Hohman’s “explorable” article The Beginner’s Guide to Dimensionality Reduction was created in collaboration with Matt Conlen of the University of Washington. Using a dataset of artworks from the Metropolitan Museum of Art in New York City, Hohman and Conlen explore the methods that data scientists use to visualize high-dimensional data.

Visualizing the myriad connections between all of the different features of each artwork in a high-dimensional graph could provide new insights. However, as Hohman says in the article, humans can’t see so many dimensions all at once.

Dimensionality reduction algorithms reduce the number of random variables by collecting a set of principal variables that retain the variation present in the data. This allows the data to be presented in fewer dimensions, which can be more easily processed by human viewers. This kind of projection is called an embedding.

The guide teaches users about embeddings and compares some of the most popular dimensionality reduction algorithms used today to create them. The article also contains a list of pros and cons for each of the algorithms to help readers use this technique for their own data. All of the algorithms mentioned are open-source Python implementations.

“Explorable and interactive articles are a great medium for teaching concepts that haven’t seen much usage and attention in academia yet,” said Hohman. “It’s really great to see recognition for our article, which helps people learn and engage with complicated concepts through interactive visualizations that are easily accessible on the web,” said Hohman.

IEEE VIS is the flagship conference on visualization and visual analytics. Hohman was also a panelist at this year’s event, and his advisor, CSE Associate Professor Polo Chau, served as a co-organizer of VISxAI. IEEE VIS was held Oct. 21-26 in Berlin, Germany.

For more information on Georgia Tech’s presence at IEEE VIS, explore highlights with the GVU Center’s interactive overview.