A Theory of Neural Computation with Clifford Algebras (2005) [pdf]

I'm not really aware about this application of Clifford algebra. I'll summarize it as the algebra that operated on sets of scalars, vectors, matrices, and up, so we can use both addition and multiplication between Clifford objects. Just my recollection.

+ I would assume then that keeping track of Clifford objects per neuron is not biological plausible.

+ I would then also assume that geometric manipulations are handled differently than through Clifford algebra.

Just saying because biology is mentioned in a few places in the thesis as motivation.

Additional information for Clifford Algebra can be found here:

https://slehar.wordpress.com/2014/03/18/clifford-algebra-a-v...

As I recall, the most interesting feature of these networks is that they do better when learning geometric features in the presence of noise. Without the added geometric structure baked into the network, the real neural nets overfit and 'learn the noise' to a much greater extent.