@Glid yes, kinda. But sometimes we cannot build it. I m working on physics modelling and you can model up to certain degree. We cannot use NN because its not feasible search all results space. This is costly and useless since there are a lot of parameters. But also our physical (and logical) models can go up to some extent. However, how can you derive a function for image classicitaon? Yes its hyped but, I dont mean that. Its overfunded i think. There are so much to do for science and its other valuable topics. We do not know its consequences. We may hate all the benefits of overfunding ( which are good contributions to our life) after some year. And I m not sure how it ll continue like that without solving inefficiency, interpretability, inexplainability, insufficiency, inflexibility problems. Academia is %60 doing applications %35 improving these models %5 working on problems i mentioned. Of course percentages are wrong, but i think you got the point.
Thank you for an informative video. However I'm still left with the question: Are these methods worth it, and if so, when are they better than manual optimization?
Great survey of approaches! In this video, they reproduce some of the results from the ENAS paper: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-Ut97E9K-ai0.html
