Designing a Multi-Touch Interface to Build Trust in Machine Learning Models

As the use of machine learning applications is increasing in our everyday lives, there is a need to allow users with a diverse set of expertise in machine learning (ML) to incorporate ML models in their workflows. This requires creating tools that make ML models easy to understand, which can be challenging as most of these models are internally complex with intertwined parameters. This thesis aims to provide a solution to this problem by leveraging visual analytics techniques to model these complex relationships between the different parameters through an interactive multi-touch interface. The system also supports real-time user interactions with the data to allow domain experts to steer the model using their domain knowledge. By modelling these relationships through an intuitive interactive visualization, we aim to create a system that bridges the gap between domain experts and building ML models. Results from the user study show an increase in the perceived levels of trust after using our system, Embodied Machine Learning (EML).