RelCon: Relative Contrastive Learning for a Motion Foundation Model for Wearable Data

2024-11-27

arXiv.org

AI Summary

To address the weak generalization of wearable motion data models on downstream tasks with limited labeled samples, this paper introduces the first self-supervised foundation model specifically designed for wearable motion data. Methodologically, it employs relative contrastive learning to model semantic similarity among accelerometer time series, proposes a novel learnable distance metric incorporating physical priors—including motion primitive similarity and rotational invariance—and adopts a softened contrastive loss. The model is pre-trained on an ultra-large-scale temporal dataset comprising 1 billion motion segments from 87,000 users. Experiments demonstrate substantial improvements in few-shot transfer performance across diverse classification and regression downstream tasks. This work provides the first systematic validation of cross-task generalization capability of self-supervised motion representations, establishing a universal, robust foundation for wearable health analytics.

Technology Category

• Computer Vision: Motion & Tracking
• Machine Learning: Unsupervised & Self-Supervised Learning
• Intelligent Robots: Motion and Path Planning

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present RelCon, a novel self-supervised *Rel*ative *Con*trastive learning approach that uses a learnable distance measure in combination with a softened contrastive loss for training an motion foundation model from wearable sensors. The learnable distance measure captures motif similarity and domain-specific semantic information such as rotation invariance. The learned distance provides a measurement of semantic similarity between a pair of accelerometer time-series segments, which is used to measure the distance between an anchor and various other sampled candidate segments. The self-supervised model is trained on 1 billion segments from 87,376 participants from a large wearables dataset. The model achieves strong performance across multiple downstream tasks, encompassing both classification and regression. To our knowledge, we are the first to show the generalizability of a self-supervised learning model with motion data from wearables across distinct evaluation tasks.

Problem

• Wearable Devices
• Action Recognition
• Data Analysis

Innovation

• RelCon
• Unsupervised Learning
• Wearable Device Data