EasyDeepLearn

Reparameterization trick in stochastic policies.

hard

Answer

  • For continuous stochastic policy π_θ(as)  =  N(μθ(s),  σθ(s)2)(a \mid s)\; = \;N({\mu}{\theta}(s), \;{\sigma}{\theta}(s)^{2}), sample a = μ + σ ⊙ ε with ε ~ N(0, I) → gradient flows through μ, σ directly.
  • Enables low-variance path-wise gradients rather than score-function REINFORCE-style gradients.
  • Used in SAC.
  • Also foundation of VAEs.
  • Doesn't work directly for discrete actions (use Gumbel-softmax as continuous relaxation).
Check yourself — multiple choice
  • Same as REINFORCE
  • a = μ_θ + σ_θ ⊙ ε with ε ~ N(0, I) → path-wise gradients through μ, σ, much lower variance than REINFORCE; SAC uses it; Gumbel-softmax for discrete
  • Random
  • Not real

Reparam: a = μ + σε; path-wise low-variance gradients; SAC / VAE.

#policy-methods#deep-rl#theory

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions