GPU Parallelization of MIO Primal Heuristics

Develop effective methods for exploiting GPU parallelization in mixed-integer optimization primal heuristics.

Background

The paper places mixed-integer optimization within the broader development of GPU-accelerated optimization algorithms. Although GPUs have enabled efficient parallel solution of continuous optimization problems, the paper identifies the extension of GPU parallelization to primal heuristics for mixed-integer optimization as an unresolved methodological challenge. The paper addresses this challenge by introducing PaNGEA, a GPU-friendly primal heuristic that combines batched linear-relaxation solves, local search, and parallel generation and exploration of restricted subproblems.

References

However, exploiting GPU parallelization for MIO primal heuristics remains an open challenge.

— PaNGEA: Parallel Node Generation and Exploration Algorithm on GPU  (2610.03090 - Pauphilet et al., 2 Oct 2026) in Section 1, Introduction