Three students received special recognition this year for their research presented at the annual Center for Statistics and Machine Learning (CSML) undergraduate research poster session: Kavya Chaturvedi ’21, Princeton School of Public and International Affairs, Byron Chin ’21, Department of Mathematics, and Alexandria Skarzynski ’21, Department of Sociology.
“We are proud of the effort the students poured into their projects, which are innovative in the datasets analyzed, the questions asked, and the theoretical results produced,” said Peter J. Ramadge, the CSML director.
Kavya Chaturvedi’s project, “Can Words Speak Louder Than Actions? A Text Analysis Based Evaluation of the U.K. Modern Slavery Act,” uses quantitative methods to analyze the impact of the U.K. Modern Slavery Act, passed in 2015. There are an estimated 40.3 million victims of modern slavery around the world, according to the Global Slavery Index. These are adults and children forced to sew clothes, make bricks for buildings, or mine cobalt for smartphones and other electronics. Policymakers have tried to combat this through various channels, such as the U.K. Modern Slavery Act, passed in 2015. A specific requirement of the act is for large companies to publish “a statement detailing what they are doing to mitigate the risk of slavery in their supply chain.”
Her project uses “multinomial inverse regression for text analysis to predict a company’s labor performance score based on their modern slavery statement. These corporate labor scores are then regressed against financial indicators to assess whether there exists a financial incentive for improved labor conditions,” explains Chaturvedi.
One of her biggest takeaways from the project is that machine learning-based text analysis can streamline elements of the laborious process to assess a company’s impact on human rights. Chaturvedi can also use her quantitative approach to distinguish companies taking the risks of modern slavery seriously and those doing the bare minimum. Chaturvedi’s advisor was Professor Omar Wasow, assistant professor of politics.
Alexandria Skarzynski became interested in intermarriages between Americans and non-citizens from her family experience and by watching the reality T.V. show, 90-Day Fiance. She spun this interest into her CSML research project, “Investigating Matching Patterns and Status Exchange in U.S. Citizen and Non-U.S. Citizen Intermarriages Using the Exchange Index.” She undertook this research in her junior year under her advisor, Yu Xie, the Bert G. Kerstetter ’66 University Professor of Sociology, and the Princeton Institute for International and Regional Studies.
“I was looking at how U.S. citizens and non-U.S. citizens would pick their marital partners based on their education and their socioeconomic status, given the fact that those are already forms of social capital. And you throw citizenship into the mix - that’s also a form of social capital since it gives you access to certain rights,” said Skarzynski. “I was wondering whether we would see a different kind of status exchange in those kinds of relationships compared to just marriages between U.S. citizens, and if so, how much that would be?”
For her analysis, she used linear regression and nearest neighbor parametric matching. After running her process, Skarzynski came up with some interesting conclusions: American men who marry out from their country tend to marry women of lower socioeconomic status. In comparison, American women pick husbands that tend to have higher socioeconomic status than American men.
“I think it was cool to be able to take such a bizarre interest, 90-Day Fiance, and make it into a piece of really awesome independent work. And so that was exciting to see these results and understand this issue in a deeper way,” said Skarzynski.
Byron Chin explores community detection algorithms in a more theoretical study, “Optimal Reconstruction of Block Models.” Community detection encompasses methods to find “interesting structure within networks, which can be anything from social media networks to biochemical networks,” said Chin.
“While there are effective algorithms in practice, the theory behind these algorithms is not as well understood,” said Chin. “Why do they work well?”
His project for the CSML certificate focused on advancing the theoretical understanding of community detection, specifically on the sparse stochastic block model. These models generate graphs, a mathematical structure that contains some objects that are related.
What Chin set out to do is to reconstruct the block model from the resultant graph. As detailed in the 2016 paper “Belief Propagation, Robust Reconstruction and Optimal Recovery of Block Models” published in The Annals of Applied Probability, a pathway to do so exists. This paper - one of the co-authors is Mathematics Professor Allan Sly and Chin’s advisor – reconstructs a graph in a limited use case.
Chin expands on this research by showing that their process can reconstruct a greater variety of sparse stochastic block models than previously been thought.
Chin was attracted to this problem because he enjoys working on probability and statistics.
“Going through my classes, it pushed me into this theoretical area,” said Chin. “It’s an interesting avenue of inquiry, and I am excited and gratified that CSML recognized my work.”