Data Science, Humanely

Abstract

It is undeniable that machine learning has fundamentally changed what computers can do, especially as access to data sources and processing power continues to become easier. At the same time, the ability for us humans to actually make sense of these techniques has not progressed at nearly the same pace. In this talk, I will present two recent projects from our group which bring methods from data mining and machine learning into novel visualization techniques. The first project, Gaussian Cubes, provides interactive, low-latency visual exploration with models fit on hundreds of millions of samples. DimReader, on the other hand, shows how automatic differentiation -- the same technique that drives modern machine learning infrastructure such as Torch and TensorFlow -- can be used to provide a much deeper understanding of popular dimensionality reduction methods like t-SNE. Time permitting, I will present some additional recent work in the recent field of fairness in machine learning and automated decision making, specifically on runaway feedback loops and the assessment of black-box models.

Bio

Since 2014, Carlos Scheidegger is an assistant professor in the Department of Computer Science at the University of Arizona. He holds a PhD in Computing from the University of Utah, where he worked on software infrastructure for scientific collaboration. His current research interests are in large-scale data analysis, information visualization and, more broadly, what happens "when people meet data", for both good and bad. His research has been supported by both industry and the government through awards from the NSF and AT&T Labs, and his honors include multiple best paper awards including at the IEEE Visualization conference, and an IBM student fellowship.