There is a good progression of effort you may want to put in: First, discuss your idea on reddit machine learning. It will be shot down quickly if there is a g

There is a good progression of effort you may want to put in: First, discuss your idea on reddit machine learning. It will be shot down quickly if there is a gaping flaw or something you overlooked. You can combine this with second, write a blogpost on what you got. Third, write a preprint, link to it on the blogpost so there is no risk of you not getting credit. Overall, the effort for each step increases significantly, you might be tempted to go straight for the preprint but let me tell you that this is most likely a waste of your time because you (very likely) don't know how to write with a certain ML flavour, how to present certain ideas, and how to design the right experiments or contextualise your arguments. If you write a good blogpost, you have a nice starting point to find someone to help you formalise your ideas.