Can we make humans understanding high-dimensional data?

Modern technologies in all application domains of data science like cyber-security, environment, health or social media, are the sources of very large and multidimensional datasets. While we witnessed 70 years of exponential increase of computing power, human cognitive capacity to understand these data remains the same. This growing gap calls for new ways to represent and summarize data enabling human discovery, analysis, and control. Visual Analytics attempts to bridge this gap using interactive visualizations and advanced processing and modeling techniques. I will present several approaches I have explored for the last 10 years to tackle this challenge: inferring topological invariants of latent structures from data, enriching dimension reduction layouts, and modeling human perception to design more trustworthy Machine Learning techniques.