Geometry-aware training of factorized layers in tensor Tucker format | Read Paper on Bytez