Tool Design

TeachTOK, a Collaborative Interactive Machine Teaching Application

During my research, I also design tools, including TeachTOK, developed for the first project of my thesis on Collaborative Interactive Machine Teaching (CIMT). Built with the Marcelle Toolkit, TeachTOK enables a collaborative workflow where users take on the role of teachers, training an image classifier to recognize categories of their choice while communicating and coordinating throughout the task. In our study, we used TeachTOK to investigate how users negotiate teaching strategies, balance different objectives, and shape model behavior together. The code is available on GitHub and the application can be tried online. If you want to know more about TeachTOK design and implementation, there is a paper about it in French. I would be thrilled to hear about your interest in TeachTOK, its applications in your domain, or ideas for future collaborations.

Auditing Tool

For my second project on User-Driven Algorithm Auditing, I developed an interactive auditing tool that enables novice users to explore an image captioning model’s behavior and identify problematic patterns such as bias. The first version, used in our study, supports auditing through input manipulation and pattern discovery. It is available to try online, with source code on GitHub. A second version incorporates an affinity diagram interface, allowing users to organize model outputs and create narratives of its behavior. While this version is not yet online, you can see my own attempt with it here. In this not-so-secret nor personal-anymore project, I used my intimate photo archive in an autoethnographic audit to reflect on bias, identity, and diasporic experience. I plan to release this version online soon so you can use it to tell your own story.