{"id":"arxiv/2606.02938","url":"https://arxiv.org/abs/2606.02938","tags":["math.OC","eess.SY"],"score":0,"title":"Koopman operator learning for predictive control via Khatri-Rao kernel regression","domain":"math.OC","source":"arxiv","authors":["Mircea Lazar"],"reviews":0,"subject":"math.OC","summary":"This paper develops a data-driven realization of the generalized Koopman operator (GeKo) [1], in which states and inputs are lifted independently and the dynamics are expressed as a tensor bilinear system. The first contribution is a time-sequenced multi-step Khatri-Rao kernel regression formulation that exposes the operator to evolved snapshots along trajectories rather than only single one-step pairs, which reduces compounded prediction error. Secondly, we develop a kernel- and input-agnostic structured SVD reduction that compresses the lifted state and input spaces while preserving the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-01","reactions":{}}