Recursive Sketched Interpolation: Efficient Hadamard Products of Tensor Trains
Recursive Sketched Interpolation (RSI) computes tensor-train Hadamard products with cubic complexity O(χ³) instead of the conventional O(χ⁴). It achieves this by recursively applying randomized tensor-train sketching and interpolative decomposition to maintain a compressed bond dimension throughout the process, offering superior scalability for large tensors.