choklatewolfy

I recently stumbled upon something called the "Supersymmetric Artificial Neural Network" or (SANN) on reddit/machine learning, which the author is to discuss at Gordon Research's next String Theory Conference in June. With the recent proliferation of machine learning in the realm of physics (such as Katie Bouman et al's recent black hole photo powered by machine learning), I think this thread is appropriate here. That said, from my understanding, Supersymmetry (which the SANN model above utilizes at its core) emerged in Superstring theory which unraveled a theory of both bosons and fermions in the same symmetry group. (See this review including the SANN model above by a physics person named Mitchell Porter: Open Review : “Applications of Super-mathematics to Machine learning” ) In short, the difference between typical machine learning models, can be observed in a few mathematical notations: 1. Typical Deep learning model notation, as seen in the "Deep Learning Book" by Bengio et al. (Bengio is a winner of the Nobel Prize like Turing Award): \( \phi(x;\theta)^{\top}w \) (See 'Deep Learning Book' Chapter 6, 'Deep Feedforward Networks', page 166, item 3) 2. Supersymmetric Artificial Neural Network notation, where the extra theta signifies supersymmetric directions: \( \phi(x;\theta, \bar{{\theta}})^{\top}w \) Could the SANN yield a practical application of String Theory?