Bridging AI and Materials Science: Unlocking Insights with Autoencoders

This talk is designed for data scientists, AI researchers, and materials scientists interested in the intersection of machine learning and experimental science. If you've ever wondered how AI can reveal hidden patterns in real-world materials data, this session is for you! I’ll share how we use autoencoders, a type of neural network, to decode the relationship between X-ray absorption spectra (XAS) and structural properties of nanomaterials. This is a crucial challenge in materials science, where understanding atomic-scale structures can lead to breakthroughs in catalysts, batteries, and fuel cells. You’ll learn about MAVEN, our multi-tasking variational autoencoder that balances reconstruction, denoising, and descriptor mapping to create a more structured and interpretable latent space. By doing so, we not only improve data-driven insights but also enable AI and science experts to collaborate more effectively. We’ll explore real experimental cases—like distinguishing structural changes in palladium nanoparticles—and compare our AI-driven approach with traditional spectroscopy methods. Ultimately, this talk highlights how machine learning can accelerate scientific discovery and invites the AI community to partner with domain experts to solve complex real-world problems. Join us to see how AI can revolutionize materials science!