Deep Generative Domain Adaptation with Temporal Attention for Cross-User Activity Recognition

TL;DR

Cross-user HAR faces distribution shifts between users and non-$i.i.d.$ time-series data, formalized as $P^{Source}\neq P^{Target}$ with source labels available and target labels inferred. The authors propose DGDATA, a deep generative domain adaptation framework with Temporal Relation Attention that aligns source and target distributions using CVAE-based generation and adversarial learning to capture user-invariant temporal relations. DGDATA comprises three interacting components—fine-grained feature representation, common temporal relations characterization with Temporal Relation Attention, and cross-user classifier learning—trained in an iterative loop with a gradient reversal layer. Evaluations on OPPORTUNITY, PAMAP2, and DSADS show DGDATA achieving superior cross-user accuracy, demonstrating the practical value of incorporating temporal dynamics into domain adaptation for sensor-based HAR.

Abstract

In Human Activity Recognition (HAR), a predominant assumption is that the data utilized for training and evaluation purposes are drawn from the same distribution. It is also assumed that all data samples are independent and identically distributed ($\displaystyle i.i.d.$). Contrarily, practical implementations often challenge this notion, manifesting data distribution discrepancies, especially in scenarios such as cross-user HAR. Domain adaptation is the promising approach to address these challenges inherent in cross-user HAR tasks. However, a clear gap in domain adaptation techniques is the neglect of the temporal relation embedded within time series data during the phase of aligning data distributions. Addressing this oversight, our research presents the Deep Generative Domain Adaptation with Temporal Attention (DGDATA) method. This novel method uniquely recognises and integrates temporal relations during the domain adaptation process. By synergizing the capabilities of generative models with the Temporal Relation Attention mechanism, our method improves the classification performance in cross-user HAR. A comprehensive evaluation has been conducted on three public sensor-based HAR datasets targeting different scenarios and applications to demonstrate the efficacy of the proposed DGDATA method.

Figures (17)

• Figure 1: The design principle overview of DGDATA.
• Figure 2: The outline of DGDATA.
• Figure 3: The architecture of VAE model.
• Figure 4: The outline of DGDATA data distributions mapping.
• Figure 5: The network architecture of feature extractor.