Apologies in advance as I know relating a neural net with a DFA is a bit along the line of ‘babby’s first comp sci + ml conjecture’ but I’d really like to attempt a self publish along these lines as the thought has been bouncing around my head for forever. (Undergrad plus decent research experience for that level)

My hypothesis is essentially that at least a vanilla FF neural net can be shown to be equivalent mathematically or in computational power to abstract computing devices like DFAs, in my mind they’re both just state machines.

I wanted to ask if this was something worth exploring or attempting to make a paper out of as opposed to a blog post, I’m a lot more interested in the theory heavy side of things in ML research versus practice (which is ambitious for my math scored admittedly).