Whew. Here's the final post in the series about my attempts to connect an approach to computing scattering amplitudes in quantum field theory to the world of machine learning. The TL;DR is that YES positroid-based networks work and can even be applied to transformers, but simply aren't as useful as your standard ReLU networks. #Mathematics

https://harrison.totty.dev/p/chasing-geometric-structure-in-transformers