To address the longstanding bottleneck in material structure analysis where traditional Electron Backscatter Diffraction (EBSD) methods force a trade-off between speed and accuracy, Assistant Professor Shao-Pu Tsai of National Taiwan University and Dr. Po-Yen Tung of Cambridge University developed “Latice,” a groundbreaking AI-driven approach. By utilizing a variational autoencoder (VAE) to process EBSD patterns, the team created a model that learns intrinsic physical properties—such as crystal rotational symmetry—achieving indexing speeds 7.5 times faster than conventional techniques alongside a massive 99.9% data compression rate. Executed by graduate student Yu-Chun Liu and supported by Walsin Lihwa and government funding, this “physics-aware” AI was validated on 316 stainless steel and published in Cell Reports Physical Science, offering a significantly more efficient solution for advanced material research.