Parallel & distributed data analytics for time-sensitive applications on fog/edge architectures

Abstract

Fog computing expands upon the conventional cloud computing model, typically integrating fog nodes at the network edge for computing and storage purposes. By situating these edge devices close to users, it enhances application response times and alleviates the burden on the central cloud server. Additionally, fog computing o↵ers computational, storage, and networking services bridging the gap between users and traditional cloud computing data centers. This thesis proposes four di↵erent frameworks for ML model partitioning on fog architecture, Intrusion Detection System on Fog Architecture, Data-Driven Deep Neural Network Task O✏oading on Edge Networks, real-time outcomes prediction in cardiac surgery. In the first work, we propose a framework that intelligently partitions ML models into smaller sub-models that can be safely executed across multiple edge devices, leveraging their parallel computing capabilities. Further, to enhance the safety and reliability of the online model training process, our approach incorporates the Triple Modular Redundancy (TMR) technique for trusted computation. The second work proposes a lightweight distributed Intrusion Detection System (IDS) framework, called FCAFE-BNET (Fog based Context Aware Feature Extraction using BranchyNET). The proposed FCAFE-BNET approach considers versatile network conditions, such as varying bandwidth and data load before allocating inference tasks on Cloud/Edge resources. Early exit DNN is used to obtain faster inference generation at the edge. The proposed FCAFE-BNET framework works for both Network-based and Host-based IDS. In the third work, we propose a D2–TONE (Data-driven Deep Neural Network Task O✏oading on the Network Edge), an approach that employs Machine Learning algorithms for accurately estimating o✏oading delays, such as computational and transmission delays. D2–TONE holistically adapts to dynamic network situations and provides optimal/near-optimal o✏oading solutions in real-time. In addition, the proposed algorithm employs distributed execution of DNN tasks on edge devices/cloud data centers. The fourth work aims to develop artificial intelligence models based on non-linear time-series data of blood pressure and heart rate to predict the ICU stay, hospital stay, and survival outcome of cardiac surgical patients. Specifically, we aim to construct an end-to-end real-time data analysis pipeline that incorporates artifact removal, non-linear noise reduction, and features engineering. We have performed model predictions on an edge device, so that the alerts to the doctor can be transmitted in real-time. This thesis also provides a detailed description of the fog computing paradigm. It summarises the state-of-the-art work in the field of ML model partitioning on fog architecture, Intrusion Detection System on Fog Architecture, Data-Driven Deep Neural Network Task O✏oading on Edge Networks, and real-time outcomes prediction in cardiac surgery. Finally, this thesis discusses future research directions in this field.