{"id":"arxiv/2606.08758","url":"https://arxiv.org/abs/2606.08758","tags":["quant-ph"],"score":0,"title":"Neural network decoder confidence as a learned proxy for the logical gap","domain":"quant-ph","source":"arxiv","authors":["David Dentelski"],"reviews":0,"subject":"quant-ph","summary":"To utilize quantum error-correcting codes, a decoder must infer the logical sector from the measured syndrome. Beyond a hard logical decision, some decoders provide soft information estimating its reliability. For minimum-weight perfect matching (MWPM), a common confidence measure is the complementary, or logical, gap. Neural-network decoders trained with binary cross entropy naturally output a different score, the logit, whose relation to the matching gap remains unclear. Here, we test whether the logit of a graph neural network (GNN) decoder behaves as a learned proxy for the logical gap.…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-06-07","reactions":{}}