Mobile health (mHealth) apps offer a range of benefits. There are nutrition-tracking apps that help us count calories, sleep apps are meant to track our sleep patterns, and stress-management apps help us calm our mind. These apps make suggestions for us to follow. However, in these apps, reinforcement learning algorithms that adapt to one’s context without learning personalized policies might fail to distinguish between the needs of individuals. Therefore, the effectiveness of any type of health app requires delivering recommendations at useful times, and at the same time avoid overtreatment.
Researchers at three major U.S. universities – Harvard, Stanford and the University of Michigan – proposed an artificial intelligence (AI) framework they claim could deliver recommendations via smartphone that encourage healthier lifestyles. They named their new system IntelligentPooling, reports Kyle Wiggers on VentureBeat.
“We present IntelligentPooling, which learns personalized policies via an adaptive, principled use of other users’ data,” the researchers wrote.
To bolster training, the recommendations can be personalized and data on other users made available. Researchers claim this approach reduces regret – i.e., the number of actions taken when there was a better choice in hindsight — by 26%.
“We show that IntelligentPooling achieves an average of 26% lower regret than state-of-the-art across all generative models. Additionally, we inspect the behavior of this approach in a live clinical trial, demonstrating its ability to learn from even a small group of users,” the researchers wrote.
The goal of IntelligentPooling is to learn an optimal policy for when and how to intrude for every individual and context. Over time, the system develops personalized treatment policies for each user, algorithmically learning from data pooled from the users’ devices.
The researchers conducted a study on 10 people with Stage 1 hypertension. Each individual was equipped with a Fitbit Versa smartwatch. Results showed that IntelligentPooling was able to determine whether to make an activity suggestion based on 107 data points from all users; autonomously, it could decide to send (or not send) a notification with a suggestion, the VentureBeat report said.