Trade-offs in Latent Representations of Microstructure Evolution

Investigated how different latent space representations impact the modeling of microstructure evolution in materials. Explored and compared dimensionality reduction techniques—such as autoencoders and diffusion maps—to analyze trade-offs between reconstruction accuracy, interpretability, and predictive capability. This work supports improved machine learning approaches for understanding and forecasting material behavior over time.

Refer paper (Desai et al., 2024) for details.

For code refer GitHub Repository.