adaptive doe

Adaptive DOE is a design of experiments approach that dynamically modifies the experimental plan based on incoming results — using algorithms (Bayesian optimization, reinforcement learning) to select each next experiment to maximize information gain or expected improvement.

How Adaptive DOE Works

• Initial Points: Start with a small space-filling design (Latin Hypercube, random).
• Surrogate Model: Fit a model (Gaussian process, random forest) to current data.
• Acquisition Function: Select the next experiment to maximize Expected Improvement, Knowledge Gradient, or other criteria.
• Iterate: Run the experiment, update the model, select the next point. Repeat until convergence.

Why It Matters

• Efficiency: Converges to the optimum in 2-5× fewer experiments than classical DOE.
• Expensive Experiments: Ideal when each experiment is costly (real wafers, long process times).
• Non-Standard: Can handle constraints, noisy responses, and multi-fidelity evaluations.

Adaptive DOE is experiments guided by AI — using models to choose the most valuable next experiment in real time.