Unassigned · 2026-10-15
W-JEPA: Wavelet Joint-Embedding Predictive Architecture for Wearable Activity Recognition
Yogesh Prabhu, Loki Natarajan, Jingjing Zou
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
W-JEPA is a self-supervised wearable activity recognition method that tokenizes inertial motion into fixed wavelet tokens indexed by scale, sensor axis, and time, then learns by masked latent prediction using a JEPA-style EMA target encoder. The authors report higher performance than prior methods on six wearable benchmarks, with the strongest subject-held-out results improving average macro-F1 by 8.7 points.
Why it matters
Wearable activity recognition faces cross-dataset heterogeneity (sensor placement, channel count, sampling rate, subjects). The paper argues that activity-discriminative motion can be scale-specific, and proposes wavelet-token units of masked latent prediction to preserve this structure for transferable representations under participant holdout.
Method
- Fixed wavelet tokenization: apply a discrete wavelet transform per sensor axis, summarize each wavelet-band bin with mean and RMS, and index tokens by (channel, band/scale, time bin).
- Masked latent pretraining: draw asynchronous paired masks across channel-band token pairs; a context encoder predicts latent representations of masked tokens from an EMA target encoder.
- Cross-dataset evaluation: pretrain on the remaining five datasets per target, fine-tune on 2–6 participants with participant-disjoint and random-window splits; use a classification head on encoder outputs.
Abstract (from arXiv)
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.
Related papers
- Learning Multimodal Sensor Fusion with JEPA Representations for Breathing Activity and Pose Recognition
- Phenomenon-Graph JEPA: Label-Efficient Representation Learning for Contactless Cardiorespiratory Sensing
- LEF-JEPA: Denoising-Trained Latent Energy Fields for Spatial Representation Learning
- Self-Supervised Latent Action-Conditioned World Modeling for Time Series
- Touch-JEPA: Tactile Representation Learning from Hand-Object Video with Limited Tactile Supervision
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-14 · Indexed by Pub-AI: 2026-10-15 UTC
W-JEPA: Wavelet Joint-Embedding Predictive Architecture for Wearable Activity Recognition
Preprint posted: 2026-10-14 · Original record ↗
- arxiv
- 2610.19411
Explicit source versions (1)
- v1 · 2026-10-14
- Preprint available
Source date: 2026-10-14 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
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.
