A library to perform deep learning (DL) using geometric algebra (GA) as fundamental primitives.
- Some sort of image classification (MNIST) to compare: Accuracy, Accuracy per # of parameters, Accuracy per dataset size, Adversarial robustness(?)
- Some sort of Capsule comparison (smallnorb): Accuracy, Accuracy per # of parameters, Accuracy per dataset size
- Replace einsum with sparse-tensor friendly version (geometric product tensor is quite sparse) and measure performance difference.