ADAPT-AI

Domain Adaptation for Time Series Data

ADAPT-AI addresses a key challenge in data-driven applications: machine learning models often do not transfer well to new environments, resulting in substantial manual effort for data annotation. The project develops methods for efficiently adapting time-series models to new domains, such as different sensor configurations or operating environments.

Building on existing research, the project advances adaptation methods for time-series data and integrates active learning approaches to minimize annotation requirements. The resulting toolbox will be implemented and evaluated with MotionMiners, focusing on the adaptation of human activity recognition models across different logistics environments and sensor setups. By reducing development time and annotation costs, ADAPT-AI aims to improve the scalability and economic viability of machine-learning-based solutions.

Publications

Moh’d Khier Al Kfari, Stefan Lüdtke. Domain Adaptation in Human Activity Recognition through Self-Training. Companion of the 2024 on ACM International Joint Conference on Pervasive and Ubiquitous Computing (Ubicomp Workshops) 2024. [web]

Contact

Moh’d Khier Al Kfari

mohd.kfari@uni-rostock.de