Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VaCDA: Variational Contrastive Alignment-based Scalable Human Activity Recognition

Soham Khisa, Avijoy Chakma

TL;DR

VaCDA tackles domain shift in wearable HAR by combining a variational autoencoder–driven latent space with contrastive learning to align and discriminate features across multiple source sensors. The method comprises a VAE-based pretraining pipeline and a classifier that leverages shared latent representations, optimized with a joint loss that includes $L_{ ext{VAE}}$ and two contrastive terms, enabling scalable multi-source adaptation. Empirical results on four public HAR datasets show improvements in cross-position and cross-device settings, with ablations confirming the positive contribution of contrastive learning, though cross-person performance is mixed. The work advances practical HAR under heterogeneous sensor conditions and lays groundwork for future extensions such as open-set DA and novel-category recognition.

Abstract

Technological advancements have led to the rise of wearable devices with sensors that continuously monitor user activities, generating vast amounts of unlabeled data. This data is challenging to interpret, and manual annotation is labor-intensive and error-prone. Additionally, data distribution is often heterogeneous due to device placement, type, and user behavior variations. As a result, traditional transfer learning methods perform suboptimally, making it difficult to recognize daily activities. To address these challenges, we use a variational autoencoder (VAE) to learn a shared, low-dimensional latent space from available sensor data. This space generalizes data across diverse sensors, mitigating heterogeneity and aiding robust adaptation to the target domain. We integrate contrastive learning to enhance feature representation by aligning instances of the same class across domains while separating different classes. We propose Variational Contrastive Domain Adaptation (VaCDA), a multi-source domain adaptation framework combining VAEs and contrastive learning to improve feature representation and reduce heterogeneity between source and target domains. We evaluate VaCDA on multiple publicly available datasets across three heterogeneity scenarios: cross-person, cross-position, and cross-device. VaCDA outperforms the baselines in cross-position and cross-device scenarios.

VaCDA: Variational Contrastive Alignment-based Scalable Human Activity Recognition

TL;DR

VaCDA tackles domain shift in wearable HAR by combining a variational autoencoder–driven latent space with contrastive learning to align and discriminate features across multiple source sensors. The method comprises a VAE-based pretraining pipeline and a classifier that leverages shared latent representations, optimized with a joint loss that includes and two contrastive terms, enabling scalable multi-source adaptation. Empirical results on four public HAR datasets show improvements in cross-position and cross-device settings, with ablations confirming the positive contribution of contrastive learning, though cross-person performance is mixed. The work advances practical HAR under heterogeneous sensor conditions and lays groundwork for future extensions such as open-set DA and novel-category recognition.

Abstract

Technological advancements have led to the rise of wearable devices with sensors that continuously monitor user activities, generating vast amounts of unlabeled data. This data is challenging to interpret, and manual annotation is labor-intensive and error-prone. Additionally, data distribution is often heterogeneous due to device placement, type, and user behavior variations. As a result, traditional transfer learning methods perform suboptimally, making it difficult to recognize daily activities. To address these challenges, we use a variational autoencoder (VAE) to learn a shared, low-dimensional latent space from available sensor data. This space generalizes data across diverse sensors, mitigating heterogeneity and aiding robust adaptation to the target domain. We integrate contrastive learning to enhance feature representation by aligning instances of the same class across domains while separating different classes. We propose Variational Contrastive Domain Adaptation (VaCDA), a multi-source domain adaptation framework combining VAEs and contrastive learning to improve feature representation and reduce heterogeneity between source and target domains. We evaluate VaCDA on multiple publicly available datasets across three heterogeneity scenarios: cross-person, cross-position, and cross-device. VaCDA outperforms the baselines in cross-position and cross-device scenarios.
Paper Structure (20 sections, 6 equations, 4 figures, 3 tables, 1 algorithm)

This paper contains 20 sections, 6 equations, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Discrepancy-based DA Approaches
  4. Adversarial Learning-based DA Approaches
  5. VAE-based DA Approaches
  6. Contrastive Learning-based DA Approaches
  7. Methodology
  8. Problem Setup
  9. Proposed Framework
  10. Optimization and Label Prediction
  11. Experiments
  12. Datasets
  13. Baselines
  14. Evaluation Criteria
  15. Implementation Details
  16. ...and 5 more sections

Figures (4)

  • Figure 1: An illustration of domain adaptation in human activity recognition. The image highlights data distribution heterogeneity due to device placement and user behavior.
  • Figure 2: VaCDA pertaining pipeline; it combines VAE and contrastive learning. The encoder processes available data from all domains. The framework optimizes reconstruction loss, KL divergence, and contrastive loss for domain adaptation.
  • Figure 3: VaCDA classifier pipeline; The labeled input from source domains is processed by the pre-trained encoder and optimized by cross-entropy loss.
  • Figure 4: VaCDA performance comparison with and without contrastive learning