Branch-And-Model (BAM) Optimization Tool

A global optimization solver for black-box and data-driven problems.
BAM Licenses

What is BAM?

BAM (Branch-And-Model) is a derivative-free global optimization algorithm designed to solve complex optimization problems where objective functions and constraints are not available in algebraic form. BAM is ideal for problems involving: 

• Expensive simulations
• Laboratory experiments
• Hyperparameter tuning
• Black-box or data-driven modeling

BAM intelligently navigates the search space using surrogate modeling and a novel domain partitioning strategy to minimize costly evaluations, making it highly efficient for optimization based on expensive simulations or laboratory experiments.

Key Features

• Black-box optimization: Optimize systems without requiring gradient or algebraic model information.
• Flexible input: Work with simulations, experimental setups, or pre-collected data.
• Efficient sampling: Uses a unique subdivision algorithm for global search convergence.
• Model-driven strategy: Builds and optimizes local surrogate models powered by ALAMO.
• Built-in optimization engine: Employs BARON to globally solve surrogate models.
• Support for discrete variables: Handles integer and categorical inputs seamlessly.
• Automation: Fully automated optimization without manual intervention.

How It Works

BAM follows a four-stage process for each iteration:

1. Domain Partitioning: Divides the search space into boxes using all previously evaluated points.
2. Box Selection: Identifies promising regions using a Lipschitz-inspired filter (without needing a Lipschitz constant).
3. Local Surrogate Modeling: Builds accurate, sparse models around selected points using ALAMO.
4. Evaluation & Update: Solves surrogate models with BARON to select new sample points.

This loop repeats until BAM meets a stopping criterion (e.g., function call or time limit).

Applications

BAM is particularly well-suited for:

How to Run BAM

BAM runs from the command line and reads a simple text-based input file. Users define:

• Number and bounds of variables
• Data provider executable (simulator or experimental data collecting program)
• Evaluation budget and options

A typical run looks like:

bam my_problem.bam

Outputs are logged both on-screen and in a .lst file for later analysis.