Few-Shot Optimization for Sensor Data Using Large Language Models: A Case Study on Fatigue Detection

2025-05-24

Italian National Conference on Sensors

AI Summary

For sensor-based few-shot classification tasks—such as fatigue detection—that exhibit high inter-subject variability and low signal-to-noise ratios, conventional example selection methods struggle to identify representative samples. To address this, we propose HED-LM: a hybrid example selection framework that integrates Euclidean-distance-based preliminary filtering with temporal feature embedding, and—novelty introduced herein—leverages a large language model (GPT-4) for context-aware semantic relevance re-ranking of candidate examples. This paradigm synergistically combines numerical similarity and semantic consistency. Evaluated on accelerometer-based fatigue detection, HED-LM achieves a macro-F1 score of 69.13±10.71%, outperforming random selection and distance-only baselines by 16.6% and 2.3%, respectively. The method significantly enhances prompt quality in few-shot settings, thereby improving robustness and generalization under limited labeled data.

Technology Category

• Machine Learning: Dimensionality Reduction/Feature Selection
• Humans and AI: Human-in-the-loop Machine Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

In this paper, we propose a novel few-shot optimization with Hybrid Euclidean Distance with Large Language Models (HED-LM) to improve example selection for sensor-based classification tasks. While few-shot prompting enables efficient inference with limited labeled data, its performance largely depends on the quality of selected examples. HED-LM addresses this challenge through a hybrid selection pipeline that filters candidate examples based on Euclidean distance and re-ranks them using contextual relevance scored by large language models (LLMs). To validate its effectiveness, we apply HED-LM to a fatigue detection task using accelerometer data characterized by overlapping patterns and high inter-subject variability. Unlike simpler tasks such as activity recognition, fatigue detection demands more nuanced example selection due to subtle differences in physiological signals. Our experiments show that HED-LM achieves a mean macro F1-score of 69.13 ± 10.71%, outperforming both random selection (59.30 ± 10.13%) and distance-only filtering (67.61 ± 11.39%). These represent relative improvements of 16.6% and 2.3%, respectively. The results confirm that combining numerical similarity with contextual relevance improves the robustness of few-shot prompting. Overall, HED-LM offers a practical solution to improve performance in real-world sensor-based learning tasks and shows potential for broader applications in healthcare monitoring, human activity recognition, and industrial safety scenarios.

Problem

• Improving few-shot example selection for sensor-based classification
• Enhancing fatigue detection using hybrid distance-contextual filtering
• Optimizing sensor data performance with LLM-augmented selection

Innovation

• Hybrid Euclidean Distance with LLMs
• Filters and re-ranks examples contextually
• Improves few-shot sensor data classification