Wearable activity recognition improves with wavelet-based multiscale learning
W-JEPA: Wavelet Joint-Embedding Predictive Architecture for Wearable Activity Recognition
Machine Learning
Summary
Recognizing human activities with wearable devices is tough because sensors differ a lot and movements can look very similar. The authors created W-JEPA, a method that breaks down motion signals into smaller parts at different scales and learns patterns by predicting missing pieces. This helps the system notice subtle differences in movement across various sensors and people. Tests on several activity datasets show that W-JEPA works better than previous approaches, especially when recognizing activities for new people.
What this means in practice
- •For wearable device developers: Create more accurate activity recognition features that work well across different sensor setups and users by using multiscale wavelet token learning.$Commercial implications: It enables selling wearables with improved adaptable activity recognition that handles diverse hardware and user variability better than existing solutions.
- •For fitness app developers: Improve activity tracking reliability by applying the W-JEPA method to sensor data from various wearable devices and user populations.
Authors
Yogesh Prabhu, Loki Natarajan, Jingjing Zou
Abstract
A major challenge in wearable device activity recognition is the heterogeneity across datasets in sensor placement, channel count, sampling rate and subjects. Moreover, the motion patterns that distinguish activities often appear as subtle, scale-specific variations within otherwise similar waveforms. As a result, raw signals are often ambiguous, and models trained on them tend to rely on cues specific to a particular sensor setup or participant, which limits both cross-dataset transfer and generalization to unseen participants. Addressing this requires representations that preserve such activity-relevant, scale-specific variation. We introduce W-JEPA, a self-supervised framework that exploits the multiscale structure of inertial motion and leverages a joint-embedding predictive architecture, by organizing windows into a fixed grid of wavelet tokens indexed by scale, sensor axis, and time. These wavelet tokens help expose activity-specific detail within broadly similar raw signals, and serve as the units of masked latent prediction. W-JEPA predicts masked token representations supplied by an exponential-moving-average (EMA) target encoder. Because recovering a masked token requires relating motion at one scale and axis to motion at other scales and axes, the model learns the cross-scale, cross-axis structure of human movement. Across six wearable activity benchmarks, W-JEPA outperforms state-of-the-art methods under random-window splits and remains the strongest when subjects are held out, improving average macro-F1 by 8.7 percentage points over the strongest published baseline. These results indicate that physically grounded multiscale tokenization, combined together with latent predictive learning, provides a promising foundation for transferable representations of heterogeneous wearable inertial data.