How AI could bring a scientific renaissance: The SOS+CD lab

The Science of Science and Computational Discovery (SOS+CD) lab focuses on the innovative use of AI and computational methods to advance scientific research and discovery. 

When we think of scientific research, said Daniel Acuna, the head of the SOS+CD lab, "it can be hard to remember that scientists are people too."

The science of science, he explained, seeks to unravel the complexities and imperfections of who gets to make scientific discoveries, what those discoveries are and how those discoveries are credited and amplified. Computational discovery, on the other hand, seeks to help scientists discover new knowledge through AI and other computational methods. 

From detecting altered images in research papers to improving information retrieval and fairness in AI systems, each member of the SOS+CD lab is engaged in projects that aim to solve complex problems in scientific research. 

Let's learn from members of the lab about what their research entails. 

Tyler Gorman (Second Year, MS, Computer Science)

Tyler Gorman is working to see if he can detect digitally altered images using AI. This is valuable because the pressure to publish can lead people to edit images, which can have large real-world consequences, such as the case of potentially doctored Alzheimer's research stretching across two decades. 

Gorman sees a goal of the SOS+CD lab as automating the impossible task of large-scale review. 

"At the end of the day, you just can't go through tens of thousands to hundreds of thousands of papers. But we could apply this tool to research papers and see if a percentage are photoshopped," he said.

When asked what he sees as the common thread of the lab, Gorman said, "We're all trying to take a hammer and microscope and look at different parts of the science of science."

Gorman said he believes that this variety of experience in the lab is a valuable part of research in a university. 

"Everyone has their own expertise, and no one is smart enough to know everything. If they were, you wouldn't need grad school," Gorman said. 

Carolina Chavez-Ruelas (First Year, PhD, Computer Science)

Carolina Chavez-Ruelas is working on a project examining whether the socioeconomic status of academic faculty influences their research interests. She is advised by Acuna and fellow computer science professor Aaron Clauset, who also investigates the science of science. 

Chavez-Ruelas sees the SOS+CD lab as a place where the power of AI can be used for something beyond private industry. 

"I think this lab is about how those tools can be used for science in general, be it discovery of unexplored gaps or maybe just helping researchers perform better and automatizing tasks that are tedious," she said.  

Chavez-Ruelas said that having a strong research interest is the most important part of deciding whether to pursue a PhD.

She also acknowledges that the process of getting a PhD is complex.

 "I think I was a bit lost when I first started my PhD, and I feel very fortunate that both Daniel and Aaron have been my advisors. It's been nice to have that guidance," she said.

Pawin Taechoyotin (Second Year, MS, Computer Science)

Pawin Taechoyotin is working on several projects related to representation learning, a machine-learning technique that breaks down complex data into simpler representations so that its relationship to other data can be seen more easily. 

Taechoyotin said this is similar to how humans learn to simplify the complex entity of an apple into an image of an apple or the word apple. 

"A limitation of AI currently is that it cannot learn something beyond the data it is given, but if we could apply the way that humans learn back to the AI, we could potentially have AI that can identify or create new concepts," Taechoyotin said. 

Among other projects, Taechoyotin is working on representation learning for the images and text of research papers. His goal is to enable researchers to retrieve papers that are similar to their own, but not the same, in the hopes of sparking creative thinking. 

Taechoyotin said that he enjoys the collaborative space the lab brings, saying that you never truly work alone in academia because it limits your ideas. 

"Also, humans are social animals, we need to talk to one another. You cannot help humanity if you can't communicate your work." 

Shubham Sati (Second Year, MS, Computer Science)

Shubham Sati is working on building a recommender system for citations in research papers. 

The system will use a dataset of 100 million research papers to provide accurate and context-based recommendations for citations. 

"Let's say you write a sentence and you know you read it somewhere but you're not able to cite it properly; you can't remember where you read it,” Sati said. “Take that sentence and feed it into my system, and it'll give you recommendations for citation." 

This work can help ensure people don't plagiarize unintentionally, and could help uncover similar scientific literature that improves their work.

Sati was inspired to pursue this research after taking a class with Acuna and working as a backend engineer at Amazon. 

Sati plans to continue working on systems engineering, working with computer science Professor Shivakant Mishra to reduce the amount of data duplication in databases by exploiting shared memory channels for edge computing systems. 

He said that graduate school offers a valuable opportunity to delve deeper into research and expand one's knowledge and expertise.

Meysam Varasteh (First Year, PhD, Computer Science)

Meysam Varasteh is working on the fairness of recommender systems toward women and Black people. He's working both with Acuna and Department of Information Science Chair and Professor Robyn Burke. 

"I think society in general isn't always fair, and I think that I cannot change the world, but maybe I can change the AI that people use every day, like Netflix, YouTube, Google, and other applications," Varasteh said. 

All of these applications and many others use what are called recommender systems, which recommend what you might want to watch next by what you've watched previously, and they can easily be biased. 

During his master's degree, Varasteh read a research paper every day, and he said it helped him greatly. He became inspired to pursue this research after reading one of Burke's papers on recommender systems. He said he also deeply appreciates Acuna's guidance. 

"He's a really supportive advisor," Varasteh said.

Varasteh, like Chavez-Ruelas, said he believes that you have to be committed to research to find a PhD useful.