MILP-StuDio is an innovative mixed integer linear programming (MILP) generation framework proposed by Professor Wang Jie's team (MIRA Lab) at the University of Science and Technology of China in 2024. The full name is Block Structure Decomposition, which aims to generate high-quality MILP instances by preserving the problem block structure to improve the solution quality of AI operations research solvers. The relevant paper results are "MILP-StuDio: MILP Instance Generation via Block Structure Decomposition", has been accepted by NeurIPS 2024, a top artificial intelligence conference.

The core idea of MILP-StuDio is to identify the block structure in the MILP constraint coefficient matrices (CCMs) and decompose the instance into block units, which serve as the basis for building MILP instances. On this basis, three operators are designed to build new instances by removing, replacing and appending block units in the original instance, so as to generate instances of different sizes.