{"id":"arxiv/2608.28941","url":"https://arxiv.org/abs/2608.28941","tags":["math.NA","cs.MS"],"score":0,"title":"A GPU-Accelerated Blocked Adaptive Randomized Range Finder Based on an Implicit Householder QR Decomposition","domain":"math.NA","source":"arxiv","authors":["Carolin Penke","Andreas Herten"],"reviews":0,"subject":"math.NA","summary":"Low-rank methods can reduce the memory and computational requirements of deep neural network training in approaches such as GaLore. Randomized range finders offer an attractive alternative to singular value decompositions, particularly when the required rank is determined adaptively from a prescribed approximation tolerance. We introduce a blocked adaptive randomized range finder based on an implicit Householder QR decomposition and an optimized hybrid CPU--GPU implementation. The proposed method avoids explicit reorthogonalization. Numerical experiments show that it preserves orthogonality…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-28","reactions":{}}