While maintaining a perfect GPA, Di has solved an open problem in theoretical Computer Science with applications to computer vision, and begun research on practical application of his ideas to image processing.

Di's work focuses on the NP-hard problem of Quadratic Unconstrained Binary Optimization (QUBO), which captures the essence of a variety of vision problems. Prior work by Prof. Endre Boros has identified a family of lower bounds of the optimal value, denoted by C2,C3,C4,..., and C2 can be computed using network flow. Two questions are of great interest about these bounds: 1) Can our understanding about C2 be generalized to C3,C4,..., and 2) Can we identify partial optimal assignment from the computation of these lower bounds. For the particular bound C3, Di's work answers both questions in the affirmative.