Interpretable and Interactive Machine Learning
Machine Learning has positioned itself as a hot topic and almost as a synonym to data analysis. However, most of the current real world ML systems use models as black-boxes. Given this, the big question that arises is what about if model performance is not the unique requirement to be fulfilled? To address this, two new sub-fields of research have been proposed to help users to interact and understand ML models, and by this opening the black-box Interactive ML and Interpretable ML. This master thesis project is developed in the context of these sub-fields, with applications to Exploratory Data Analysis (EDA) supported by the t-SNE algorithm, a common technique for dimensionality reduction of high-dimensional data. The contributions of this project are (1) provide a set of guidelines for designing more user-centric ML systems in terms of interactivity and interpretability and (2) design and develop of a visual tool to interact and interpret t-SNE models for performing EDA. The proposal is evaluated in two case studies, the animals dataset and the urban environment dataset from SALURBAL project.
People
José Tiberio Hernández
