Tensor Field Network
A general equivariant graph neural network proposed in 2018 that uses spherical harmonic expansions to represent features, enabling the network to strictly satisfy SO(3) rotational equivariance. It is applicable to modeling three-dimensional physical systems such as molecules and materials.
Timeline (last 90 days)
IKNO matches Tensor Field Networks (TFN) in prediction accuracy while having only approximately 1/5.6 the number of parameters of TFN.
Tensor Field Networks (TFN) are a class of general equivariant graph neural networks proposed in 2018, which use spherical harmonic expansions to represent features, making the network strictly equivariant to SO(3) rotations.
All Facts (2)
IKNO matches Tensor Field Networks (TFN) in prediction accuracy while having only approximately 1/5.6 the number of parameters of TFN.
50%UnverifiedTensor Field Networks (TFN) are a class of general equivariant graph neural networks proposed in 2018, which use spherical harmonic expansions to represent features, making the network strictly equivariant to SO(3) rotations.
50%