LLM-Guided Exemplar Selection for Few-Shot Wearable-Sensor Human Activity Recognition

TL;DR

This work tackles data scarcity and semantic confusion in few-shot wearable HAR by introducing an LLM-Guided Exemplar Selection framework. The method generates semantic features and structured knowledge priors with an LLM, and fuses them with margin-based, graph-based, and hubness cues in a facility-location optimization to select compact, representative exemplars. A machine-learning gate splits static and dynamic activities, allowing specialized modeling, while a suite of classic classifiers is trained on the selected exemplars. On the UCI-HAR dataset, the approach yields notable improvements over traditional exemplar-selection baselines, with ablations showing the critical role of the facility-location step and semantic priors in boosting generalization under limited labels. Overall, the framework demonstrates that integrating semantic reasoning into exemplar selection can enhance interpretability and robustness in real-world, data-scarce HAR applications.

Abstract

In this paper, we propose an LLM-Guided Exemplar Selection framework to address a key limitation in state-of-the-art Human Activity Recognition (HAR) methods: their reliance on large labeled datasets and purely geometric exemplar selection, which often fail to distinguish similar weara-ble sensor activities such as walking, walking upstairs, and walking downstairs. Our method incorporates semantic reasoning via an LLM-generated knowledge prior that captures feature importance, inter-class confusability, and exemplar budget multipliers, and uses it to guide exemplar scoring and selection. These priors are combined with margin-based validation cues, PageRank centrality, hubness penalization, and facility-location optimization to obtain a compact and informative set of exemplars. Evaluated on the UCI-HAR dataset under strict few-shot conditions, the framework achieves a macro F1-score of 88.78%, outperforming classical approaches such as random sampling, herding, and $k$-center. The results show that LLM-derived semantic priors, when integrated with structural and geometric cues, provide a stronger foundation for selecting representative sensor exemplars in few-shot wearable-sensor HAR.

Figures (10)

• Figure 1: Motivation for the proposed LLM-Guided Exemplar Selection. (a) Overlapping signal distributions cause Ambiguity between similar walking activities. (b) LLM-derived semantic priors reflect conceptual closeness (e.g., walking vs. walking upstairs).
• Figure 2: Overall workflow of the proposed LLM-Guided Exemplar Selection Framework. The process consists of three main stages: (a) Exemplar Selection, which combines LLM-guided semantic reasoning and facility-location selection; (b) Training, where selected exemplars are used to train the ML models; and (c) Inference: the ML Gate sends each test sample to the static or dynamic ML models, and the arrows show which model group produces the final prediction.
• Figure 3: Prompt design for LLM-based semantic feature generation.
• Figure 4: Prompt design for LLM knowledge representation.
• Figure 5: LLM response for semantic feature synthesis prompt.