I have been working on a computer vision problem with the use of cnns, but quite frustratingly I'm often in the situation of not knowing what to do to improve my results. It seems to me that most of the time I am mostly making random changes and experimenting in hope that this change will bring some improvement. I notice how this is different from non-AI software development where debugging can be performed by trying to pinpoint where exactly in the code lies an unexpected behaviour.

I wonder if there is a technique that could better orient the research effort.

  • 1
    $\begingroup$ There are definitely things that you can do. However, I doubt there is a top-level system which starts with "not good enough results" and works you through all possible deep learning scenarios to discover what the most likely issue is. So, I would suggest that as well as this question, you open a new question with more specific details of what is currently blocking your progress (maybe on Data Science SE, but heremay also be OK). That may produce some useful leads, even if it does not give you some kind of DL cheat sheet for troubleshooting $\endgroup$ Aug 2 at 8:30
  • $\begingroup$ I guess that's why machine deep learning is a research topic and not something any 10-year-old can do reliably. $\endgroup$
    – user253751
    Aug 2 at 16:40
  • $\begingroup$ If there is really one systematic & pretty reliable way to improve results and performance without human intervention, then we, the AI-Engineers, would lose our jobs. $\endgroup$
    – Yahya
    Aug 2 at 16:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.