Search papers, labs, and topics across Lattice.
This paper introduces stochastic Riemannian optimizers tailored for tree tensor networks (TTNs), enhancing their application in machine learning. By employing a hybrid CNN-TTN architecture and testing on datasets like Fashion-MNIST and CIFAR10, the authors demonstrate that these optimizers can achieve predictive performance on par with traditional unconstrained methods. Notably, the proposed techniques also facilitate numerically stable compression, addressing a critical challenge in deploying tensor networks in practical scenarios.
Stochastic Riemannian optimizers for tree tensor networks achieve competitive performance while enabling stable compression, revolutionizing their utility in machine learning applications.
Tensor networks, originally developed for quantum many-body physics, are promising models for machine learning. We derive stochastic Riemannian optimizers for tree tensor networks (TTNs) on both their parameter and quotient manifolds, including adaptive and learning-rate-free schemes suitable for minibatch training. Using a hybrid CNN-TTN architecture, we evaluate the methods on Fashion-MNIST, CIFAR10, and Imagenette. The proposed optimizers achieve predictive performance comparable to unconstrained optimization while enabling numerically stable downstream compression.
