Skill Performance Estimation from cARdiac Signals

Introduction

In any learning situation, be it math education, language learning, or sport training, different learners have different abilities, motivations, and capacities at any given time. Thus, an optimal learning experience can only be achieved with personalized training solutions, dynamically adapted to each learner’s cognitive and/or physical states.

Background

The scientific literature shows that such states could be estimated from Cardiac Signals (CS). In ERC PoC SPEARS, we propose to redefine consumer training apps by enabling them to suggest personalized and adaptive training plans according to an estimation of their users’ cognitive and/or physical states from their CS measured with consumer-grade sensors, e.g., smartwatches.

Previous Work

The outcome of the ERC project BrainConquest should enable us to tackle this challenge. In BrainConquest, we explored a personalized training approach for users of Brain-Computer Interfaces (BCI). In doing so, we developed Machine Learning (ML) and Signal Processing (SP) algorithms to estimate users’ mental states and predict their upcoming performances from their brain and physiological signals, including CS.

Objectives

In SPEARS, we aim to:

1. Adapt and improve BrainConquest ML & SP algorithms, initially designed for BCI performance prediction from research-grade brain and CS sensors in the lab.
2. Predict cognitive and physical performance from consumer-grade CS sensors in the wild.
3. Explore a commercial application of this technology for sport training, particularly in collaboration with the startup Flit Sport.

Collaboration with Flit Sport

Flit Sport sells an app for providing personalized training exercises for endurance sport athletes, based on their past performances and ML. By integrating our CS-based prediction into the Flit Sport training app, we aim to design optimally personalized training solutions for millions of runners worldwide.