Prof. Bareinboim Named IEEE AI's 10 to Watch

2-22-2016
Writer(s): Jesica E. Hollinger

Assistant Professor Elias Bareinboim, one of the CS Department’s most recent hires, was recently named one of AI’s 10 to Watch – a biennial list of young stars in the fieldElias Bareinboim compiled by the Institute of Electrical and Electronics Engineers (IEEE).

Bareinboim joined the department in Fall 2015 with an appointment in CS and a courtesy appointment in Statistics. He conducted his graduate studies under the Turing Award winner Judea Pearl. Bareinboim’s research focuses on causal inference and their applications to data-driven fields, including the health, behavioral, and social sciences. This is a rapidly evolving field that lies on the boundary between computer science, statistics, and the empirical sciences.

“One important application of the theory is policy evaluation — each time the government is considering to deploy a new program, its causal impact must first be assessed. If we do not have the tools to conduct this type of assessment, we end up with claims that are disproven when the program is in fact implemented, millions of dollars are spent and people may have been hurt," Bareinboim said. “The need for evaluating causal effects is pervasive in many facets of our society. For example, it may appear in genetics when evaluating the effect of a gene on a certain phenotype, in medicine when considering the effects of a new drug, or in business, when assessing the effect of a new advertisement campaign. The list goes on.”

One of the main focuses of Bareinboim’s research is the problem of policy evaluation in ‘big data’ scenarios, which he notes, is different than in traditional settings and comes with its own opportunities and challenges. “It is challenging because data are coming from all over the place, and the sources are almost invariably heterogeneous – each dataset entails a different type of bias, it may be confounding, selection, or many others. The opportunity comes from the fact that from any single dataset we may not be able to figure out certain causal relations, but after accounting for commonalities and differences, they beautifully emerge from a combination of several datasets.”

Prof. Bareinboim’s work is credited with being the first to formalize and systematize the data-fusion problem, bringing to light the principles and mechanics that license the combination of datasets collected from disparate studies conducted under different experimental conditions and from heterogeneous populations. “After formalizing the mathematics behind data-fusion, we can now play scientists and combine our dataset in a principled way. We can go on and play computer scientists and design algorithms to streamline and automate the process of scientific discovery,” he explained.

Dr. Pearl spoke about Bareinboim’s research, “The data-fusion challenge, decorated with titles such as "external validity" or "meta-analysis," is one of those perennial, blue sky problems that I never thought would be solved in our generation. I was surprised every time Elias came back with another piece of the puzzle solved and, now that he obtained a complete and general solution, scientists are much better equipped to tackle causal inference challenges than at any time in the past.” He went on to say,  “Anyone who has multiple heterogeneous datasets and wants to make a causal claim using these data must first understand Elias’s solution.”

Prof. Bareinboim’s other recognitions include the 2014 AAAI Outstanding Paper Award, the Dan David Prize Scholarship, the Yahoo! Key Scientific Challenges Award, and the UCLA's Edward K. Rice Outstanding Doctoral Student Award. 

To learn more about Bareinboim's research, visit his webpage (www.cs.purdue.edu/~eb/).