Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Models are Zero-Shot Recognizers for Activities of Daily Living

Gabriele Civitarese, Michele Fiori, Priyankar Choudhary, Claudio Bettini

TL;DR

The paper addresses the challenge of recognizing Activities of Daily Living (ADLs) from smart-home sensor data under limited labeled data. It introduces ADL-LLM, which converts sensor windows into natural language and uses pre-trained LLMs in zero-shot and few-shot prompting to identify ADLs without extensive labeling. On MARBLE and UCI ADL datasets, zero-shot ADL-LLM achieves recognition rates close to strong supervised baselines, and few-shot prompting with semantic example selection further improves performance in data-scarce scenarios. The work discusses prompts, open-world HAR, privacy considerations, model evolution, and robustness to noisy data, highlighting the practical potential of LLM-based ADL recognition for scalable, privacy-aware healthcare applications, with code available for reproducibility and future clinical deployment envisioned in SERENADE.

Abstract

The sensor-based recognition of Activities of Daily Living (ADLs) in smart home environments enables several applications in the areas of energy management, safety, well-being, and healthcare. ADLs recognition is typically based on deep learning methods requiring large datasets to be trained. Recently, several studies proved that Large Language Models (LLMs) effectively capture common-sense knowledge about human activities. However, the effectiveness of LLMs for ADLs recognition in smart home environments still deserves to be investigated. In this work, we propose ADL-LLM, a novel LLM-based ADLs recognition system. ADLLLM transforms raw sensor data into textual representations, that are processed by an LLM to perform zero-shot ADLs recognition. Moreover, in the scenario where a small labeled dataset is available, ADL-LLM can also be empowered with few-shot prompting. We evaluated ADL-LLM on two public datasets, showing its effectiveness in this domain.

Large Language Models are Zero-Shot Recognizers for Activities of Daily Living

TL;DR

The paper addresses the challenge of recognizing Activities of Daily Living (ADLs) from smart-home sensor data under limited labeled data. It introduces ADL-LLM, which converts sensor windows into natural language and uses pre-trained LLMs in zero-shot and few-shot prompting to identify ADLs without extensive labeling. On MARBLE and UCI ADL datasets, zero-shot ADL-LLM achieves recognition rates close to strong supervised baselines, and few-shot prompting with semantic example selection further improves performance in data-scarce scenarios. The work discusses prompts, open-world HAR, privacy considerations, model evolution, and robustness to noisy data, highlighting the practical potential of LLM-based ADL recognition for scalable, privacy-aware healthcare applications, with code available for reproducibility and future clinical deployment envisioned in SERENADE.

Abstract

The sensor-based recognition of Activities of Daily Living (ADLs) in smart home environments enables several applications in the areas of energy management, safety, well-being, and healthcare. ADLs recognition is typically based on deep learning methods requiring large datasets to be trained. Recently, several studies proved that Large Language Models (LLMs) effectively capture common-sense knowledge about human activities. However, the effectiveness of LLMs for ADLs recognition in smart home environments still deserves to be investigated. In this work, we propose ADL-LLM, a novel LLM-based ADLs recognition system. ADLLLM transforms raw sensor data into textual representations, that are processed by an LLM to perform zero-shot ADLs recognition. Moreover, in the scenario where a small labeled dataset is available, ADL-LLM can also be empowered with few-shot prompting. We evaluated ADL-LLM on two public datasets, showing its effectiveness in this domain.
Paper Structure (43 sections, 11 figures, 7 tables, 1 algorithm)

This paper contains 43 sections, 11 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data Scarcity in Human Activity Recognition
  4. Data-Driven Methods
  5. Knowledge-Based Methods
  6. Neuro-Symbolic Methods
  7. Using LLMs for Human Activity Recognition
  8. LLM-based approaches for supervised activity recognition
  9. Zero-shot approaches for human activity recognition using LLMs
  10. The ADL-LLM framework
  11. Problem Formulation
  12. Overall architecture of ADL-LLM
  13. The main components of ADL-LLM
  14. Sensor States Generation
  15. Segmentation
  16. ...and 28 more sections

Figures (11)

  • Figure 1: Overall architecture of ADL-LLM. When the pool of examples is empty ADL-LLM, acts as a zero-shot ADLs recognition method. Otherwise, it is a few-shot approach.
  • Figure 2: An example of Window2Text in action on the UCI ADL dataset
  • Figure 3: An example of system prompt in ADL-LLM for the UCI ADL dataset (Home B). The sets of locations, household elements captured by sensors, and activities are variables depending on the specific home where ADL-LLM is installed.
  • Figure 4: ADL-LLM: An example of input and corresponding output.
  • Figure 5: ADL-LLM: An example of the prompt in the few shot setting.
  • ...and 6 more figures

Theorems & Definitions (2)

  • Example 3.1
  • Example 3.2