification(Institute of Electrical and Electronics Engineers Inc., 2024) Sindhura, S.; Annappa, B.; Sachin, D.N.The integration of Artificial Intelligence (AI) in the medical field has significantly enhanced patient healthcare by enabling the development of sophisticated predictive models. However, the conventional method of building these AI models often requires the aggregation of vast amounts of centralized data, which poses substantial risks to patient data privacy and presents major barriers to scaling these solutions. Federated Learning (FL), a distributed machine learning paradigm, addresses this issue by allowing multiple clients to collaboratively train a model without exchanging their data, thus preserving privacy. Despite its benefits, FL faces significant data and computational heterogeneity challenges. Medical environments frequently feature devices with varied computational capacities, which can severely impede the performance of FL. This paper presents a novel model partitioning approach within the Split Learning (SL) framework by incorporating Dynamic Hypernetworks, which dynamically adjust network parameters in real-time. This method optimizes the’cut layer’—the point at which data is split between client and server—thereby enhancing computational efficiency and significantly reducing the risk of sensitive information leakage. It enables efficient model training on resource-constrained devices and demonstrates superior performance in a medical classification task when compared to traditional centralized, FL, and standard SL methods. The results show improved effectiveness in maintaining data privacy and highlight the potential of this approach in facilitating the broader adoption of AI in healthcare. ©2024 IEEE.