Machine learning applied to quantum computing.
Quantum computers need error correction to be useful, and the software that does the correcting is currently one of the things standing between the machines that exist and the machines people want. These posts explain that problem for a machine learning audience, then measure where a learned decoder beats the standard classical one.
Quantum error correction explained for people who train models. No results, just the background I wish I'd had before starting.
How minimum-weight perfect matching works, the two assumptions it rests on, and the one design decision that makes the experiment mean anything.
The numbers. A learned decoder matches MWPM given a correct noise model and beats a realistic wrong one by 2x, and the failures are as informative as the wins.