Machine Teaching-Based Efficient Labelling for Cross-unit Healthcare Data Modelling
- Publisher:
- Springer International Publishing
- Publication Type:
- Chapter
- Citation:
- AI 2021: Advances in Artificial Intelligence, 2022, 13151 LNAI, pp. 320-331
- Issue Date:
- 2022-01-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
A data custodian of a big organization (such as a Commonwealth Data Integrating Authority), namely teacher, can easily build an intelligent model which is well trained by comprehensive data collected from multiple sources. However, due to information security and privacy-related regulation requirements, full access to the well-trained intelligent model and the comprehensive training data is usually limited to the teacher only and not available to any unit (or branch) of that organization. Therefore, if a unit, namely student, needs an intelligent function similar to the trained intelligent model, the student has to train a similar model from scratch using the student’s own dataset. Such a dataset is usually unlabelled, requiring a big workload on labelling. Inspired by the Iterative Machine Teaching, we propose a novel collaboration pipeline. It enables the teacher to iteratively guide the student to select samples that are most worth labelling from the student’s own dataset, which significantly reduces the requirement for human labelling and, at the same time, prevents regulation and information security breaches. The effectiveness and efficiency of the proposed pipeline is empirically demonstrated on two publicly available healthcare datasets in comparison with baseline methods. This work has broad implications for the healthcare sector to facilitate data modelling in instances where the large labelled datasets are not accessible to each unit.
Please use this identifier to cite or link to this item: