XDecomposer: Learning Prior-Free Set Decomposition for Multiphase X-ray Diffraction
Published in Conference on Neural Information Processing Systems (NeurIPS 2026), to appear, 2026
Real-world synthesis often produces complex mixtures whose constituent phases cannot be reliably disentangled from powder X-ray diffraction (PXRD). XDecomposer infers an unordered set of phase-resolved components, their mixture proportions and structural representations within a unified architecture, using phase-query-driven decomposition together with diffraction-consistent physical reconstruction. It substantially improves reconstruction accuracy and phase identification on simulated and experimental data.
Recommended citation: Hanyu Gao, Bin Cao, Yunyue Su, Tong-Yi Zhang, Qiang Liu. (2026). "XDecomposer: Learning Prior-Free Set Decomposition for Multiphase X-ray Diffraction." Conference on Neural Information Processing Systems (NeurIPS), to appear.
Download Paper