DomusFM: Foundation Model for Smart-Home Behavioral Monitoring
A new foundation model named DomusFM has been developed by researchers for monitoring behavior based on events within smart homes. This model overcomes the drawbacks of current methods that depend on supervised learning, which often necessitates an impractical volume of labeled activity data. Unlike traditional foundation models that concentrate on continuous inertial or physiological information, DomusFM extracts transferable representations from sparse, discrete, and semantically rich event streams generated by smart-home sensors. The research, detailed in a paper on arXiv (2602.01910), emphasizes the promise of smart-home monitoring in healthcare, independent living, and early detection of cognitive or functional changes. DomusFM aims to enhance smart-home technology by effectively capturing variability among residents and environments without the need for extensive labeled datasets.
Key facts
- DomusFM is a foundation model for event-based behavioral monitoring in smart homes.
- It addresses limitations of supervised approaches that require impractical labeled data.
- Current foundation models are designed for continuous inertial or physiological data, not sparse event streams.
- Smart-home monitoring supports healthcare, independent living, and early detection of functional or cognitive changes.
- Tasks include activity recognition, prediction, and pattern discovery.
- The paper is available on arXiv with ID 2602.01910.
- DomusFM learns transferable representations of latent behavioral patterns.
- The model captures variability across residents and environments.
Entities
Institutions
- arXiv