Charlie Marx ’20 Publishes and Presents Research on Machine Learning
The senior computer science and mathematics double major has his research accepted by NeurIPS and will present his work on understanding the variables that go into human-facing model predictions at the upcoming NeurIPS conference in Vancouver.
For humans, following the golden rule means treating everyone equally, and Charlie Marx ’20 is making sure that computers and artificial intelligence are held to the same standard.
A computer science and mathematics double major, Marx is headed to Vancouver later this month to present his work on reducing bias in machine-learning models in order to produce more equitable results. His paper is titled “Disentangling Influence: Using Disentangled Representations to Audit Model Predictions.”
"For example, a model may use a person’s race in making a prediction by using the proxy variable of zip code.” said Marx in the description of his paper. “This ‘indirect influence’ can be difficult to detect, and has important ethical implications when models are used in high-impact domains.”
In Vancouver, Marx will present at the NeurIPS conference to a crowd of experts in the field of neural information processing systems. Accompanying this, Marx’s research will be published in the prestigious NeurIPS, which is a rare opportunity for an undergraduate student. The research paper is coauthored by Marx, Richard Phillips ’18 (who is currently pursuing a Ph.D. in computer science at Cornell University), and Assistant Professor of Computer Science Sorelle Friedler.
"Charlie has been the lead author in developing and executing the ideas in the paper—it really is his paper,” said Friedler. “The acceptance rate at NeurIPS is about 19%, so it's a big accomplishment to have a paper published there, even for established researchers, and is unusual for undergraduates to be lead authors on such papers.”
Marx, who also minors in statistics, began researching this topic two years ago, when he received funding from the Arnold & Mabel Beckman Foundation to analyze images of the deep sea using machine learning.
"We first encountered the problem when working with the deep-sea dataset, and soon recognized it was similar to issues we see when working on fairness in machine learning,” he said. “We were fortunate to have motivating examples in both the natural sciences and social sciences while designing our methods. We hope our tools will be useful to researchers across many disciplines.”
Marx is now looking forward to sharing his work at the NeurIPS Conference, which runs from December 8 to 14.
"It is inspiring to have the opportunity to engage with researchers from around the world at a top-publication venue,” he said. “I hope conversations with other researchers at the conference will allow us to build upon and improve our current tools.”
After graduation this spring, Marx hopes to pursue a Ph.D. in machine learning, so this is definitely only the beginning of his publishing career.
"A special thanks to my advisors Sorelle Friedler and Helen White, as well as our co-authors Richard Phillips ’18, Carlos Scheidegger, and Suresh Venkatasubramanian,” he said. “We were very fortunate to have a great group of researchers on this project.”