9
Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Embed Size (px)

DESCRIPTION

Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3. From utility to optimal policy. The utility function U(s) allows the agent to select the action that maximizes the expected utility of the subsequent state:. The Bellman equation. - PowerPoint PPT Presentation

Citation preview

Page 1: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Markov Decision Processes

AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Page 2: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

From utility to optimal policyโ€ข The utility function U(s) allows the agent to select

the action that maximizes the expected utility of the subsequent state:

Page 3: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

The Bellman equation

Now, if the utility of a state is the expected sum of discounted rewards from that point onwards, then there is a direct relationship between the utility of a state and the utility of its neighbors:

3

The utility of a state is the immediate reward for that state plus the expected discounted utility of the next state, assuming that the agent chooses the optimal action

๐‘ˆ (๐‘  )=๐‘… (๐‘  )+๐›พ max๐‘Žโˆˆ ๐ด(๐‘  )

โˆ‘๐‘  โ€ฒ๐‘ƒ (๐‘  โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ (๐‘  โ€ฒ) the Bellman

equation

Page 4: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

The Bellman equation

4

๐‘ˆ (๐‘  )=๐‘… (๐‘  )+๐›พ max๐‘Žโˆˆ ๐ด(๐‘  )

โˆ‘๐‘  โ€ฒ๐‘ƒ (๐‘  โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ (๐‘  โ€ฒ)

Page 5: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

The value iteration algorithm

For problem with n states, there are n Bellman equations, and n unknowns, however NOT linear

5

๐‘ˆ (๐‘  )=๐‘… (๐‘  )+๐›พ max๐‘Žโˆˆ ๐ด(๐‘  )

โˆ‘๐‘  โ€ฒ๐‘ƒ (๐‘  โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ (๐‘  โ€ฒ)

๐‘ˆ ๐‘–+1(๐‘  )โ†๐‘… (๐‘  )+๐›พ max๐‘Žโˆˆ๐ด (๐‘ )

โˆ‘๐‘ โ€ฒ๐‘ƒ (๐‘ โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ ๐‘–(๐‘ 

โ€ฒ )

Start with random U(s), update iteratively Guaranteed to converge to the unique solution

Demo: http://people.cs.ubc.ca/~poole/demos/mdp/vi.html

Page 6: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Policy iteration algorithm

It is possible to get an optimal policy even when the utility function estimate is inaccurate

If one action is clearly better than all others, then the exact magnitude of the utilities on the states involved need not be precise

6

Compute utilities of states

Compute optimal policy

Compute utilities of states for a given policy

Compute policy for the given state utilities

Value iteration Policy iteration

Page 7: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Policy iteration algorithm

7

๐œ‹โˆ—=argmax๐‘Žโˆˆ๐ด (๐‘  )

โˆ‘๐‘ โ€ฒ๐‘ƒ (๐‘ โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ (๐‘  โ€ฒ)

๐‘ˆ (๐‘  )=๐‘… (๐‘  )+๐›พ max๐‘Žโˆˆ ๐ด(๐‘  )

โˆ‘๐‘  โ€ฒ๐‘ƒ (๐‘  โ€ฒ|๐‘  ,๐‘Ž )๐‘ˆ (๐‘  โ€ฒ)

๐‘ˆ (๐‘  )=๐‘… (๐‘  )+๐›พโˆ‘๐‘  โ€ฒ๐‘ƒ ( ๐‘ โ€ฒ|๐‘  ,๐œ‹๐‘– (๐‘ ) )๐‘ˆ (๐‘  โ€ฒ) Linear

equation

Page 8: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Policy evaluation

n linear equations, n unknowns for problem with n states, solved in n cubic time, can also use iterative scheme 8

Page 9: Markov Decision Processes AIMA: 17.1, 17.2 (excluding 17.2.3), 17.3

Summary

Markov decision processes Utility of state sequence Utility of states Value iteration algorithm Policy iteration algorithm

9