UNDERSTANDING HUMANS
IN INTELLIGENT DEVICES AT SCALE

The age of autonomous machines and intelligent IoT devices is upon us. They will affect many aspects of our lives. They will bring about a new generation of diagnostic instruments to clinics, improve cars and transportation, and create new consumer experiences, while being embedded into the networks that power and inform society underpinning the efficient provision of public and private services. Increasingly they sense their surroundings, operate autonomously and collaborate with other devices seamlessly.

Critical to their operation, will be improving the safety, security, and quality of life of humans that interact with such intelligent systems. To interact efficiently and effectively, sensing and interpreting human behaviour will become mission critical to connecting the human user with automated systems. This white paper outlines the core horizontal topics necessary to teaching machines to sense and understand humans and use cases where “AI at the Extreme Edge” is not only critical but transformative.

HUMAN BEHAVIOUR AI
AI-DRIVEN HUMAN MACHINE INTERFACES

The task of understanding humans by using smart sensing has become increasingly important in modern society, given the proliferation and ubiquity of embedded sensors in a huge range of devices ranging from smartphones and smart city infrastructure to consumer IoT devices and health monitoring systems. A key to enabling the human user to connect with automated systems is artificial intelligence, which enables the plethora of sensing data to be processed in a way that gives context and meaningful insights to aid smart devices to make decisions and perform tasks.

Although there are a number of challenges in developing and integrating artificial intelligence into smart sensing and IoT applications (ranging from memory footprint and computational complexity, to privacy and robustness), the detection and understanding of human activities using artificial intelligence can be divided into three fundamental layers as described below.

DATA CENTRIC AI
AI FLYWHEEL EFFECT

An Artificial Intelligence Application (AI App) is the entire software including any machine learning component(s) of a data-driven algorithm necessary to transform data input(s) into output(s). An AI App is represented by its architecture that shows how to transform its input(s) into its output(s) where the transformation is given in the form of machine learning algorithm(s) (normally specified in frameworks such as Pytorch, Tensorflow, etc.) including pre and post processing algorithms and code.

Data Centric AI is the discipline of systematically engineering data used to build and continuously improve an AI system over time. The virtuous cycle of AI Apps, also called the "AI Flywheel Effect" is one of the most exciting ideas in Artificial Intelligence and it is also incredibly simple. Essentially, when AI technologies are integrated with a product properly, they create a feedback loop where the product continuously improves with use, generating more usage and a better competitive position relative to other products.

Although any product tends to improve with usage regardless of its underlying technology because a good team will use qualitative feedback and analytics data to bring it closer in line with user needs. This improvement, though, tends to reach an asymptote where additional usage and data no longer provide much marginal insight to the product.

When a product's core technology is AI-driven, though, it adds another layer on top of the typical team-driven product improvement cycle. With today's AI technologies (for example Deep Learning), additional data continues to provide marginal improvement for a very long time, allowing the cycle to continue much further than it previously could and improving the core product for a long time.

THE EXTREME EDGE
COMPUTING AT THE SENSOR

With the explosive growth in the adoption of artificial intelligence to address a large range of problems that were deemed very hard, data-intensive applications have placed a high demand on hardware performance, in terms of short access latency, high capacity, large bandwidth, low cost, and ability to execute artificial intelligence (AI) tasks. So far, the majority of attention and investment has been directed towards “Cloud AI”, with “Data” being the largest common denominator in creating value for industries, governments, and individuals’ lives. However, the quest for intelligence is fast becoming a prominent and essential feature at the “Edge” as well, where trillions of “things” will combine to generate even more data. Given the severe constraints that govern edge devices in terms of efficiency, footprint, robustness and cost, it is self-evident that bringing true intelligence to the edge will require profound innovation at all levels of the stack from the computational concepts all the way down the implementation technology.

SMART HEALTH
APPLICATIONS