Tight Dimensionality Reduction for Sketching Low Degree Polynomial Kernels

Neural Information Processing Systems 

We revisit the classic randomized sketch of a tensor product of $q$ vectors $x_i\in\mathbb{R}^n$.