As someone who follows (or tries to) the cutting edge in machine learning and data analysis I have found that I often see really useful ideas on arxiv first whi

As someone who follows (or tries to) the cutting edge in machine learning and data analysis I have found that I often see really useful ideas on arxiv first which may latter appear in some form somewhere else (conference, journal). Conversely some more prestigious/general interest journals like science and nature tend to favor papers based on novelty in a way which often results in them highlighting an interesting problem domain but not a best-in-class method. Indeed they often reject follow up pieces that improve on the published method.