Stanford Webinar - Democratizing Model Discovery with Neural Networks

Professor Ellen Kuhl's Background and Research

• Professor Ellen Kuhl is a professor at Stanford's School of Engineering and focuses on improving human health through computational simulation.
• She is the founder of the Living Matter Lab, which uses physics-based tools embedded in machine learning to model living systems, such as the heart, arteries, lungs, muscles, and the brain.
• The lab collaborates with medical professionals and device designers to help develop therapies and understand the impact of forces on the heart.

Traditional Model Discovery

• In the past, researchers would select a model from a library of existing models that best fit their data points.
• This required expertise in mechanics and physics, limiting its use to experts in the field.
• The process of selecting a model was manual and time-consuming, relying on the researcher's knowledge and experience.

Automated Model Discovery

• The Living Matter Lab has developed a method to automate the model discovery process using neural networks.
• Neural networks are trained to select the best model based on input data, such as mechanical tests of tissues.
• The neural network output includes model parameters that represent stiffness, shear modulus, and other physical properties.
• The automated process removes the need for manual selection and allows non-experts to use these tools safely.

Constitutive Neural Networks

• Constitutive neural networks are a type of neural network that incorporates physics and engineering principles into the model discovery process.
• Physics parameters and constraints, such as symmetry and elastic behavior, are hardwired into the network.
• The resulting models are more interpretable, with parameters that have physical meanings.

Applications and Impact

• The automated model discovery process has applications beyond medical research, such as modeling the mechanical properties of artificial meat.
• In healthcare, it can be used to model the human brain and better understand conditions like Alzheimer's disease and traumatic brain injuries.
• It can aid in the design of medical devices, choosing the safest regions to perform surgery, and optimizing therapies for individual patients.
• The Living Matter Lab's code and resources for this technology are open-source and available on GitHub.
• The lab offers a graduate-level course in automated model discovery for professionals interested in learning more about the methods and applications.