Takanori Fujiwara

High-dimensional data contains a rich set of measured observations of a phenomenon and is ubiquitous for data analysis. However, its high dimensionality makes data analysis challenging. One promising approach to solve this challenge is to extract essential information by applying dimensionality reduction (DR) and then visualize the result for further data analysis. In this talk, I will discuss three aspects necessary to advance this analytical approach: (1) develop interactive DR methods, (2) investigate the reliability of DR results, and (3) design effective visualizations for DR results. Specifically, I will present new DR algorithms that are designed to interactively compare data groups and uncover the uniqueness of each group. Second, I will demonstrate how existing DR algorithms can potentially hide prominent data patterns and lead to biased analytical insights, and I will present new algorithms designed to mitigate such potential biases. Lastly, I will showcase two types of 3D visualizations: a data physicalization (i.e., physical representation of data) and an immersive dome-scale visualization. Advancing these three aspects will enable a more effective, trustworthy, and intuitive way of communicating and deriving insights from high-dimensional data.