We have at the present two threads on canonical solutions to common neural network pitfalls, thanks to initiative of @Sycorax and @DeltaIV:
- What should I do when my neural network doesn't learn?
- What should I do when my neural network doesn't generalize well?
These threads are probably good candidates to become community wikis, so that it is possible (or even welcome) to expand on existing answers, creating a coherent list of tips and tricks and best practices etc., instead of posting semi-repetitive answers where one has to cherry pick to find the relevant part.
For example, I feel like I could improve some of the answers by adding extra references, listing a few additional methods in a relevant section of the existing answer etc. However, at the moment I consider doing so a bad practice, as it generally changes the original message of the answer author. On the other hand, posting these bits and pieces in my own answer would probably result in a one-or-two-liner at the bottom of the thread, not useful to anyone.
However, community wiki posts do not provide score needed for attaining golden tag-badges and our current consensus is that we would profit from having a user or two who would have the ultimate closing powers of a golden [neural-networks] tag badge. In that sense, going community wiki is undesirable.
Should we convert these threads to community wikis?