Yeah, I have the deepest respect for the people who do a lot of AI. I took Ng's online course and know that I don't know a ton but I'm more a "learn the basics,

Yeah, I have the deepest respect for the people who do a lot of AI. I took Ng's online course and know that I don't know a ton but I'm more a "learn the basics, read papers after that" kind of person so I just started trying to replicate papers after that a bit. Now my problem is mostly that I just don't have the energy to scour the research journals and books to put together datasets worth training with (for heterogeneous catalyst performance). Same as everyone I guess. The algorithms don't seem that bad, albeit obviously I want to run it past some friends that do it before claiming anything since I know I'll do it badly and just not know, so maybe in a perfect world it's just a matter of needing more data. I honestly think it will work because I with a human brain can look at trends in catalyst manufacturing methods and composition and guess at least reasonably locally what changes will do. The parameter space is just too big between atomic composition, preparation methods, support acidity, surface area, etc etc etc. I think honestly a computer could do it because it feels like a problem of fitting the data in my head enough to model it beyond super locally. But every time I try to improve a catalyst I succeed, which I think says more about how poorly optimized we are today than it says about my cleverness -- literally almost any random change is likely one that is a variable no one has tried yet. It's an awesomely open feeling field, and more exciting now that biomolecules are getting thrown into the mix from fermentation facilities.