Lie Group Decompositions for Equivariant Neural Networks

Mircea Mironenco, Patrick Forré

Invariance and equivariance to geometrical transformations have proven to be very useful inductive biases when training (convolutional) neural network models, especially in the low…