Search papers, labs, and topics across Lattice.
This work constructs and analyzes a class of 1-Lipschitz neural networks specifically designed for Hadamard manifolds, leveraging Busemann functions to create geometry-preserving layers. The proposed architecture demonstrates robustness in two numerical experiments: achieving robust classification on the Poincaré disk and enhancing masked-Wishart covariance reconstruction using SPD-valued denoisers. Results indicate significant improvements over traditional denoising methods, showcasing the effectiveness of the nonexpansive denoiser in maintaining stability under hyperbolic perturbations.
Achieving robust neural network performance on non-Euclidean spaces could redefine stability benchmarks in machine learning applications.
Controlling the Lipschitz constant of a neural network is a standard way to promote robustness and stability. Most existing constraining strategies are designed for Euclidean spaces. In this work, we construct and analyze a class of 1-Lipschitz neural networks on Hadamard manifolds. Our layers are of gradient-descent type, $1$-Lipschitz, and quasi-$α$-firmly nonexpansive. The core building blocks of the proposed architecture are Busemann functions, and we exploit the properties of Busemann gradient flows to design $1$-Lipschitz geometry-preserving layers. We provide explicit constructions and examples for hyperbolic manifolds and the manifold of symmetric positive definite (SPD) matrices. We test the proposed architecture in two numerical experiments: robust classification on the Poincaré disk and masked-Wishart covariance reconstruction. On the Poincaré disk, the proposed networks yield robust classifiers under hyperbolic perturbations. On the SPD manifold, we train SPD-valued denoisers and adopt them as a Plug-and-Play prior for a masked-Wishart covariance reconstruction problem. We show improved results from the nonexpansive denoiser over static, data-only, and Log-Euclidean denoising baselines, and empirically test its convergence properties.
