Ramya Korlakai Vinayak

Mathematics—and its potential to increase our understanding of how the world works—has always fascinated Ramya Korlakai Vinayak. In grad school, this enthusiasm for math, combined with her curiosity about the use of data and algorithms for making predictions, scientific discoveries, and decisions that benefit society, piqued her interest in machine learning.

Ramya is interested in data that either comes from, or is about, people. Her research focuses both on algorithms and query design for learning from this data. The data used to train algorithms for machine learning is commonly prepared through a process called labelling, where it is tagged with descriptive information. When people label data, they often make mistakes, and their errors and biases are amplified in machine learning models. Ramya has received support through the American Family Funding Initiative to develop methods for providing high-quality, affordable training data for specialized machine learning applications in industry.

Ramya collaborates with domain experts in her work, with the goal of producing algorithms that capture the variability in large, diverse populations. She stresses the importance of ensuring that people from different demographic backgrounds are represented in data. Ideally, this results in better machine learning models and inferences, leading to better decisions and policies in sectors like healthcare that impact people’s lives.

“If we just rely on historic data, sometimes we might be carrying the biases that were in that data, and then there is a real problem of that getting amplified more and more when we train algorithms using that data. My hope is that we create environments where experts from different domains, as well as stakeholders, can come together to address these questions.”