1: \begin{abstract}

2:   Nonparametric extension of tensor regression is proposed.

3:   Nonlinearity in a high-dimensional tensor space is broken into

4:   simple local functions by incorporating low-rank tensor

5:   decomposition. Compared to naive nonparametric approaches, our

6:   formulation considerably improves the convergence rate of estimation

7:   while maintaining consistency with the same function class under

8:   specific conditions. To estimate local functions, we develop a

9:   Bayesian estimator with the Gaussian process prior. Experimental

10:   results show its theoretical properties and high performance in

11:   terms of predicting a summary statistic of a real complex network.

12: \end{abstract}

13: