Why not model tabular data correlations as an adjacency matrix? This transforms them into undirected networks, enabling to find optimal partitions through spectral clustering. I discovered the core, orthogonal skill groups of NBA players using this method.
For my graduate-level statistical mechanics final project, I explored the Fokker-Planck equation dynamics in diffusion generative models and examined how phase transitions relate to generating samples across different classes.
Check out my blog posts on the cool facts that I discovered in my research journey!