EasyDeepLearn

What is the Markov property and why does it matter?

medium

Answer

  • P(st+1    st,  at,  st1,  ,  s0)  =  P(st+1    st,  at)P(s_{t + 1}\; \mid \;s_{t}, \;a_{t}, \;s_{t - 1}, \;, \;s_{0})\; = \;P(s_{t + 1}\; \mid \;s_{t}, \;a_{t}) — the future depends only on the current state, not the history.
  • Enables efficient RL: policy and value functions depend on state alone, not history.
  • When violated (partial observability), we use POMDPs, recurrent nets, or state-stacking (Atari frame stack) to recover a Markovian representation.
Check yourself — multiple choice
  • Random
  • P(st+1    st,  at)P(s_{t + 1}\; \mid \;s_{t}, \;a_{t}) is independent of history; violated → POMDP + RNN or state-stacking
  • Same as Bellman
  • Not real

Markov: future depends only on (s, a); POMDP + RNN / stack when violated.

#fundamentals#theory

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions