Search papers, labs, and topics across Lattice.
This paper introduces Equivariant Poincaré ResNets, which integrate hyperbolic geometry with discrete symmetry groups to enhance the optimization of visual representations in hyperbolic space. The authors address the inefficiencies of standard hyperbolic networks by implementing geometrically safe tensor reshaping and left-regular permutations, which significantly reduce the optimization space and improve convergence rates. Key results show that embedding equivariance not only accelerates training but also maintains the boundary constraints of the Poincaré ball and preserves spatial-group equivariance, leading to more efficient learning.
Embedding equivariance in hyperbolic networks can drastically accelerate convergence and reduce parameter redundancy, reshaping how we approach visual representation learning.
While recent advancements like the Poincaré ResNet have demonstrated the potential of learning visual representations directly in hyperbolic space, their optimisation remains hampered by the computationally intensive nature of Riemannian gradients and the strict boundaries of the manifold. Furthermore, standard hyperbolic networks treat spatial transformations of the same object as distinct hierarchical concepts, leading to redundant parameter usage and vanishing signals. We propose Equivariant Poincaré ResNets, combining hyperbolic geometry with discrete symmetry groups ($C_4$ and $D_4$). We identify critical roadblocks in applying Euclidean equivariance to hyperbolic space and propose geometrically safe tensor reshaping, left-regular permutations for hyperbolic group convolutions, and joint-orientation Poincaré Midpoint Batch normalisation. Empirically, embedding equivariance drastically reduces the optimisation space, accelerating convergence while accelerating convergence while respecting the boundary constraints of the Poincaré ball and preserving spatial-group equivariance.
