Flash EQ-Linear: Accelerating Equivariant Linear Layers via Group-wise Discrete Fourier Transform
Flash EQ-Linear is an exact acceleration algorithm for equivariant linear layers that reduces computational complexity from O(NDC) to O(NDC/T). It achieves this by exploiting the Fourier convolution theorem and conjugate symmetry of the real DFT. This results in a theoretical speedup of T/2, allowing equivariant networks to match or surpass non-equivariant models in accuracy, efficiency, and inference speed.