Justin Solomon, an associate professor at the Massachusetts Institute of Technology (MIT) Department of Electrical Engineering and Computer Science, is applying modern geometric techniques to solve complex problems in machine learning, data science, and computer graphics. He leads the Geometric Data Processing Group, half of which works on optimizing two- and three-dimensional geometric data in applications such as medical imaging and autonomous vehicle sensor technology. The rest of his team uses geometric tools for high-dimensional statistical research to improve generative AI models.
An important part of this research involves understanding how datasets are arranged in high-dimensional space. Certain geometrical similarities between datasets can offer insights into the performance of machine learning models, according to Solomon. He notes that such tools help in areas ranging from comparing training and testing datasets to complex applications in computer animation.
Much of Solomon’s interest in the intersection of geometry, computer science, and machine learning was sparked by a high school internship where he developed 3D face recognition algorithms. His subsequent experience at Pixar Animation Studios throughout college further reinforced his interest in the mathematical challenges within visual content and computer graphics. In grad school, Solomon focused on the problem of optimal transport, which aims to characterize the most efficient way to distribute items across different points.
In addition to leading his research team, Solomon seeks to increase diversity in the field through the Summer Geometry Initiative, a six-week research program for underrepresented undergraduate students. The program offers participants an introduction to geometric research and has been influential in diversifying the demographics of incoming doctoral students not just at MIT, but also at other institutions.
Looking ahead, Solomon plans to apply geometric tools to improve unsupervised machine learning models. In unsupervised machine learning, models must learn to recognize patterns without labeled training data, a complex process that could greatly benefit from geometric insights, Solomon explains.
In his leisure time, he enjoys playing classical music on the piano or cello, often participating in local orchestras. With his professional work closely connected to artistic practice, he notes a “mutual benefit” between his research and his passion for music.
