Tensorization of neural networks for improved privacy and interpretability

We present an algorithm for constructing tensor-train representations of functions. The method only requires black-box access to the target function and a small set of sample points defining the domain of interest. We demonstrate that our approach can be used to enhance the privacy and interpretability of neural network models. Additionally, we show that it can serve as an efficient initialization method for optimizing tensor trains in general settings, and that, in model compression, our algorithm achieves a superior trade-off between memory and time complexity compared to conventional tensorization methods of neural networks.