International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 05 | May 2024
p-ISSN: 2395-0072
www.irjet.net
Comparative Analysis of Federated Learning Aggregation Techniques For Alzheimer’s Disease Diagnosis Akshada Tari1 1Student, Department of Information Technology and Engineering, Goa College of Engineering, Farmagudi, Goa,
India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Federated Learning (FL) is a machine learning
One of the key advantages of federated learning is its ability to operate without the necessity of transferring raw data to a central server. This feature is crucial in maintaining the confidentiality of sensitive information, making federated learning particularly attractive in scenarios where data privacy is of utmost importance. The decentralization of model training in federated learning not only enhances data privacy but also mitigates concerns related to computational power limitations. By distributing the training process across edge devices, federated learning leverages the collective computing power of a network of devices, enabling more efficient and scalable machine learning model training.
technique that decentralizes the training process across multiple devices or servers, each holding its own local dataset. This approach contrasts with traditional centralized machine learning techniques where all data is combined into one central point for training. In FL, an aggregator, typically a central server, plays a crucial role. It is responsible for collecting model updates from all participating nodes (clients), aggregating these updates, and then distributing the aggregated model back to the nodes. In our research, we conducted a comparative analysis of federated learning aggregation techniques for Alzheimer’s disease diagnosis. We used the Flower framework to train the Machine Learning (ML) model. The aggregators used in this research include FedAvg (Federated Averaging), FedYogi (Adaptive Federated Optimization using Yogi), FedOpt (Federated Optim strategy), FedMedian (Configurable FedMedian strategy), FedTrimmedAvg (Federated Averaging with Trimmed Mean). These aggregators were applied on the Alzheimer MRI Preprocessed Dataset. Among these, FedTrimmedAvg yielded the best accuracy result.
Security risks associated with centralized models are also alleviated through federated learning. The decentralized nature of this approach reduces the likelihood of a single point of failure compromising the entire system. This makes federated learning more resilient to attacks and ensures the robustness of the overall system. The transformative potential of federated learning is particularly evident in its application to healthcare, where issues of data privacy and security are paramount. In the context of Alzheimer's disease diagnosis, federated learning offers a promising solution. By allowing model training to occur on local devices, the sensitive health data involved in Alzheimer's diagnosis can be kept on individual devices, minimizing the risk of unauthorized access and ensuring patient privacy.
Key Words: Federated learning, aggregator, CNN, alzheimer’s disease
1. INTRODUCTION Federated learning, also known as collaborative learning, is a groundbreaking approach that has revolutionized the way machine learning models are trained. Unlike traditional centralized methods, federated learning operates on a decentralized paradigm, eliminating the need for exchanging raw data between client devices and global servers. This novel methodology enhances data privacy by conducting model training locally on edge devices, utilizing the raw data present on each individual device. The shift from traditional centralized server models to federated learning is driven by several critical issues associated with the former. Centralized approaches often face challenges such as data privacy concerns, computational power limitations, security risks, and vulnerability to single points of failure. Federated learning addresses these challenges directly, offering a more secure and privacy-preserving alternative.
© 2024, IRJET
|
Impact Factor value: 8.226
2. RELATED WORKS This paper [1] explores and investigates several federated learning aggregation strategies, algorithms, concept, advantages and disadvantages. It also explains the working of federated learning. This study [2] replicates experiments using four clients and 2482 chest X-ray images from the Kaggle repository. The dataset is divided into training and testing parts, with each client receiving 25% of the entire data. The accuracy rates on testing data are assessed after each federated learning round. The decentralized and distributed nature of the training process results in significant variation in accuracy compared to typical existing learning models. The model with federated learning compromises performance and is ahead regarding privacy. The loss function evaluates the
|
ISO 9001:2008 Certified Journal
|
Page 862