{"id":"arxiv/2607.18745","url":"https://arxiv.org/abs/2607.18745","tags":["cs.LG","cs.IT","math.NA"],"score":0,"title":"Contraction-Gauge Preconditioning for Quantized Matrix Multiplication","domain":"cs.LG","source":"arxiv","authors":["Piyush Sao","Narasinga Miniskar","Pedro Valero-Lara","Keita Teranishi","Sudip Seal"],"reviews":0,"subject":"cs.LG","summary":"We study low-precision computation of C=AB with both factors quantized. We derive an exact finite-dimensional identity for the expected squared product error under independent, zero-mean entrywise errors with known variance fields; it holds exactly for non-overloading subtractive dither and for independent stochastic rounding, and we empirically assess deterministic round-to-nearest (RTN). Using the product-preserving equivalence AB=(AT)(T^{-1}B), we formulate contraction-gauge preconditioning: jointly choosing a factor representation and its sharing pattern before quantization.…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-07-21","reactions":{}}