[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

DailyLLM: Context-Aware Activity Log Generation Using Multi-Modal Sensors and LLMs

Created by
  • Haebom

Author

Ye Tian, Xiaoyuan Ren, Zihao Wang, Onat Gungor, Xiaofan Yu, Tajana Rosing

Outline

In this paper, we propose DailyLLM, the first activity log generation and summarization system that comprehensively integrates four dimensions of contextual information: location, motion, environment, and physiological information, using only common sensors of smartphones and smartwatches. DailyLLM enables high-dimensional activity understanding by integrating a lightweight LLM-based framework with structured prompting and efficient feature extraction. It is proposed to overcome the limitations of existing methods in accuracy, efficiency, and semantic richness, and achieves 17% improved BERTScore precision and nearly 10x faster inference speed than the state-of-the-art method with 70B parameters using a 1.5B-parameter LLM model.

Takeaways, Limitations

•
Takeaways:
◦
Demonstrates that rich, context-aware activity logs and summaries can be generated using just smartphone and smartwatch sensors.
◦
Achieving both high accuracy and efficiency through a lightweight LLM-based framework.
◦
Efficient deployment even on low-spec devices such as personal computers and Raspberry Pi.
◦
High-level activity understanding possible through integration of four-dimensional situational information.
•
Limitations:
◦
Currently we use LLM with 1.5B parameters, but using larger LLMs could potentially improve performance (as implied by the 1.5B vs 70B comparison results).
◦
Further validation of generalization performance across diverse sensor data and environments is needed.
◦
Consideration needs to be given to privacy and data security.
◦
An efficient strategy for data accumulation and management is needed for long-term use.
👍
Made with Slashpage