Student: Jakob Gürtl
Github: @jguertl
Organisation: KNIME
Mentor: @ChristianAlbrecht, @dietzc
The goal of this project is to implement an interactive JavaScript view to enable “Active Learning” within KNIME Analytics Platform. While KNIME Analytics Platform already contains a JavaScript enabled node for the manual annotation of samples (Table Editor), a dedicated active learning node would allow users to interact with a KNIME workflow more intuitively and allow for active learning specific functionality such as a labeling wizard or fast multiobject labeling. After the implementation phase, the person working on the project is encouraged to build active learning examples that involve text or images in order to test implemented node(s).
Before starting to implement code, information about current methods used in the field of active learning has been gathered. The key findings of this research task can be found here.
In addition, an user interface design for the new node has been created. The focus lies on assigning labels to object instances.
As part of the implementation phase, several features have been realized. All commits can be found here. The code branch can be found here.
- Node dialog options
- Queue table view
- Detail tile view
- Done table view
- Label class buttons
- Label class edit dialog
After the implementation phase, the node has been tested with newly created workflows which involve text and image objects. An example workflow can be found attached.