Researchers have developed STELLA, a novel framework designed to enable large language models (LLMs) for on-device human activity recognition (HAR). This system efficiently translates raw sensor data into compact latent tokens, which are then processed by a frozen LLM. STELLA also incorporates on-device personalization by adapting a lightweight tokenizer, allowing for continuous improvement as user-specific data accumulates. The framework achieves state-of-the-art performance across multiple datasets and settings, demonstrating real-time inference capabilities within practical mobile and edge computing constraints. AI
IMPACT Enables efficient, private, and personalized LLM-based human activity recognition on edge devices.
RANK_REASON The item is a research paper detailing a new framework for LLM application in human activity recognition. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Human Activity Recognition
- Nirhoshan Sivaroopan
- ScienceCast
- STELLA
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
