AI and the World's Hard Problems · Quantum physics & quantum computing · arXiv · 2026-10-20
LiftDiff: Guided Discrete Diffusion for Quantum Code Construction
Fault-tolerant quantum computation needs quantum error-correcting codes with both a high encoding rate and sufficient distance. The authors introduce LiftDiff, which trains a discrete diffusion model to generate base matrices that are expanded into lifted product codes, guided by an energy function that scores code properties. The method found codes such as [[169,13,11]] and [[247,19,14]], and several of them outperformed the Gross code across a broad range of tested physical error rates.