These online demos accompany the following paper published at the ACM UIST '21 conference:
Jules Françoise, Baptiste Caramiaux, Téo Sanchez. Marcelle: Composing Interactive Machine Learning Workflows and Interfaces. Annual ACM Symposium on User Interface Software and Technology (UIST ’21), Oct 2021, Virtual, France. DOI: 10.1145/3472749.3474734
Case Study 1: Sketch Recognition
This scenario involves a HCI researcher who focuses on democratizing ML systems for the general public. To that end, she runs workshops and studies with playful scenarios where participants can teach concepts to a classifier and she collects data on user interactions. This demo illustrates how Marcelle can be used to develop prototypes for a sketch recognition application.
Initial Prototype
Detailed Dashboard
Final Application
Case Study 2: Skin Cancer Recognition
The second scenario focuses on the collaboration between a machine learning expert and a clinician to build a skin lesion classifier. The machine learning expert, Louise, trains models with Python and logs the training to a Marcelle data store. She develops two dashboards for monitoring the training and assessing the performance of various models. She can share particular models with Michel, a clinician who can test the classifier with his own images, correcting the predictions if necessary.
For testing, example data from the HAM10000 dataset can be downloaded from Harvard Dataverse.