Postdoctoral Associate
Massachusetts Institute of TechnologyHello, this is Bo Tang, an operations researcher.
My work draws on machine learning and modern computing to develop reliable and scalable methods and reusable computational frameworks for decision-making.
H-PDLP: GPU-Accelerated Linear Programming via Generalized PDHG and Homogeneous Self-Dual Embedding
arXiv preprint, 2026
Develops a generalized primal-dual hybrid gradient method for a compact homogeneous self-dual embedding with linear convergence guarantees. Enables matrix-free GPU computation of solutions and infeasibility certificates for large-scale linear programs.
Learning to Optimize for Mixed-Integer Non-linear Programming with Feasibility Guarantees
INFORMS Journal on Computing (Major revision)
Learns to solve parametric mixed-integer nonlinear programs without an optimization solver during training or inference, using differentiable integer correction and a gradient-based feasibility projection with convergence guarantees. Produces feasible, high-quality solutions efficiently.
PyEPO: a PyTorch-based end-to-end predict-then-optimize library for linear and integer programming
Mathematical Programming Computation, 2024
Introduces a unified framework and benchmarks integrating linear and integer programming solvers with PyTorch and JAX for decision-focused learning, which trains predictive models on the quality of the decisions they lead to rather than on prediction error alone.
I currently work with Haihao Lu at MIT Sloan. I received my PhD in Operations Research from the University of Toronto in 2025, advised by Elias B. Khalil. During my PhD, I spent summer 2023 as a research intern at Pacific Northwest National Laboratory, working with Ján Drgoňa.
I'll be at the 2026 INFORMS Annual Meeting in San Francisco, November 1–4, presenting my work on GPU-accelerated linear programming and joining a panel on GPU-accelerated optimization. If you're attending, I'd be happy to connect.