Apple’s new Heart Rate Dynamics paper provides a clue around interesting innovations to come on Apple’s Wearables

A new research paper from Apple researchers, accepted at the NeurIPS 2022 conference, shows the potential of machine learning algorithms to provide accurate heart rate (HR) dynamics in response to workout intensity and duration.

The findings of the new study were also published in Apple’s machine learning research library.

Apple’s researchers, along with key opinion leaders from academia, showcase a hybrid machine learning model that combines a physiological HR model during exercise with complex neural networks to learn user-specific fitness representations.

Related reading:

• Researchers find Apple Watch Sensor to be accurate when used for predicting Stress Levels
• Use the new Race Route Workouts on your Apple Watch to compete against yourself
• New Mayo Clinic study shows Apple Watch can help detect a potentially life-threatening heart condition

Details on Apple’s heart rate dynamics research

The researchers applied this model at scale to a large set of workout data collected with Apple Watch and showed that it could accurately predict HR response to exercise demand in new workouts.

The paper introduces a personalized physiological model of heart rate (HR) based on ordinary differential equations (ODEs) that uses neural networks and representation learning to estimate user-specific parameters.

The research shows that these algorithms can predict HR for workout (w) happening at  date (T), just based on prior workout data of the user.

Here are some of the significant inferences from this new study:

1. Apple’s machine learning algorithms can estimate your heart rate before you work out.
2. Unlike other existing work out there that can perform short-term HR prediction, Apple’s method can predict the full HR trend of a workout of up to 2 hours, even for those workouts the user has yet to experience!
3. The implications of this are huge. This forecasting model can be used in new personalized workout planning and estimating HR zones or calories burned during a workout.
4. The research also shows that learned representations of HR dynamics correlate well with traditional metrics of cardiorespiratory fitness.
5. Estimating the calories burned during exercise can be accurately measured using a linear formula with heart rate measurements collected during prior workouts. This can be used to design activities accurately based on calorie burn goals. Even when the user is not wearing an Apple Watch, the model can predict calories burned accurately using workout metrics collected from the user’s iPhone.
6. Apple’s model can consider local geographical data to fine-tune its predictions. The researchers quantified the relative effect of weather on the body’s oxygen demand (this constitutes one of the most extensive studies of this kind (with over 270,000 workouts)). In short, if your training is going to be at a higher temperature or a humid place, the algorithm can take that into account and project out your HR and other workout details.

The new study used workout measurements contributed to the Apple Heart and Movement Study by users. The researchers analyzed 270,707 outdoor runs from 7,465 users enrolled between 2019 and 2022.

The heart and movement study provided four measures of the exercise intensity: speed from the pedometer and the GPS, step cadence, and elevation gain apart from the heart rate.

The Apple researchers believe that can help track fitness levels over time and aid personalized workout planning. According to the researchers, future work investigates how such measures can predict changes in cardiovascular health.