The Intuition of Algorithms
We live in a world of ever smarter machines and devices. For all their intelligence, though, we still find ourselves tapping on screens or pressing buttons to communicate with most of our gadgets. When will devices spend more time thinking about us, than we do about them? Algorithmic Intuition is focused on ways to close the gap. We develop deep-learning-based algorithms that recognize activity and context. From these algorithms, we can derive actionable insight for devices to better understand human needs.
Aging Adults and Care
As our population ages, the need for high quality care becomes ever more urgent. Already there are 20 million people in the US over the age of 75 years. Gone is the assumption of residential care. Now Aging Adults want to live their lives in the comfort of their own home. Today with proper care and monitoring, aging-in-place is becoming the standard. The challenge then is understanding when care plans should change and when more immediate help is needed.
Activity is complicated. We move, shift, reach, stretch and walk. For fellow humans, we readily recognize each of these activities without a thought. Knowing a person is sitting on sofa, we can guess they will relax for a while. Knowing they are drinking a beverage, we can guess that they are thirsty. Human activity provides tremendous understanding about people.
Machine Learning Meets Sensors
Recent advances in deep neural network (DNN) algorithms are providing unprecedented capability to resolve activities from fused motion and sensor data. When taken from data across large populations, these DNNs are especially powerful and allow for subtle distinctions between activities (ADLs) such as Drinking, Eating, Dressing, Walking, Sitting etc with high accuracy and low-computational cost.
Sensing the World
There are a vast number of human-centric sensors being used today. Thanks to advances from the Mobile phone industry, low-cost, high-volume sensors can measure motion, rotation, particulates, light, pressure, humidity, sound and location — to name a few. With their small size, these allow a broad range of solutions for identification of human activity.
Human Activity Recognition
Human Activity Recognition is a set of algorithmic techniques that takes fused sensor data from motion and other sensors and identifies activity. The range of activities is broad and varied. From identifying activities in the elderly, to maximizing athletic performance, to optimizing factory efficiency, to tracking medical care and regimen compliance.