Improving future statistical analysis for researchers and students

A recently funded grant could change how researchers in multiple disciplines measure statistics. Sewoong Oh, CSL assistant professor, is looking at how to improve machine data analysis in his project, “Robust Intelligence: Small: Information theoretic measure of

dependencies and novel sample-based estimators.”

One of the first things Sewoong, an assistant professor of Industrial and Enterprise Engineering, will address is the practical issues already present in these analyses. This includes situations in which data is made up of continuous and discrete variables or when data is highly correlated. The research will look into how the data estimators can overcome these practical challenges and potentially provide new analyses to researchers.

Results of this project could have an immediate impact on biological datasets and analyzing deep neural networks. Analyses of inter-layer dependencies of the networks, such as the information bottleneck, provide valuable insight into the training process. The proposed estimators can provide efficient tools to compute mutual information and other dependency measures, giving it the potential for a greater long-term impact. Having ways to find new data and analyze data could influence data collection in fields like genomics, machine-learning, biology and artificial intelligence and others.

In addition to the results on future data collection, the project will have an immediate and ongoing impact on students at the University of Illinois Urbana-Champaign. As part of the broader impact portion of the grant, Oh plans to create graduate-level course on statistical learning to share his work. Undergraduate students will also have the opportunity to be involved with the project and experience academic research firsthand.