The Inverse Design Lab: Multi-Objective Bayesian Optimization for Complex Formulations

• Introduction to Gaussian Processes (GPs) for modeling uncertainty in experimental data.
• Practical setup: Building a GP model in Python (BoTorch or GPyTorch) using experimental spreadsheet data (e.g., Composition % vs. Compressive Strength).
• Hands-on: Configure and run a GP model using personal experimental data.
• Deliverable: A plot visualizing the uncertainty and predictions from the Gaussian Process model.

• Defining Acquisition Functions to guide AI in balancing exploration and exploitation.
• Visualizing the Pareto Frontier for trade-offs (e.g., Strength vs. Sorption). Learn how to optimize competing objectives simultaneously.
• Hands-on: Run a closed-loop simulation where AI suggests the next 5 experiments to maximize two conflicting variables.
• Deliverable: Visualize the Pareto Frontier and interpret the optimal balance between competing properties.

• Integrating human intuition with AI through Human-in-the-loop AI for refining experimental designs.
• Hands-on: Create a recommender system that takes previous experiment results and generates the next optimal synthesis recipe.
• Deliverable: A Python script that recommends the next experiment based on past results, optimizing for the desired outcomes.

Take your research to the next level with NanoSchool.