Pierre Bonami

I am a Principal Developer at Gurobi Optimization based in Madrid, Spain.

Background

My career spans academia and industry, with positions at:

  • Gurobi Optimization — Principal Developer (current)
  • IBM — Research Scientist in the CPLEX team.
  • CNRS — Permanent Researcher at LIF Marseille (now LIS).
  • Carnegie Mellon University — Postdoctoral research in optimization

I earned my PhD in Operations Research at Paris 6 (LIP6), focusing on algorithms for Mixed-Integer Nonlinear Programming.

Currently, I work on the algorithms and solvers that power Gurobi, one of the world’s leading mathematical optimization platforms. My work spans Mixed-Integer Programming (MIP), Mixed-Integer Nonlinear Programming (MINLP), cutting-planes, and interactions of optimization with Machine Learning.

I am the creator and project manager of Bonmin (Basic Open-source Nonlinear Mixed INteger programming), an open-source C++ solver for convex MINLP problems, distributed through the COIN-OR initiative.

Research Interests

My research spans Mixed-Integer Programming (MIP), Mixed-Integer Nonlinear Programming (MINLP), and cutting-plane theory. I design algorithms that tighten LP relaxations via split disjunctions and wide split cuts, solve nonconvex quadratic programs via integer programming, automate Benders decomposition inside solvers, and apply machine learning to guide algorithm selection. A growing focus of my work is the interaction between optimization and machine learning: using ML to improve solver decisions, and conversely, embedding trained predictors inside optimization models.

Gurobi Machine Learning

I maintain gurobi-machinelearning, an open-source Python package that lets you embed trained regression models — from scikit-learn, Keras/TensorFlow, or PyTorch — directly inside a gurobipy optimization model. This makes it possible to optimize over the input space of a learned predictor using Gurobi’s exact MIP solver, bridging the gap between machine learning and mathematical optimization. Full documentation is available on ReadTheDocs.

Bonmin

I created Bonmin (Basic Open-source Nonlinear Mixed INteger programming), an open-source C++ solver for convex MINLP, distributed through COIN-OR. Bonmin is very loosely maintained.