Article Preview
Top1. Introduction
Critical infrastructures are systems which are essential for the smooth functioning of a society and its economy and these include such as transportation systems, communication networks, and financial systems. These infrastructures are often complex, interdependent, and vulnerable to disruptions, which can have serious repercussions for public safety, economic stability, and national security. Deep Learning (DL) (Aggarwal et al., 2022; Mengi et al., 2023) provides critical infrastructure operators with valuable insights and predictive capabilities that can help them make more informed decisions, improve system resilience, and enhance public safety. By analyzing data from sensors and other sources, DL models can identify patterns and anomalies that may indicate equipment failure or maintenance needs so that it could be repaired before failures occur, reducing downtime (Mandle et al., 2022). Deep learning in critical infrastructure requires large amounts of training data for accurate and reliable modeling of the task. Traditionally, training data for these systems has been collected in data centers or on a single machine, which can be costly and time-consuming. Storing large amounts of critical infrastructure data in the cloud can pose significant risks and responsibilities, such as the potential for data breaches and cyber attacks. To address these challenges, training data from these systems must be collected and managed, in a decentralized manner. These approaches could help ensure the accuracy and quality of the training data while also reducing the risks and responsibilities associated with storing large amounts of sensitive data in the cloud. Federated Learning (FL) (D. Li et al., 2023) can be used to secure critical infrastructure making it possible to achieve the benefits of improved efficiency and performance while utilizing DL without compromising the safety and security of the system. Enhancing privacy(M. Singh et al., 2023) in critical infrastructure systems is crucial for safeguarding sensitive information and ensuring the reliability of essential services. Developing different prediction algorithms for critical infrastructure system (I. Singh, Singh, Singh, et al., 2022) (Peñalvo, Maan, et al., 2022) (S. Gupta et al., 2023) with sustainable development (Chopra et al., 2022; Peñalvo, Sharma, et al., 2022) (Bouncken et al., 2022; M. Singh et al., 2023) is an art which involves a deep understanding of the underlying systems, the potential risks they face, and a creative approach to designing algorithms with minimum overheads (S. Kumar et al., 2021) (S. Kumar et al., 2022) (P. S. Kumar, 2022; S. Kumar et al., 2023).
Federated deep learning is a technique and architecture which allows multiple clients to interact and train a deep learning model without having to share their raw data with each other. In federated learning, the training data is distributed across multiple clients or devices and the model is trained locally on each client using its own data. The updates from the local models are then aggregated to create a global model that is more accurate and robust. In this approach, since the data remains on the local devices or servers, and only the model updates are exchanged between the clients or with a central server. Therefore, federated deep learning helps to preserve the privacy and security (A. Sharma et al., 2023; I. Singh, Singh, Kumar, et al., 2022) of sensitive data and reduces the risk of data breaches and cyber attacks. In this paper a novel framework using federated deep learning for critical infrastructure has been proposed in which multiple devices or clients collaboratively train a global model without sharing their raw data. The global model is a convolutional neural network (CNN) (Kaur et al., 2021) that is trained initially on a centralized dataset. The framework is demonstrated on the MNIST dataset, a commonly used benchmark dataset for image classification.