EasyDeepLearn

Random Network Distillation (RND) — mechanism.

hard

Answer

  • Fixed random target network ftarget(s)f_{\mathrm{target}}(s) initialized randomly + trainable predictor fpred(s)f_{\mathrm{pred}}(s).
  • Intrinsic reward = ||fpred(s)    ftarget(s)f_{\mathrm{pred}}(s)\; - \;f_{\mathrm{target}}(s)||².
  • Predictor learns quickly on frequently-visited states → low intrinsic reward.
  • Rare states have high error → high intrinsic reward.
  • Simple, cheap, effective — solved Montezuma's Revenge (Burda et al. 2018).
  • Foundation of modern exploration methods.
Check yourself — multiple choice
  • Same as count
  • Fixed random target NN + trainable predictor; intrinsic reward = ||pred - target||²; predictor learns fast on visited states → high error = novel state; solved Montezuma
  • Random
  • Not real

RND: predict-random-network error as novelty; solved Montezuma.

#exploration#deep-rl

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions