{"id":"arxiv/2609.06672","url":"https://arxiv.org/abs/2609.06672","tags":["quant-ph","cond-mat.dis-nn"],"score":0,"title":"Efficient Representation of multicategorical local Hilbert spaces: nonlinear Restricted Boltzmann Machines to Kolmogorov-Arnold Networks","domain":"quant-ph","source":"arxiv","authors":["Abhiroop Lahiri"],"reviews":0,"subject":"quant-ph","summary":"Neural network quantum states (NQS) for representing spin-$\\frac{1}{2}$ systems are typically built from Multi-Layer Perceptrons (MLPs) with binary visible variables, in which the input spins are first combined through linear affine maps before more expressive nonlinear transformations are applied. The standard way for representing multi-categorical systems, namely spin-1 and spin-2, or the q-state quantum Potts systems with more than two local degrees of freedom, is a unary or one-hot encoded MLP architecture. Following a few similar studies, I observe that while the one-hot construction is…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-09-06","reactions":{}}