Hardware-Aware Probabilistic Machine Learning Models: Learning, Inference and Use Cases

This book proposes probabilistic machine learning models that represent the hardware properties of the device hosting them, with the goal of balancing resource consumption and performance of machine learning tasks in resource-constrained devices. It augments and exploits the properties of Bayesian Networks and Probabilistic Circuits to encode the properties of device sub-systems, bringing about resource-saving opportunities that traditional approaches fail to uncover. The performance of the proposed models and strategies is empirically evaluated for several use cases, showing significant resource-saving opportunities with minimal accuracy losses.

Format: Paperback / softback
Length: 163 pages
Publication date: 21 May 2022
Publisher: Springer Nature Switzerland AG

This book presents probabilistic machine learning models that capture the hardware characteristics of the devices hosting them. These models serve as valuable tools for assessing the impact of specific device configurations on the resource consumption and performance of machine learning tasks, with the ultimate objective of achieving optimal balance between the two.

In the introductory chapter, the book explores the realm of extreme-edge computing within the context of the Internet of Things (IoT) paradigm. It provides a brief overview of the steps involved in executing machine learning tasks and highlights the implications associated with deploying such workloads on resource-constrained devices.

The heart of this book revolves around augmenting and harnessing the properties of Bayesian Networks and Probabilistic Circuits to imbue them with hardware awareness. By doing so, the proposed models can encode the characteristics of various device sub-systems that are often overlooked by traditional resource-aware strategies. This results in the discovery of resource-saving opportunities that traditional approaches fail to uncover.

To evaluate the performance of the proposed models and strategies, extensive empirical evaluations are conducted for a diverse range of use cases. These examples demonstrate the potential of achieving significant resource savings with minimal accuracy losses at application time.

In conclusion, this book offers a groundbreaking approach to hardware-algorithm co-optimization, further bridging the domains of Machine Learning and Electrical Engineering. It provides valuable insights and practical methodologies for optimizing resource utilization in machine learning systems deployed on resource-constrained devices.

Weight: 279g
Dimension: 235 x 155 (mm)
ISBN-13: 9783030740443
Edition number: 1st ed. 2021