Together let’s unlock the full potential of Geometric Algebra in Deep Learning.

Two years ago, upstride engaged in an exciting and challenging adventure: bringing Geometric Algebra (GA) into Neural Networks (NNs) to achieve better accuracy and compression than traditional methods . We are pioneering a new field of research and have already laid down the foundations of a very promising technology.

However, given the magnitude of the challenges ahead, and thanks to our progress so far, we believe that it is now time to share with the community and to collaborate with other AI labs and researchers . Here is why we believe our work matters — we hope that this could trigger the same spark in you and lead to a fruitful collaboration.

Linear algebra remains the default mode for Neural Networks

Linear algebra (LA) is the mathematical lingua franca across scientific disciplines. This is also the case for Machine Learning (ML) and Deep Learning (DL), where the entire theory can be described by matrices, vectors, and a norm induced by an inner product.

The promise of hypercomplex algebras

In the last couple of years, several research initiatives have shown that ML algorithms (from adaptive filters to NNs) can have their estimation…