{"id":"arxiv/2603.25530","url":"https://arxiv.org/abs/2603.25530","tags":["stat.ML","math.NA"],"score":0,"title":"Adaptive Subspace Modeling With Functional Tucker Decomposition","domain":"stat.ML","source":"arxiv","authors":["Noah Steidle","Joppe De Jonghe","Mariya Ishteva"],"reviews":0,"subject":"stat.ML","summary":"Tensors provide a structured representation for multidimensional data, yet discretization discards the underlying continuous structure when the data originate from continuous processes. We address this limitation by introducing a functional Tucker decomposition (FTD) that embeds a mode-wise continuity constraint directly into the factorization. The FTD models the continuous mode as a function in a reproducing kernel Hilbert space (RKHS), avoiding a prespecified basis while preserving the multilinear subspace structure of the Tucker model. We derive a reconstruction error bound for the…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-03-26","reactions":{}}