You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
abstract: "We propose LResNet, a new residual connection in the Lorentz formulation of hyperbolic spaces. With provable properties including guaranteed numerical stability and generalizing previous methods, LresNet addresses previous limitations such as runtime inefficiencies, numerical instability, mapping errors, and lack of geometric meaning on the manifolds.",
27
-
impact: "Beyond theoretical guarantees, we demonstrate the improvement achieved by LResNet in building hyperbolic deep learning models, conducting extensive experiments to show its superior performance in graph and image modalities across CNNs, GCNs, and graph Transformers.",
27
+
impact: "Beyond theoretical guarantees, we demonstrate the improvements achieved by LResNet in building hyperbolic deep learning models, where we conduct extensive experiments to show its superior performance in graph and image modalities across CNNs, GNNs, and graph Transformers.",
0 commit comments