Johannes Knittel
The Symbiosis of Visualization and AI: Explaining Large Models and Obtaining Insights into Big Data
When: Tuesday 17th of January, 14:30h
Where: Amphitheatre, Bat 660 How to get to there?
Abstract
The combination of machine learning and visualization has received increased attention in recent years, in part due to the impactful advances of recent neural architectures and models for natural language processing and human-like text or image generation. In this talk I will first summarize and characterize the different aims and tasks of recent approaches that combine ML and VIS. The main goal of my talk is to illustrate the symbiosis of interactive visualizations and machine learning techniques for two important pillars of VIS and ML: visually explaining large (language) models and obtaining more comprehensive insights into big datasets. Several visual analytics approaches have been developed to better understand, assess, and debug machine learning models, taking advantage of the fact that we can benefit from our visual cognition for pattern mining in addition to simply parsing the outputs of algorithmic analyses. However, interpreting what a model has learnt may also reveal interesting insights about the training data itself. I will talk about that, on the one hand, making ML models more visually interpretable not only helps to better understand said models, but it also offers new ways of leveraging advanced ML techniques to extract and visualize more complex, multidimensional relationships in large datasets. On the other hand, developing better ways of visualizing large corpora as well as multidimensional relationships will also advance the visual analysis of machine learning models with ever growing sets of weights and training data.
Bio:
Johannes Knittel received his PhD in Computer Science from the University of Stuttgart in 2022, working on the visual analysis of large text collections and streaming data, as well as on extracting and visualizing multidimensional relationships in multivariate datasets. His research focuses on the intersection of machine learning and interactive visualizations. He is particularly interested in developing new approaches for visually explaining large machine learning models, and how we can leverage machine learning techniques to gain insights into big datasets.