About Workshop 

This workshop is focused on enhancing and nurturing the research in the niche domain of AutoML for tiny edge systems like wearables / implantables and smart fabrics. In recent years, the applications of AI have become nearly indispensable in healthcare and other pervasive sensing domains. This is facilitated even more by the higher usage of wearables and other forms of personal devices. Creation of a seamless experience with multiple applications running on such battery-powered devices with limited on-device compute power requires efficient optimization of AI-ML models. When designing such systems across multiple applications and devices, manual tuning and pruning becomes extremely cumbersome and time-consuming, even for domain experts with years of experience. It also makes it difficult to reproduce the same optimizations for different sets of target constraints, which are very significant in embedded systems. Automation becomes very useful in such scenarios, where models need to be customized for different hardware and applications.

Existing approaches for AI in Pervasive Sensing Systems include manual design and optimization of machine learning and deep neural network models. These approaches are often tedious and sub-optimal, with more layers and operations used in applications where very small networks will suffice. Exploring this domain will also contribute to energy conservation by enabling the use of smaller models that can run continuously on personal monitoring devices. Automated optimization techniques hold the potential to solve this problem. Along with reduction of model footprint, it is also necessary to reduce the number of GPU hours consumed during search of such models.

We welcome papers with novel ideas on automation of machine learning for edge devices for pervasive sensing applications. This covers Hyperparameter Optimization Techniques, Neural Architecture Search Techniques, Wearable and Edge implementation of AutoML-generated models, real-time parameters and their optimization and impact on on-device inference. The IEEE LaTeX and Microsoft Word templates, as well as formatting instructions, can be found here. All papers must be at most 6 pages of technical content, typeset in double-column IEEE format using 10pt fonts on US letter paper, with all fonts embedded. The accepted papers will be published in IEEE Xplore.

The primary goals of this workshop are:

• To discuss the different opportunities of automation in machine learning, and how they can lead to a robust and scalable methodology to design pervasive sensing and AI systems with minimized effort.

• To create a forum for researchers in academia and industry to discuss and collaborate on the best solutions for AutoML in pervasive computing systems.

• To exchange ideas and opinions about the advantages and disadvantages of Automation in the design of AI systems and techniques to overcome or minimize them.