When my network is performing regression (like DQN) it makes sense to use batchnorm in network when output of my network should vary from [0, 100000]?

one way to tackle it is to normalize output but output is used for ranking choices in RL system so scaling can lead to small numerical errors that can explode after upscaling...

  • $\begingroup$ Use boosting methods :) $\endgroup$
    – Aditya
    Mar 14, 2019 at 12:04
  • $\begingroup$ I don't see benefit of boosting methods there especially for RL problem where you have to develop performan algorithm for decision making $\endgroup$
    – quester
    Mar 21, 2019 at 14:16
  • $\begingroup$ Well I am sorry, little to no knowledge on RL $\endgroup$
    – Aditya
    Mar 21, 2019 at 18:01


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.