Laura Isabel Galindez Olascoaga - Hardware-Aware Probabilistic Machine Learning Models: Learning, Inference and Use Cases

This Springer imprint is published by the registered company Springer Nature Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Throughout the last decade, cloud-computing paradigms have proven successful at exploiting the opportunities and addressing the challenges of the ever-increasing number of electronic devices populating our environment. However, the privacy, latency, and efficiency concerns of this approach have motivated users and developers to seek computing paradigms that remain closer to the data-collecting devices, otherwise known as edge devices.

One of the main challenges to overcome in the quest towards edge-computing is the significant energy and computational bandwidth constraints of the involved devices, which tend to be battery operated and portable, especially when devices have to execute complex algorithms relying on large amounts of high-quality sensory data, such as in the case with machine learning applications. Furthermore, smart portable devices are prone to dynamically changing environmental conditions and noise, and the sensory data they rely on is inherently uncertain.

Probabilistic models constitute a suitable approach to address some of these challenges, since they can represent the inherent uncertainty of edge applications, and are robust to noise and missing data. However, their implementation in resource-constrained edge devices has not been studied extensively, unlike other machine learning paradigms, such as neural networks, which have already seen tremendous progress in this research field.

This book proposes to endow probabilistic models with hardware-awareness, in an attempt to enable their efficient implementation in resource-constrained edge devices. These models can represent scalable properties of the devices that host them, such as the quality with which their sensors operate, or the complexity of the inference algorithm they ought to perform.

The strategies proposed herewith can exploit these models to evaluate the impact that a specific device configuration may have on the machine learning tasks resource consumption and performance, with the overarching goal of balancing the two optimally. The proposed models can also consider various sub-systems and their properties holistically, bringing about resource-saving opportunities that other resource-aware approaches fail to uncover. Specifically, this book focuses on augmenting Bayesian networks and probabilistic circuits with hardware-awareness. These models performance is empirically evaluated in several use cases that consider different types of systems and various scalable properties. The range of use cases proves to benefit from this thesiss contributions, with the potential of attaining significant resource-saving opportunities with minimal accuracy losses at application time.

We would like to acknowledge several people and institutions for their valuable contributions to this work.

We thank Nilesh Ahuja, Herman Bruyninckx, Jesse Davis, Guy Van den Broeck, and Patrick Wambacq for their feedback on previous versions of this manuscript and for all the engaging and inspiring discussions throughout the last years. We would also like to acknowledge our collaborators contributions in the articles discussed throughout this book: Martin Andraud, Komail Badami, Steven Lauwereins, and Nimish Shah.

Finally, we thank Intel for their financial support, as well as the European Research Council for their funding under the Re-SENSE project grant number ERC-2016-STG-71503.