Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CLAD-Net: Continual Activity Recognition in Multi-Sensor Wearable Systems

Reza Rahimi Azghan, Gautham Krishna Gudur, Mohit Malu, Edison Thomaz, Giulia Pedrielli, Pavan Turaga, Hassan Ghasemzadeh

TL;DR

CLAD-Net (Continual Learning with Attention and Distillation), a framework enabling wearable-sensor models to be updated continuously without sacrificing performance on past tasks, is proposed, demonstrating robustness to label scarcity.

Abstract

The rise of deep learning has greatly advanced human behavior monitoring using wearable sensors, particularly human activity recognition (HAR). While deep models have been widely studied, most assume stationary data distributions - an assumption often violated in real-world scenarios. For example, sensor data from one subject may differ significantly from another, leading to distribution shifts. In continual learning, this shift is framed as a sequence of tasks, each corresponding to a new subject. Such settings suffer from catastrophic forgetting, where prior knowledge deteriorates as new tasks are learned. This challenge is compounded by the scarcity and inconsistency of labeled data in human studies. To address these issues, we propose CLAD-Net (Continual Learning with Attention and Distillation), a framework enabling wearable-sensor models to be updated continuously without sacrificing performance on past tasks. CLAD-Net integrates a self-supervised transformer, acting as long-term memory, with a supervised Convolutional Neural Network (CNN) trained via knowledge distillation for activity classification. The transformer captures global activity patterns through cross-attention across body-mounted sensors, learning generalizable representations without labels. Meanwhile, the CNN leverages knowledge distillation to retain prior knowledge during subject-wise fine-tuning. On PAMAP2, CLAD-Net achieves 91.36 percent final accuracy with only 8.78 percent forgetting, surpassing memory-based and regularization-based baselines such as Experience Replay and Elastic Weight Consolidation. In semi-supervised settings with only 10-20 percent labeled data, CLAD-Net still delivers strong performance, demonstrating robustness to label scarcity. Ablation studies further validate each module's contribution.

CLAD-Net: Continual Activity Recognition in Multi-Sensor Wearable Systems

TL;DR

CLAD-Net (Continual Learning with Attention and Distillation), a framework enabling wearable-sensor models to be updated continuously without sacrificing performance on past tasks, is proposed, demonstrating robustness to label scarcity.

Abstract

The rise of deep learning has greatly advanced human behavior monitoring using wearable sensors, particularly human activity recognition (HAR). While deep models have been widely studied, most assume stationary data distributions - an assumption often violated in real-world scenarios. For example, sensor data from one subject may differ significantly from another, leading to distribution shifts. In continual learning, this shift is framed as a sequence of tasks, each corresponding to a new subject. Such settings suffer from catastrophic forgetting, where prior knowledge deteriorates as new tasks are learned. This challenge is compounded by the scarcity and inconsistency of labeled data in human studies. To address these issues, we propose CLAD-Net (Continual Learning with Attention and Distillation), a framework enabling wearable-sensor models to be updated continuously without sacrificing performance on past tasks. CLAD-Net integrates a self-supervised transformer, acting as long-term memory, with a supervised Convolutional Neural Network (CNN) trained via knowledge distillation for activity classification. The transformer captures global activity patterns through cross-attention across body-mounted sensors, learning generalizable representations without labels. Meanwhile, the CNN leverages knowledge distillation to retain prior knowledge during subject-wise fine-tuning. On PAMAP2, CLAD-Net achieves 91.36 percent final accuracy with only 8.78 percent forgetting, surpassing memory-based and regularization-based baselines such as Experience Replay and Elastic Weight Consolidation. In semi-supervised settings with only 10-20 percent labeled data, CLAD-Net still delivers strong performance, demonstrating robustness to label scarcity. Ablation studies further validate each module's contribution.

Paper Structure

This paper contains 27 sections, 12 equations, 6 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Approach
  4. System overview
  5. Theoretical Background
  6. CLAD-Net
  7. Self-supervised Transformer
  8. Supervised Convolutional Neural Network
  9. Theoretical Motivation
  10. Experiments
  11. Benchmarks
  12. PAMAP2 Physical Activity Monitoring pamap2_physical_activity_monitoring_231
  13. Daily and Sports Activities daily_and_sports_activities_256 (DnSA)
  14. RealWorld HAR sztyler2016realworld
  15. Evaluation Metrics
  16. ...and 12 more sections

Figures (6)

  • Figure 1: Fine-tuning the model on new subjects leads to a decline in accuracy on previously seen subjects
  • Figure 2: The complete pipeline of the proposed system, composed of three core components: data collection, data preprocessing, and the final model, CLAD-Net.
  • Figure 3: Overview of CLAD-Net, including its architectural components and training workflow.
  • Figure 4: Visualization of four different time-series augmentation methods applied to a sample accelerometer signal.
  • Figure 5: Forgetting comparison across different models under 10% and 20% label availability on the PAMAP2, DnSA and RealWorld datasets. CLAD-Net consistently shows lower forgetting than other methods.
  • ...and 1 more figures