Название: Model Optimization Methods for Efficient and Edge AI: Federated Learning Architectures, Frameworks and ApplicationsАвтор: Pethuru Raj Chelliah, Amir Masoud Rahmani, Robert Colby, Gayathri Nagasubramanian, Sunku Ranganath
Издательство: Wiley-IEEE Press
Год: 2025
Страниц: 414
Язык: английский
Формат: pdf (true), epub
Размер: 29.4 MB
Comprehensive overview of the fledgling domain of Federated Learning (FL), explaining emerging FL methods, architectural approaches, enabling frameworks, and applications.
Model Optimization Methods for Efficient and Edge AI explores AI model engineering, evaluation, refinement, optimization, and deployment across multiple cloud environments (public, private, edge, and hybrid). It presents key applications of the AI paradigm, including computer vision (CV) and Natural Language Processing (NLP), explaining the nitty-gritty of Federated Learning (FL) and how the FL method is helping to fulfill AI model optimization needs. The book also describes tools that vendors have created, including FL frameworks and platforms such as PySyft, Tensor Flow Federated (TFF), FATE (Federated AI Technology Enabler), Tensor/IO, and more.
The first part of the text covers popular AI and ML methods, platforms, and applications, describing leading AI frameworks and libraries in order to clearly articulate how these tools can help with visualizing and implementing highly flexible AI models quickly. The second part focuses on federated learning, discussing its basic concepts, applications, platforms, and its potential in edge systems (such as IoT).
TensorFlow Federated (TFF) is a cutting-edge open-source framework spearheaded by Google, constructed as a specialized extension to the TensorFlow platform [98]. This enhancement was done specifically to cater to the intricacies and demands of federated environments. At its core, TFF is powered by TensorFlow, a highly scalable ML framework. When applied to the Intelligent Internet of Things (ITIoT) sector, TFF’s scalability ensures it can effortlessly handle vast networks of interconnected devices, making it an invaluable asset in large-scale intelligent operations. One of the significant breakthroughs of TFF is its compatibility with TensorFlow Lite. TensorFlow Lite is a lightweight solution, designed explicitly for on-device ML. This means that TFF, in conjunction with TensorFlow Lite, can facilitate training and inference right on the edge devices. For ITIoT, this feature is indispensable. Given the remote and often decentralized nature of many intelligent devices, being able to process and infer data on the spot can make operations smoother and more efficient.
PySyft is an advanced library that augments the PyTorch framework to facilitate encrypted, privacy-preserving ML. It was developed by the OpenMined community, which is dedicated to democratizing access to private data for ML purposes without violating the data’s privacy. PySyft represents a paradigm shift in how data access is approached, emphasizing data privacy and security while still harnessing the value locked in the data. Encrypted Computation: PySyft enables data operations and computations to be carried out on encrypted data. This means models can be trained and inferences made without ever having to access the raw, unencrypted data. This approach, often termed “encrypted deep learning,” is critical in ITIoT where data security and privacy are paramount.
Other topics covered include:
Building AI models that are destined to solve several problems, with a focus on widely articulated classification, regression, association, clustering, and other prediction problems
Generating actionable insights through a variety of AI algorithms, platforms, parallel processing, and other enablers
Compressing AI models so that computational, memory, storage, and network requirements can be substantially reduced
Addressing crucial issues such as data confidentiality, data access rights, data protection, and access to heterogeneous data
Overcoming cyberattacks on mission-critical software systems by leveraging Federated Learning
Скачать Model Optimization Methods for Efficient and Edge AI: Federated Learning Architectures, Frameworks and Applications