A typical artificial intelligence class teaches students how to use and understand existing AI tools. But too often, these classes neglect to explain the complex math behind them, said Aaron Lauda, the divisional dean for physical sciences and mathematics at USC.

“You get some sort of rudimentary understanding of why the algorithms work, but you’re not going to be the person who’s making the next algorithm,” Lauda said.

This fall, the Dornsife College of Letters, Arts and Sciences launched a new master’s of science graduate program in mathematical data science to teach students the math behind AI algorithms and how to build them. 

“These students will actually be more equipped to understand when an algorithm works and also be able to modify it,” Lauda said.

Nathan Nguyen, a junior majoring in mathematics and the co-president of DataSC — a club for students interested in AI and data science — said that now is a good time to develop a program focused on the mathematical foundations of data science.

“Especially in an age of AI where you could basically just tell ChatGPT ‘Create me this model’ or just ‘Make this code for me,’ it’s more important than ever that we’re actually understanding the algorithms we’re creating, how to modify them and how to improve them,” Nguyen said.

Xiaohui Chen, an associate professor of mathematics and the co-director of the program, said the two-year program will teach its students mathematical theories, such as how to extract information from data sets, and give students hands-on experience applying their knowledge.

The degree currently features a set of industry speaker events, Chen said. At these events, the speakers not only provide students with practical applications of their work but also frequently share their own data sets with students. These data sets will be used for a capstone project that students undertake during their second year.

The capstone project will involve students being separated into groups, with each trying to find improvements for a given data set.

“Students can have different solutions, wrangle with the data and play around with AI and machine learning models with this data,” Chen said. “They’re trying to get a dirty, hands-on experience with the real data.”

Lauda said that students will also engage with real-world companies or USC’s scientific departments during their second year — learning to apply the mathematical background they developed in the first year.

In an era when AI systems seem to be sparking concerns that computer science degrees may hold less value than in the past, Lauda said this new major will help students become “future-proof.”

“[AI algorithms] have made it more and more user-friendly to the point where you can ask ChatGPT to code your Python now,” he said. “One of the ways to stay future-proof is to be someone who understands the mathematics and analytical thinking behind those.”

The adoption of the new master’s program came in part from the popularity of the statistics master’s program, Lauda said. The statistics program had two types of students: those who wanted to earn a Ph.D. and those who wanted to work in the field immediately.

“The problem is those groups are often very different,” Lauda said. “If you’re running a very theoretical-based math class, it may not resonate as well with some of the master’s students … [The new program’s] classes are dedicated to people who want to build that math background, but also then go and be able to do projects and work at a company.”

Chen said that the program can empower students to improve current large language models like ChatGPT. He also said that the existing models are inefficient at processing data.

“You have a very powerful racing car, but it’s going to have a few gas problems, so you consume a lot of gas,” Chen said. “Data science is a part of the gas, and AI is the engine.”

He said it’s the role of this new cohort of students to figure out how to make AI more efficient using data science techniques.

Although data science is frequently approached from a computer science and programming perspective, Nguyen said that more input from math students can help strengthen the field.

“I’m hoping [the program] can attract more people who are interested in mathematics into developing more advanced algorithms for the purposes of data science and advancing the field,” Nguyen said.