AI · Theory

Markov Decision Processes

Theory

165/215

Markov Decision Processes

MDP: states, actions, transition P(s′|s,a), reward R, discount γ. Policy π(s) → action. Value functions estimate expected return. RL learns policies when you don’t know P. Robot navigation and game AI use this language.

Output — idea / do / check for Markov Decision Processes. Fill those three in the viva.

Exam tip

Name the MDP tuple.

Example

# Markov Decision Processes
print("idea :", "Markov Decision Processes")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Markov Decision Processes. Fill those three in the viva.

Short notes

DefS, A, P, R, γ. Policy + value.

AI · Theory

Markov Decision Processes

← All stacks

Theory

165/215

Markov Decision Processes

Output — idea / do / check for Markov Decision Processes. Fill those three in the viva.

Exam tip

Name the MDP tuple.

Example

# Markov Decision Processes
print("idea :", "Markov Decision Processes")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Markov Decision Processes. Fill those three in the viva.

Short notes

DefS, A, P, R, γ. Policy + value.

Markov Decision Processes

Example

Short notes

Questions

Markov Decision Processes

Example

Short notes

Questions