Multi-physical learning framework for optimizing ice cream sensory appeal

Overview

This innovative solution leverages a hybrid machine learning (ML) and molecular dynamics (MD) framework to predict and optimize the sensory appeal of ice cream. By combining Latent-Variable Gaussian Process (LVGP) models with Bayesian Optimization (BO), this framework explores multiple sensory outcomes, guiding both formulation and process design. The approach provides a comprehensive understanding of how ingredient interactions influence sensory qualities, enabling the discovery of innovative ice cream formulations.

Technical specifications

Key features:

• LVGP Model: Develops a response surface from formulation, processing, and sensory data, transforming input variables into meaningful latent representations.
• Bayesian Optimization: Utilizes acquisition functions to explore Pareto fronts, optimizing multiple sensory outcomes efficiently and with minimal experimentation.
• Molecular Dynamics Simulations: Captures interfacial interactions between ingredients, enriching the framework with physical insights into structure-property relationships.
• Causal Inference Methods: Identifies true cause-effect relationships using Directed Acyclic Graphs (DAGs) and other analytical techniques.

Technology readiness level

This ice cream innovation framework software is at Technology Readiness Level 6, indicating that it has been validated in a relevant environment and is ready for further development and potential commercialization.