CS 7180: Behavioral Modeling and Decisionmaking

Size: px
Start display at page:

Download "CS 7180: Behavioral Modeling and Decisionmaking"

Transcription

1 CS 7180: Behavioral Modeling and Decisionmaking in AI Markov Decision Processes for Complex Decisionmaking Prof. Amy Sliva October 17, 2012

2 Decisions are nondeterministic In many situations, behavior and decisions may have more than one possible outcome c Action failures e.g., gripper drops its load a c b grasp(c) a b Intended outcome Exogenous events e.g., road closed a b Unintended outcome One approach: Markov Decision Processes Still need to be able to plan and make decisions in such situations

3 Decision-making under uncertainty Decisions traditionally represented in decision trees Example decision problem: Should I have my party inside or outside? Exponential in number of decisions Limitations of decision trees What if there are many stages to consider? What if there are many different outcomes? Markov decision processes (MDPs) Framework for complex multi-stage decision problems under uncertainty More compact representation of problem with efficient solutions Further specialized influence diagram No explicit representation of qualitative dependencies in out dry wet dry wet Regret Relieved Perfect! Disaster

4 Stochastic systems of actions We have already learned about stochastic systems Markov chain/markov process Stochastic system: a triple Σ = (S, A, P) S = finite set of states A = finite set of actions P a (sʹ s) = probability of going to sʹ if we execute a in s sʹ S P a (sʹ s) = 1 Several different possible action representations e.g., Bayes networks (or influence diagrams), probabilistic state-space operators, etc. Same underlying semantics Fully specified transition model explicit enumeration of each P a (sʹ s)

5 MDPs are stochastic systems with utility Model the dynamics of the environment under different actions Formally defined as a 4-tuple MDP = (S, A, T, R) S = set of states A = set of actions T = transition model T(s, a, sʹ ) = P a (sʹ s) R = reward model R(s) and C(s,a) Environmental context Possible behaviors What states can result from an action? Reward for each state Cost for each state and action Markov assumption next state depends only on current state and action Full observability cannot predict exactly which state will result from an action, but once it is realized know what it is

6 Graphical MDP representation Nodes are possible states of the world E.g., Location of robot Arcs are actions E.g., Move to new location Each arc has associated transition probability Each state has associated reward or cost

7 First-order Markov dynamics and rewards First-Order Markov dynamics (history independence) P(s t+1 a t,s t,a t-1,s t-1,...,s 0 ) = P(s t+1 a t,s t ) Next state only depends on current state and current action First-Order Markov reward process P(r t a t,s t,a t-1,s t-1,...,s 0 ) = P(r t a t,s t ) Reward only depends on current state and action Assume reward is specified by a deterministic function R(s) Stationary dynamics and reward P(s t+1 a t,s t ) = P(s k+1 a k,s k ) for all t, k The world dynamics do not change over time

8 Planning actions in MDPs Robot r1 starts at location l1 State s1 in the diagram Objective is to get r1 to location l4 State s4 in the diagram Start Goal

9 Probability of initial states For every state s, there will be a probability P(s) that the system starts in s Sometimes assume a unique state s 0 such that the system always starts in s 0 In the example, s 0 = s1 P(s1) = 1 P(s) = 0 for all s s1 Start Goal

10 Planning actions in MDPs Robot r1 starts at location l1 State s1 in the diagram Objective is to get r1 to location l4 State s4 in the diagram Start Goal No classical plan (sequence of actions) can be a solution Why not?

11 Planning actions in MDPs Robot r1 starts at location l1 State s1 in the diagram Objective is to get r1 to location l4 State s4 in the diagram Start Goal No classical plan (sequence of actions) can be a solution No guarantee we ll be in a state where the next action is applicable π = move(r1,l1,l2), move(r1,l2,l3), move(r1,l3,l4)

12 Policies for choosing actions Policy is a function mapping states to actions Stationary policy π:s A π(s) is action to do at state s (regardless of time) set of state-action pairs Specifies a continuously reactive controller Nonstationary policy π:s x N A, where N is the non-negative integers π(s,t) is action to do at state s with t stages-to-go What if we want to keep acting indefinitely?

13 Example stationary policy Robot r1 starts at location l1 State s1 in the diagram Objective is to get r1 to location l4 State s4 in the diagram π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} Start Goal

14 Histories of previous states History is a sequence of system states h = s 0, s 1, s 2, s 3, s 4, Examples: h 0 = s1, s3, s1, s3, s1, h 1 = s1, s2, s3, s4, s4, h 2 = s1, s2, s5, s5, s5, h 3 = s1, s2, s5, s4, s4, h 4 = s1, s4, s4, s4, s4, h 5 = s1, s1, s4, s4, s4, h 6 = s1, s1, s1, s4, s4, h 7 = s1, s1, s1, s1, s1, Start Goal Each policy induces a probability distribution over histories If h = s 0, s 1, then P(h π) = P(s 0 ) i 0 P π(si) (s i+1 s i )

15 Example history distribution π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} Start Goal h 1 = s1, s2, s3, s4, s4, P(h 1 π 1 ) = = 0.8 h 2 = s1, s2, s5, s5 P(h 2 π 1 ) = = 0.2 P(h π 1 ) = 0 for all other h

16 Example history distribution π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} π 1 reaches the goal with probability 0.8 Start Goal h 1 = s1, s2, s3, s4, s4, P(h 1 π 1 ) = = 0.8 h 2 = s1, s2, s5, s5 P(h 2 π 1 ) = = 0.2 P(h π 1 ) = 0 for all other h

17 Example history distribution π 2 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, move(r1,l5,l4))} Start Goal h 1 = s1, s2, s3, s4, s4, P(h 1 π 2 ) = = 0.8 h 3 = s1, s2, s5, s4, s4, P(h 3 π 2 ) = = 0.2 P(h π 2 ) = 0 for all other h

18 Example history distribution π 2 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, move(r1,l5,l4))} π 2 reaches the goal with probability 1.0 Start Goal h 1 = s1, s2, s3, s4, s4, P(h 1 π 2 ) = = 0.8 h 3 = s1, s2, s5, s4, s4, P(h 3 π 2 ) = = 0.2 P(h π 2 ) = 0 for all other h

19 Example history distribution π 3 = {(s1, move(r1,l1,l4)), (s2, move(r1,l2,l1)), (s3, move(r1,l3,l4)), (s4, ), (s5, move(r1,l5,l4))} Start Goal h 4 = s1, s4, s4, s4, P(h 4 π 3 ) = = 0.5 h 5 = s1, s1, s4, s4, s4, P(h 5 π 3 ) = = 0.25 h 6 = s1, s1, s1, s4, s4, P(h 6 π 3 ) = = h 7 = s1, s1, s1, s1, s1, s1, P(h 7 π 3 ) = = 0

20 Example history distribution π 3 = {(s1, move(r1,l1,l4)), (s2, move(r1,l2,l1)), (s3, move(r1,l3,l4)), (s4, ), (s5, move(r1,l5,l4))} π 3 reaches the goal with probability 1.0 Start Goal h 4 = s1, s4, s4, s4, P(h 4 π 3 ) = = 0.5 h 5 = s1, s1, s4, s4, s4, P(h 5 π 3 ) = = 0.25 h 6 = s1, s1, s1, s4, s4, P(h 6 π 3 ) = = h 7 = s1, s1, s1, s1, s1, s1, P(h 7 π 3 ) = = 0

21 Determining the value of a policy How good is a policy π? How do we measure accumulated reward? Utility function V: S R associates value with each state (or each state and time for non-stationary π) V π (s) denotes value of policy at state s Depends on immediate reward, but also what you achieve subsequently by following π Optimal policy is no worse than any other policy at any state The goal of MDP reasoning is to compute an optimal policy (method depends on how we define utility)

22 Compute utility using reward function Numeric cost C(s,a) for each state s and action a Numeric reward R(s) for each state s r = 100 No explicit goals now Desirable states have high rewards r = 0 r = 0 Example: C(s,) = 0 at every state except s3 C(s,a) = 1 for each horizontal action C(s,a) = 100 for each vertical action R as shown r = 0 Start r = 100 Utility of a history/policy: If h = s 0, s 1,, then V(h π) = i 0 [R(s i ) C(s i,π(s i ))]

23 Example utility computation r = 100 π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} r = 0 h 1 = s1, s2, s3, s4, s4, P(h 1 h 2 = s1, s2, s5, s5 r = 0 Start r = 100 V(h 1 π 1 ) = [R(s1) C(s1,π 1 (s1))] + [R(s2) C(s2,π 1 (s2))] + [R(s3) C(s3,π 1 (s3))] + [R(s4) C(s4,π 1 (s4))] + [R(s4) C(s4,π 1 (s4))] + = [0 100] + [0 1] + [0 100] + [100 0] + [100 0] + = V(h 2 π 1 ) = [0 100] + [0 1] + [ 100 0] + [ 100 0] + [ 100 0] + = r = 0

24 Discounted utility In a long history of states, distant rewards/costs have less influence on the current decision Often need to use a discount factor γ, 0 γ 1 Discounted utility of a history V(h π) = i 0 γ i [R(s i ) C(s i,π(s i ))] Convergence is guaranteed if 0 γ < 1 Expected utility of a policy: E(π) = h P(h π) V(h π)

25 Discounted and expected utility r = 100 π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} r = 0 h 1 = s1, s2, s3, s4, s4, P(h 1 h 2 = s1, s2, s5, s5 r = 0 Start V(h 1 π 1 ) =.9 0 [0 100] [0 1] [0 100] [100 0] [100 0] + = V(h 2 π 1 ) =.9 0 [0 100] [0 1] [ 100 0] [ 100 0] + = E(π 1 ) = 0.8 V(h 1 π 1 ) V(h 2 π 1 ) = 0.8(547.9) + 0.2( 910.1) = r = 0 r = 100

26 Planning as optimization Consider special case with start state s 0 and all rewards are 0 Consider cost rather than utility the negative of what we had before Equations slightly simpler can generalize to the case of nonzero rewards Discounted cost of a history h and policy π: C(h π) = i 0 γ i C(s i, π(s i )) Expected cost of a policy π: E(π) = h P(h π) C(h π) A policy π is optimal if for every π', E(π) E(π') A policy π is everywhere optimal if for every s and every π', E π (s) E π' (s) where E π (s) is the expected utility if we start at s rather than s 0

27 Bellman s theorem If π is any policy, then for every s, s π(s) s 1 s 2 s n E π (s) = C(s, π(s)) + γ s S P π(s) (sʹ s) E π (sʹ ) Let Q π (s,a) be the expected cost in a state s if we start by executing the action a, and use the policy π from then onward Q π (s,a) = C(s,a) + γ sʹ S P a (sʹ s) E π (sʹ ) Bellman s theorem: Suppose π* is everywhere optimal. Then for every s, E π* (s) = min a A(s) Q π* (s,a)

28 Intuition behind Bellman s theorem Bellman s theorem: Suppose π* is everywhere optimal. Then for every s, E π* (s) = min a A(s) Q π* (s,a) If we use π* everywhere else, then the optimal actions at s is arg min a A(s) Q π* (s,a) If π* is optimal, then at each state it will pick one of those Otherwise we can construct a better policy by using an action in arg min a A(s) Q π* (s,a) instead of the action that π* uses From Bellman s theorem it follows that for all s, E π* (s) = min a A(s) {C(s,a) + γ s S P a (sʹ s) E π* (sʹ )}

29 Finite horizon decision-making First look at policy optimization over a finite horizon Assumes the agent has n time steps to live γ =1 To act optimally, should we use stationary or nonstationary policy? Put another way: If you had only one week to live would you act the same way as if you had fifty years to live?

30 Finite horizon problems Value (utility) depends on stage-to-go, hence so should policy Nonstationary π(s,k) V πk (s) is the k-stage-to-go value function for π Expected total reward/cost after executing π for k time steps V πk (s) = t = 0 to k C(s t, π(s t ))P(h t π) where C(s t, π(s t )) denotes the cost received at stage t and P(h t π) is the probability of the history seen until state t, given the policy π

31 Computing the finite-horizon value Markov property facilitates dynamic programming Use to compute V πk (s) Work backward from final stage to determine optimal policy V π0 (s) = C(s,π(s)) π(s,k) 0.7 V πk (s) = C(s,π(s)) + sʹ P π(s,k) (sʹ s) V π k-1 (sʹ ) Vk 0.3 Vk-1

32 Computing the finite-horizon value Markov property facilitates dynamic programming Use to compute V πk (s) Work backward from final stage to determine optimal policy V π0 (s) = C(s,π(s)) Vk π(s,k) V πk (s) = C(s,π(s)) + sʹ P π(s,k) (sʹ s) V π k-1 (sʹ ) Vk-1 Immediate cost Expected future cost with k-1 stages to go

33 Bellman backup Use Bellman s theorem to incrementally compute utility value Backup the values from V t to V t+1 using DP and choosing minimum path V t s1 a s2 V t+1 (s) s 0.4 s3 V t+1 (s) = min a A(s) {C(s,a) + 0.7V t (s1) + 0.3V t (s4), C(s,a) + 0.4V t (s2) + 0.6V t (s3) } Uses procedure called value iteration a s4

34 General case value iteration algorithm 1. Start with an arbitrary cost E 0 (s) for each s and a small ε > 0 2. For i = 1, 2, a) For every s in S and a in A, i. Q i (s,a) := C(s,a) + γ sʹ S P a (sʹ s) E i 1 (sʹ ) E i (s) = min a A(s) Q i (s,a) ii. iii. π i (s) = arg min a A(s) Q i (s,a) b) If max s S E i (s) E i 1 (s) < ε for every s then exit π i converges to π * after finitely many iterations, but how to tell it has converged? In general, E i Eπ i When π i doesn t change, E i may still change The changes in E i may make π i start changing again

35 General case value iteration algorithm 1. Start with an arbitrary cost E 0 (s) for each s and a small ε > 0 2. For i = 1, 2, a) For every s in S do i. For each a in A do Q (s,a) := C(s,a) + γ sʹ S P a (sʹ s) E i 1 (sʹ ) ii. E i (s) = min a A(s) Q(s,a) iii. π i (s) = arg min a A(s) Q(s,a) b) If max s S E i (s) E i 1 (s) < ε for every s then exit If E i changes by < ε and if ε is small enough, then π i will no longer change In this case π i has converged to π* How small is small enough?

36 Value iteration in finite-horizon case Use Markov property and Bellman backup to avoid enumerating all possibilities Value iteration Initialize to final stage cost: V 0 (s) = C(s) Compute V k (s) optimal k-stage-to-go value function V k (s) = min a A(s) {C(s,a) + s S P a (sʹ s) V k-1 (sʹ )} Derive optimal k-stage-to-go policy π*(s,k) = argmin a A(s) {C(s,a) + s S P a (sʹ s) V k-1 (sʹ )} Optimal value function is unique, but optimal policy is not Many policies can have same value

37 Finite-horizon value iteration example V 3 V 2 V 1 V 0 s1 a1 = s2 a2 = s s4 V 1 (s4) = min a A(s) {C(s4,a1) + 0.7V 0 (s1) + 0.3V 0 (s4), C(s4,a2) + 0.4V 0 (s2) + 0.6V 0 (s3) }

38 Finite-horizon value iteration example V 3 V 2 V 1 V 0 s1 a1 = s2 a2 = s s4 π*(s4,1) = argmin a1,a2 {C(s,a) + s S P a (sʹ s) V k-1 (sʹ )}

39 Complexity of finite-horizon value iteration Optimal solution to k-1 stage problem can be used without modification as part of optimal solution to k-stage problem Dynamic programming Because of finite horizon, policy nonstationary What is the computational complexity? T iterations At each iteration, each of n states computes expectation for A actions Each expectation takes O(n) time

40 Complexity of finite-horizon value iteration Optimal solution to k-1 stage problem can be used without modification as part of optimal solution to k-stage problem Dynamic programming Because of finite horizon, policy nonstationary What is the computational complexity? T iterations At each iteration, each of n states computes expectation for A actions Each expectation takes O(n) time Total time complexity: O(T A n2) Polynomial in number of states. Is this good?

41 Discounted infinite-horizon MDPs Defining value as total reward is problematic with infinite horizons Many or all policies have infinite expected reward Use our discounted utility to discount the future reward at each time step discount factor γ, 0 γ 1 Can restrict attention to stationary policies Discounted cost of a history h and policy π: C(h π) = i 0 γ i C(s i, π(s i )) Expected cost of a policy π: E(π) = h P(h π) C(h π) A policy π is optimal if for every π', E(π) E(π')

42 Value iteration in infinite horizon case Use value iteration and Bellman backup to compute optimal policies just like in finite-horizon case Now just include the discount factor in the computation Initialize to start state cost: E 0 (s) = arbitrary cost Compute V k (s) optimal value function E i (s) = min a A(s) {C(s,a) + γ s S P a (sʹ s) E i-1 (sʹ )} Will converge to the optimal value function as k gets large

43 Infinite horizon value iteration example Let a ij be the action that moves from s i to s j e.g., a 11 = and a 12 = move(r1,l1,l2) r = 0 00 Start with E 0 (s) = 0 for all s, and ε = 1 Start c = 0

44 Infinite horizon value iteration example For each s and a compute Q (s,a) := C(s,a) + γ sʹ S P a (sʹ s) E i 1 (sʹ ) Q(s1, a 11 ) = = 1 Q(s1, a 12 ) = = 100 Q(s1, a 14 ) = 1 +.9( ) = 1 Q(s2, a 21 ) = = 100 Q(s2, a 22 ) = = 1 Q(s2, a 23 ) = 1 +.9( ) = 1 Q(s3, a 32 ) = = 1 Q(s3, a 34 ) = = 100 Q(s4, a 41 ) = = 1 Q(s4, a 43 ) = = 1 Q(s4, a 44 ) = = 0 Q(s4, a 45 ) = = 100 Start 00 c = 0 00 r = 0 00 Q(s5, a 52 ) = = 1 Q(s5, a 54 ) = = 100 Q(s5, a 55 ) = = 100

45 Infinite horizon value iteration example For each s and a compute E 1 (s,a) and π 1 (s) E 1 (s1) = 1 E 1 (s2) = 1 E 1 (s3) = 1 E 1 (s4) = 0 E 1 (s5) = 1 π 1 (s1) = a 11 = π 1 (s2) = a 22 = π 1 (s3) = a 32 = move(r1,l3,l2) π 1 (s4) = a 44 = π 1 (s5) = a 52 = move(r1,l5,l2) Start 00 c = 0 00 r = 0 00

46 Infinite horizon value iteration example For each s and a compute E 1 (s,a) and π 1 (s) E 1 (s1) = 1 E 1 (s2) = 1 E 1 (s3) = 1 E 1 (s4) = 0 E 1 (s5) = 1 π 1 (s1) = a 11 = π 1 (s2) = a 22 = π 1 (s3) = a 32 = move(r1,l3,l2) π 1 (s4) = a 44 = π 1 (s5) = a 52 = move(r1,l5,l2) Start 00 c = 0 00 r = 0 What other actions could we have chosen? Is ε small enough? 00

47 Deciding how to act using MDPs Given an E i from value iteration that closely approximates E π*, what should we use as our policy? Use greedy policy π(s) = greedy[e i (s)] = argmin a A(s) Q(s,a) where Q(s,a) = C(s,a) + γ sʹ S P a (sʹ s) V(sʹ ) Note that the value of greedy policy may not be equal to E i Let E G be the value of the greedy policy? How close is E G to E π*?

48 Deciding how to act using MDPs Given an E i from value iteration that closely approximates E π*, what should we use as our policy? Use greedy policy π(s) = greedy[e i (s)] = argmin a A(s) Q(s,a) where Q(s,a) = C(s,a) + γ sʹ S P a (sʹ s) V(sʹ ) Greedy is not too far from optimal if E i close to E π* In particular, if E i within ε of E π*, then E G within 2εγ /1-γ of E π* Furthermore, there exists a finite ε s.t. greedy policy is optimal I.e., even if value estimate is off, greedy is optimal once it is close enough

49 Policy iteration for infinite-horizon planning Policy iteration is another way to find π * Suppose there are n states s 1,, s n Start with an arbitrary initial policy π 1 For i = 1, 2, Compute π i s expected costs by solving n equations with n unknowns n instances of the discounted expected cost equation For every s j, n E πi (s 1 ) = C(s,π i (s 1 ))+γ P k=1 πi (s 1 )(s k s 1 ) E πi (s k ) n E πi (s n ) = C(s,π i (s n ))+γ P πi (s n )(s k s n ) E πi (s k ) π i+1 (s j ) = argmin a A Q πi (s j, a) If π i+1 = π i then exit Converges in a finite number of iterations k=1 n = argmin a A C(s j, a)+γ P a (s k s j ) E πi (s k ) k=1

50 Policy iteration example Assume discount factor γ = 0.9 r = 0 00 s5 undesirable C(s5, ) = 100 Incentive to leave non-goal states: C(s1,) = 1 C(s2,) = 1 Start 00 c = 0 00

51 Policy iteration example Start with arbitrary policy π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} r = 0 00 Compute expected costs across all states Start c = 0

52 Policy iteration example Start with arbitrary policy π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} r = 0 00 Compute expected costs across all states Start c = 0

53 Policy iteration example Start with arbitrary policy π 1 = {(s1, move(r1,l1,l2)), (s2, move(r1,l2,l3)), (s3, move(r1,l3,l4)), (s4, ), (s5, )} r = 0 00 Compute expected costs across all states Start c = 0

54 Policy iteration example Compute expected costs across all states r = At each state s, let Start π 2 (s) = argmin a A(s) Q π (s,a) π 2 = {(s1, move(r1,l1,l4)), (s2, move(r1,l2,l1)), (s3, move(r1,l3,l4)), (s4, ), (s5, move(r1,l5,l4))} c = 0

55 Policy iteration vs. value iteration Policy iteration computes an entire policy in each iteration, and computes values based on that policy More work per iteration needs to solve a set of simultaneous equations Usually converges in a smaller number of iterations Value iteration computes new values in each iteration, and chooses a policy based on those values In general, the values are not the values that one would get from the chosen policy or any other policy Less work per iteration Usually takes more iterations to converge

56 Policy iteration vs. value iteration For both, iterations is polynomial in the number of states But the number of states is usually quite large Need to examine the entire state space in each iteration Thus, these algorithms can take huge amounts of time and space Use real-time dynamic programming and heuristics to improve efficiency

57 Real-time dynamic programming basics Explicitly specify goal states If s is a goal, then actions at s have no cost and produce no change For each state s, maintain value V(s) that gets updated as algorithm proceeds Initially V(s) = h(s), where h is a heuristic function Greedy policy π(s) = greedy[e i (s)] = argmin a A(s) Q(s,a) where Q(s,a) = C(s,a) + γ sʹ S P a (sʹ s) V(sʹ )

58 RTDP algorithm procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s'

59 RTDP algorithm procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' Forward search

60 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' Start c = 0

61 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' V = 0 Start 00 c = 0 00 Example: γ = 0.9 h(s) = 0 for all s

62 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' Example: γ = 0.9 h(s) = 0 for all s V = 0 Start Q = *0 = c = 0 00 Q = 1+.9(½*0+½*0) = 1 V = 0 Q = *0 = 100

63 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' V = 0 Start c = 0 V = 0 Example: γ = 0.9 h(s) = 0 for all s Q = 1+.9(½*0+½*0) = 1

64 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' V = 1 Start c = 0 V = 0 Example: γ = 0.9 h(s) = 0 for all s Q = 1+.9(½*0+½*0) = 1

65 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' V = 1 Start 00 c = 0 00 Example: γ = 0.9 h(s) = 0 for all s

66 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' Example: γ = 0.9 h(s) = 0 for all s V = 1 Start Q = *0 = c = 0 00 Q = 1+.9(½*1+½*0) = 1.45 V = 0 Q = *0 = 100

67 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' V = 1 Start c = 0 V = 0 Example: γ = 0.9 h(s) = 0 for all s Q = 1+.9(½*1+½*0) = 1.45

68 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) V = 0 r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' 00 V = 1.45 Start 00 c = 0 V = 0 Example: γ = 0.9 h(s) = 0 for all s Q = 1+.9(½*1+½*0) = 1.45

69 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' 00 V = 1.45 Start c = 0 00 Example: γ = 0.9 h(s) = 0 for all s

70 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' 00 V = 1.45 Start c = 0 00 Example: γ = 0.9 h(s) = 0 for all s

71 Example of RTDP procedure RTDP(s) 1. Loop until termination condition a) RTDP-trial(s) r = 0 00 procedure RTDP-trial(s) 1. While s is not a goal state a) a := arg min a A(s) Q(s,a) b) V(s) := Q(s,a) c) Randomly pick s' with probability P a (sʹ s) d) s := s' 00 V = 1.45 Start c = 0 00 Example: γ = 0.9 h(s) = 0 for all s

72 Performance of RTDP In practice, it can solve much larger problems than policy iteration and value iteration Won t always find an optimal solution, won t always terminate If h does not overestimate, and if a goal is reachable (with positive probability) at every state then it will terminate h should be an admissible heuristic If in addition to the above, there is a positive-probability path between every pair of states Then it will find an optimal solution

Chapter 16 Planning Based on Markov Decision Processes

Chapter 16 Planning Based on Markov Decision Processes Lecture slides for Automated Planning: Theory and Practice Chapter 16 Planning Based on Markov Decision Processes Dana S. Nau University of Maryland 12:48 PM February 29, 2012 1 Motivation c a b Until

More information

Some AI Planning Problems

Some AI Planning Problems Course Logistics CS533: Intelligent Agents and Decision Making M, W, F: 1:00 1:50 Instructor: Alan Fern (KEC2071) Office hours: by appointment (see me after class or send email) Emailing me: include CS533

More information

CSE 573. Markov Decision Processes: Heuristic Search & Real-Time Dynamic Programming. Slides adapted from Andrey Kolobov and Mausam

CSE 573. Markov Decision Processes: Heuristic Search & Real-Time Dynamic Programming. Slides adapted from Andrey Kolobov and Mausam CSE 573 Markov Decision Processes: Heuristic Search & Real-Time Dynamic Programming Slides adapted from Andrey Kolobov and Mausam 1 Stochastic Shortest-Path MDPs: Motivation Assume the agent pays cost

More information

Markov Decision Processes Chapter 17. Mausam

Markov Decision Processes Chapter 17. Mausam Markov Decision Processes Chapter 17 Mausam Planning Agent Static vs. Dynamic Fully vs. Partially Observable Environment What action next? Deterministic vs. Stochastic Perfect vs. Noisy Instantaneous vs.

More information

Reinforcement Learning and Control

Reinforcement Learning and Control CS9 Lecture notes Andrew Ng Part XIII Reinforcement Learning and Control We now begin our study of reinforcement learning and adaptive control. In supervised learning, we saw algorithms that tried to make

More information

MARKOV DECISION PROCESSES (MDP) AND REINFORCEMENT LEARNING (RL) Versione originale delle slide fornita dal Prof. Francesco Lo Presti

MARKOV DECISION PROCESSES (MDP) AND REINFORCEMENT LEARNING (RL) Versione originale delle slide fornita dal Prof. Francesco Lo Presti 1 MARKOV DECISION PROCESSES (MDP) AND REINFORCEMENT LEARNING (RL) Versione originale delle slide fornita dal Prof. Francesco Lo Presti Historical background 2 Original motivation: animal learning Early

More information

2534 Lecture 4: Sequential Decisions and Markov Decision Processes

2534 Lecture 4: Sequential Decisions and Markov Decision Processes 2534 Lecture 4: Sequential Decisions and Markov Decision Processes Briefly: preference elicitation (last week s readings) Utility Elicitation as a Classification Problem. Chajewska, U., L. Getoor, J. Norman,Y.

More information

Markov decision processes

Markov decision processes CS 2740 Knowledge representation Lecture 24 Markov decision processes Milos Hauskrecht milos@cs.pitt.edu 5329 Sennott Square Administrative announcements Final exam: Monday, December 8, 2008 In-class Only

More information

Markov Decision Processes Chapter 17. Mausam

Markov Decision Processes Chapter 17. Mausam Markov Decision Processes Chapter 17 Mausam Planning Agent Static vs. Dynamic Fully vs. Partially Observable Environment What action next? Deterministic vs. Stochastic Perfect vs. Noisy Instantaneous vs.

More information

Reinforcement Learning. Introduction

Reinforcement Learning. Introduction Reinforcement Learning Introduction Reinforcement Learning Agent interacts and learns from a stochastic environment Science of sequential decision making Many faces of reinforcement learning Optimal control

More information

Course 16:198:520: Introduction To Artificial Intelligence Lecture 13. Decision Making. Abdeslam Boularias. Wednesday, December 7, 2016

Course 16:198:520: Introduction To Artificial Intelligence Lecture 13. Decision Making. Abdeslam Boularias. Wednesday, December 7, 2016 Course 16:198:520: Introduction To Artificial Intelligence Lecture 13 Decision Making Abdeslam Boularias Wednesday, December 7, 2016 1 / 45 Overview We consider probabilistic temporal models where the

More information

Markov Decision Processes (and a small amount of reinforcement learning)

Markov Decision Processes (and a small amount of reinforcement learning) Markov Decision Processes (and a small amount of reinforcement learning) Slides adapted from: Brian Williams, MIT Manuela Veloso, Andrew Moore, Reid Simmons, & Tom Mitchell, CMU Nicholas Roy 16.4/13 Session

More information

Chapter 17 Planning Based on Model Checking

Chapter 17 Planning Based on Model Checking Lecture slides for Automated Planning: Theory and Practice Chapter 17 Planning Based on Model Checking Dana S. Nau CMSC 7, AI Planning University of Maryland, Spring 008 1 Motivation c Actions with multiple

More information

CAP Plan, Activity, and Intent Recognition

CAP Plan, Activity, and Intent Recognition CAP6938-02 Plan, Activity, and Intent Recognition Lecture 10: Sequential Decision-Making Under Uncertainty (part 1) MDPs and POMDPs Instructor: Dr. Gita Sukthankar Email: gitars@eecs.ucf.edu SP2-1 Reminder

More information

Chapter 17 Planning Based on Model Checking

Chapter 17 Planning Based on Model Checking Lecture slides for Automated Planning: Theory and Practice Chapter 17 Planning Based on Model Checking Dana S. Nau University of Maryland 1:19 PM February 9, 01 1 Motivation c a b Actions with multiple

More information

Motivation for introducing probabilities

Motivation for introducing probabilities for introducing probabilities Reaching the goals is often not sufficient: it is important that the expected costs do not outweigh the benefit of reaching the goals. 1 Objective: maximize benefits - costs.

More information

Markov Decision Processes

Markov Decision Processes Markov Decision Processes Noel Welsh 11 November 2010 Noel Welsh () Markov Decision Processes 11 November 2010 1 / 30 Annoucements Applicant visitor day seeks robot demonstrators for exciting half hour

More information

CS788 Dialogue Management Systems Lecture #2: Markov Decision Processes

CS788 Dialogue Management Systems Lecture #2: Markov Decision Processes CS788 Dialogue Management Systems Lecture #2: Markov Decision Processes Kee-Eung Kim KAIST EECS Department Computer Science Division Markov Decision Processes (MDPs) A popular model for sequential decision

More information

CMU Lecture 11: Markov Decision Processes II. Teacher: Gianni A. Di Caro

CMU Lecture 11: Markov Decision Processes II. Teacher: Gianni A. Di Caro CMU 15-781 Lecture 11: Markov Decision Processes II Teacher: Gianni A. Di Caro RECAP: DEFINING MDPS Markov decision processes: o Set of states S o Start state s 0 o Set of actions A o Transitions P(s s,a)

More information

Today s Outline. Recap: MDPs. Bellman Equations. Q-Value Iteration. Bellman Backup 5/7/2012. CSE 473: Artificial Intelligence Reinforcement Learning

Today s Outline. Recap: MDPs. Bellman Equations. Q-Value Iteration. Bellman Backup 5/7/2012. CSE 473: Artificial Intelligence Reinforcement Learning CSE 473: Artificial Intelligence Reinforcement Learning Dan Weld Today s Outline Reinforcement Learning Q-value iteration Q-learning Exploration / exploitation Linear function approximation Many slides

More information

Introduction to Reinforcement Learning

Introduction to Reinforcement Learning CSCI-699: Advanced Topics in Deep Learning 01/16/2019 Nitin Kamra Spring 2019 Introduction to Reinforcement Learning 1 What is Reinforcement Learning? So far we have seen unsupervised and supervised learning.

More information

Partially Observable Markov Decision Processes (POMDPs)

Partially Observable Markov Decision Processes (POMDPs) Partially Observable Markov Decision Processes (POMDPs) Geoff Hollinger Sequential Decision Making in Robotics Spring, 2011 *Some media from Reid Simmons, Trey Smith, Tony Cassandra, Michael Littman, and

More information

Decision Theory: Markov Decision Processes

Decision Theory: Markov Decision Processes Decision Theory: Markov Decision Processes CPSC 322 Lecture 33 March 31, 2006 Textbook 12.5 Decision Theory: Markov Decision Processes CPSC 322 Lecture 33, Slide 1 Lecture Overview Recap Rewards and Policies

More information

Planning in Markov Decision Processes

Planning in Markov Decision Processes Carnegie Mellon School of Computer Science Deep Reinforcement Learning and Control Planning in Markov Decision Processes Lecture 3, CMU 10703 Katerina Fragkiadaki Markov Decision Process (MDP) A Markov

More information

CS 188 Introduction to Fall 2007 Artificial Intelligence Midterm

CS 188 Introduction to Fall 2007 Artificial Intelligence Midterm NAME: SID#: Login: Sec: 1 CS 188 Introduction to Fall 2007 Artificial Intelligence Midterm You have 80 minutes. The exam is closed book, closed notes except a one-page crib sheet, basic calculators only.

More information

CMU Lecture 12: Reinforcement Learning. Teacher: Gianni A. Di Caro

CMU Lecture 12: Reinforcement Learning. Teacher: Gianni A. Di Caro CMU 15-781 Lecture 12: Reinforcement Learning Teacher: Gianni A. Di Caro REINFORCEMENT LEARNING Transition Model? State Action Reward model? Agent Goal: Maximize expected sum of future rewards 2 MDP PLANNING

More information

Administration. CSCI567 Machine Learning (Fall 2018) Outline. Outline. HW5 is available, due on 11/18. Practice final will also be available soon.

Administration. CSCI567 Machine Learning (Fall 2018) Outline. Outline. HW5 is available, due on 11/18. Practice final will also be available soon. Administration CSCI567 Machine Learning Fall 2018 Prof. Haipeng Luo U of Southern California Nov 7, 2018 HW5 is available, due on 11/18. Practice final will also be available soon. Remaining weeks: 11/14,

More information

Introduction to Reinforcement Learning. CMPT 882 Mar. 18

Introduction to Reinforcement Learning. CMPT 882 Mar. 18 Introduction to Reinforcement Learning CMPT 882 Mar. 18 Outline for the week Basic ideas in RL Value functions and value iteration Policy evaluation and policy improvement Model-free RL Monte-Carlo and

More information

Markov Decision Processes and Solving Finite Problems. February 8, 2017

Markov Decision Processes and Solving Finite Problems. February 8, 2017 Markov Decision Processes and Solving Finite Problems February 8, 2017 Overview of Upcoming Lectures Feb 8: Markov decision processes, value iteration, policy iteration Feb 13: Policy gradients Feb 15:

More information

Reinforcement learning an introduction

Reinforcement learning an introduction Reinforcement learning an introduction Prof. Dr. Ann Nowé Computational Modeling Group AIlab ai.vub.ac.be November 2013 Reinforcement Learning What is it? Learning from interaction Learning about, from,

More information

Christopher Watkins and Peter Dayan. Noga Zaslavsky. The Hebrew University of Jerusalem Advanced Seminar in Deep Learning (67679) November 1, 2015

Christopher Watkins and Peter Dayan. Noga Zaslavsky. The Hebrew University of Jerusalem Advanced Seminar in Deep Learning (67679) November 1, 2015 Q-Learning Christopher Watkins and Peter Dayan Noga Zaslavsky The Hebrew University of Jerusalem Advanced Seminar in Deep Learning (67679) November 1, 2015 Noga Zaslavsky Q-Learning (Watkins & Dayan, 1992)

More information

, and rewards and transition matrices as shown below:

, and rewards and transition matrices as shown below: CSE 50a. Assignment 7 Out: Tue Nov Due: Thu Dec Reading: Sutton & Barto, Chapters -. 7. Policy improvement Consider the Markov decision process (MDP) with two states s {0, }, two actions a {0, }, discount

More information

Reinforcement Learning. Yishay Mansour Tel-Aviv University

Reinforcement Learning. Yishay Mansour Tel-Aviv University Reinforcement Learning Yishay Mansour Tel-Aviv University 1 Reinforcement Learning: Course Information Classes: Wednesday Lecture 10-13 Yishay Mansour Recitations:14-15/15-16 Eliya Nachmani Adam Polyak

More information

A Gentle Introduction to Reinforcement Learning

A Gentle Introduction to Reinforcement Learning A Gentle Introduction to Reinforcement Learning Alexander Jung 2018 1 Introduction and Motivation Consider the cleaning robot Rumba which has to clean the office room B329. In order to keep things simple,

More information

Distributed Optimization. Song Chong EE, KAIST

Distributed Optimization. Song Chong EE, KAIST Distributed Optimization Song Chong EE, KAIST songchong@kaist.edu Dynamic Programming for Path Planning A path-planning problem consists of a weighted directed graph with a set of n nodes N, directed links

More information

16.4 Multiattribute Utility Functions

16.4 Multiattribute Utility Functions 285 Normalized utilities The scale of utilities reaches from the best possible prize u to the worst possible catastrophe u Normalized utilities use a scale with u = 0 and u = 1 Utilities of intermediate

More information

Probabilistic Planning. George Konidaris

Probabilistic Planning. George Konidaris Probabilistic Planning George Konidaris gdk@cs.brown.edu Fall 2017 The Planning Problem Finding a sequence of actions to achieve some goal. Plans It s great when a plan just works but the world doesn t

More information

Lecture 25: Learning 4. Victor R. Lesser. CMPSCI 683 Fall 2010

Lecture 25: Learning 4. Victor R. Lesser. CMPSCI 683 Fall 2010 Lecture 25: Learning 4 Victor R. Lesser CMPSCI 683 Fall 2010 Final Exam Information Final EXAM on Th 12/16 at 4:00pm in Lederle Grad Res Ctr Rm A301 2 Hours but obviously you can leave early! Open Book

More information

Heuristic Search Algorithms

Heuristic Search Algorithms CHAPTER 4 Heuristic Search Algorithms 59 4.1 HEURISTIC SEARCH AND SSP MDPS The methods we explored in the previous chapter have a serious practical drawback the amount of memory they require is proportional

More information

Preference Elicitation for Sequential Decision Problems

Preference Elicitation for Sequential Decision Problems Preference Elicitation for Sequential Decision Problems Kevin Regan University of Toronto Introduction 2 Motivation Focus: Computational approaches to sequential decision making under uncertainty These

More information

An Introduction to Markov Decision Processes. MDP Tutorial - 1

An Introduction to Markov Decision Processes. MDP Tutorial - 1 An Introduction to Markov Decision Processes Bob Givan Purdue University Ron Parr Duke University MDP Tutorial - 1 Outline Markov Decision Processes defined (Bob) Objective functions Policies Finding Optimal

More information

Final Exam December 12, 2017

Final Exam December 12, 2017 Introduction to Artificial Intelligence CSE 473, Autumn 2017 Dieter Fox Final Exam December 12, 2017 Directions This exam has 7 problems with 111 points shown in the table below, and you have 110 minutes

More information

Internet Monetization

Internet Monetization Internet Monetization March May, 2013 Discrete time Finite A decision process (MDP) is reward process with decisions. It models an environment in which all states are and time is divided into stages. Definition

More information

Prof. Dr. Ann Nowé. Artificial Intelligence Lab ai.vub.ac.be

Prof. Dr. Ann Nowé. Artificial Intelligence Lab ai.vub.ac.be REINFORCEMENT LEARNING AN INTRODUCTION Prof. Dr. Ann Nowé Artificial Intelligence Lab ai.vub.ac.be REINFORCEMENT LEARNING WHAT IS IT? What is it? Learning from interaction Learning about, from, and while

More information

Lecture notes for Analysis of Algorithms : Markov decision processes

Lecture notes for Analysis of Algorithms : Markov decision processes Lecture notes for Analysis of Algorithms : Markov decision processes Lecturer: Thomas Dueholm Hansen June 6, 013 Abstract We give an introduction to infinite-horizon Markov decision processes (MDPs) with

More information

Artificial Intelligence

Artificial Intelligence Artificial Intelligence Dynamic Programming Marc Toussaint University of Stuttgart Winter 2018/19 Motivation: So far we focussed on tree search-like solvers for decision problems. There is a second important

More information

CSE250A Fall 12: Discussion Week 9

CSE250A Fall 12: Discussion Week 9 CSE250A Fall 12: Discussion Week 9 Aditya Menon (akmenon@ucsd.edu) December 4, 2012 1 Schedule for today Recap of Markov Decision Processes. Examples: slot machines and maze traversal. Planning and learning.

More information

The exam is closed book, closed calculator, and closed notes except your one-page crib sheet.

The exam is closed book, closed calculator, and closed notes except your one-page crib sheet. CS 188 Spring 2017 Introduction to Artificial Intelligence Midterm V2 You have approximately 80 minutes. The exam is closed book, closed calculator, and closed notes except your one-page crib sheet. Mark

More information

An Adaptive Clustering Method for Model-free Reinforcement Learning

An Adaptive Clustering Method for Model-free Reinforcement Learning An Adaptive Clustering Method for Model-free Reinforcement Learning Andreas Matt and Georg Regensburger Institute of Mathematics University of Innsbruck, Austria {andreas.matt, georg.regensburger}@uibk.ac.at

More information

1 MDP Value Iteration Algorithm

1 MDP Value Iteration Algorithm CS 0. - Active Learning Problem Set Handed out: 4 Jan 009 Due: 9 Jan 009 MDP Value Iteration Algorithm. Implement the value iteration algorithm given in the lecture. That is, solve Bellman s equation using

More information

Prioritized Sweeping Converges to the Optimal Value Function

Prioritized Sweeping Converges to the Optimal Value Function Technical Report DCS-TR-631 Prioritized Sweeping Converges to the Optimal Value Function Lihong Li and Michael L. Littman {lihong,mlittman}@cs.rutgers.edu RL 3 Laboratory Department of Computer Science

More information

CS 4100 // artificial intelligence. Recap/midterm review!

CS 4100 // artificial intelligence. Recap/midterm review! CS 4100 // artificial intelligence instructor: byron wallace Recap/midterm review! Attribution: many of these slides are modified versions of those distributed with the UC Berkeley CS188 materials Thanks

More information

Machine Learning I Reinforcement Learning

Machine Learning I Reinforcement Learning Machine Learning I Reinforcement Learning Thomas Rückstieß Technische Universität München December 17/18, 2009 Literature Book: Reinforcement Learning: An Introduction Sutton & Barto (free online version:

More information

Reinforcement Learning II

Reinforcement Learning II Reinforcement Learning II Andrea Bonarini Artificial Intelligence and Robotics Lab Department of Electronics and Information Politecnico di Milano E-mail: bonarini@elet.polimi.it URL:http://www.dei.polimi.it/people/bonarini

More information

Reinforcement Learning. Donglin Zeng, Department of Biostatistics, University of North Carolina

Reinforcement Learning. Donglin Zeng, Department of Biostatistics, University of North Carolina Reinforcement Learning Introduction Introduction Unsupervised learning has no outcome (no feedback). Supervised learning has outcome so we know what to predict. Reinforcement learning is in between it

More information

Reinforcement Learning

Reinforcement Learning CS7/CS7 Fall 005 Supervised Learning: Training examples: (x,y) Direct feedback y for each input x Sequence of decisions with eventual feedback No teacher that critiques individual actions Learn to act

More information

Reinforcement Learning

Reinforcement Learning 1 Reinforcement Learning Chris Watkins Department of Computer Science Royal Holloway, University of London July 27, 2015 2 Plan 1 Why reinforcement learning? Where does this theory come from? Markov decision

More information

Final Exam December 12, 2017

Final Exam December 12, 2017 Introduction to Artificial Intelligence CSE 473, Autumn 2017 Dieter Fox Final Exam December 12, 2017 Directions This exam has 7 problems with 111 points shown in the table below, and you have 110 minutes

More information

Reinforcement Learning. Machine Learning, Fall 2010

Reinforcement Learning. Machine Learning, Fall 2010 Reinforcement Learning Machine Learning, Fall 2010 1 Administrativia This week: finish RL, most likely start graphical models LA2: due on Thursday LA3: comes out on Thursday TA Office hours: Today 1:30-2:30

More information

Grundlagen der Künstlichen Intelligenz

Grundlagen der Künstlichen Intelligenz Grundlagen der Künstlichen Intelligenz Reinforcement learning Daniel Hennes 4.12.2017 (WS 2017/18) University Stuttgart - IPVS - Machine Learning & Robotics 1 Today Reinforcement learning Model based and

More information

Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm

Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm Michail G. Lagoudakis Department of Computer Science Duke University Durham, NC 2778 mgl@cs.duke.edu

More information

A Decentralized Approach to Multi-agent Planning in the Presence of Constraints and Uncertainty

A Decentralized Approach to Multi-agent Planning in the Presence of Constraints and Uncertainty 2011 IEEE International Conference on Robotics and Automation Shanghai International Conference Center May 9-13, 2011, Shanghai, China A Decentralized Approach to Multi-agent Planning in the Presence of

More information

Reinforcement Learning: An Introduction

Reinforcement Learning: An Introduction Introduction Betreuer: Freek Stulp Hauptseminar Intelligente Autonome Systeme (WiSe 04/05) Forschungs- und Lehreinheit Informatik IX Technische Universität München November 24, 2004 Introduction What is

More information

Decision Theory: Q-Learning

Decision Theory: Q-Learning Decision Theory: Q-Learning CPSC 322 Decision Theory 5 Textbook 12.5 Decision Theory: Q-Learning CPSC 322 Decision Theory 5, Slide 1 Lecture Overview 1 Recap 2 Asynchronous Value Iteration 3 Q-Learning

More information

MDP Preliminaries. Nan Jiang. February 10, 2019

MDP Preliminaries. Nan Jiang. February 10, 2019 MDP Preliminaries Nan Jiang February 10, 2019 1 Markov Decision Processes In reinforcement learning, the interactions between the agent and the environment are often described by a Markov Decision Process

More information

Outline. CSE 573: Artificial Intelligence Autumn Agent. Partial Observability. Markov Decision Process (MDP) 10/31/2012

Outline. CSE 573: Artificial Intelligence Autumn Agent. Partial Observability. Markov Decision Process (MDP) 10/31/2012 CSE 573: Artificial Intelligence Autumn 2012 Reasoning about Uncertainty & Hidden Markov Models Daniel Weld Many slides adapted from Dan Klein, Stuart Russell, Andrew Moore & Luke Zettlemoyer 1 Outline

More information

Machine Learning. Reinforcement learning. Hamid Beigy. Sharif University of Technology. Fall 1396

Machine Learning. Reinforcement learning. Hamid Beigy. Sharif University of Technology. Fall 1396 Machine Learning Reinforcement learning Hamid Beigy Sharif University of Technology Fall 1396 Hamid Beigy (Sharif University of Technology) Machine Learning Fall 1396 1 / 32 Table of contents 1 Introduction

More information

Markov Decision Processes Infinite Horizon Problems

Markov Decision Processes Infinite Horizon Problems Markov Decision Processes Infinite Horizon Problems Alan Fern * * Based in part on slides by Craig Boutilier and Daniel Weld 1 What is a solution to an MDP? MDP Planning Problem: Input: an MDP (S,A,R,T)

More information

Hidden Markov Models (HMM) and Support Vector Machine (SVM)

Hidden Markov Models (HMM) and Support Vector Machine (SVM) Hidden Markov Models (HMM) and Support Vector Machine (SVM) Professor Joongheon Kim School of Computer Science and Engineering, Chung-Ang University, Seoul, Republic of Korea 1 Hidden Markov Models (HMM)

More information

Multiagent Value Iteration in Markov Games

Multiagent Value Iteration in Markov Games Multiagent Value Iteration in Markov Games Amy Greenwald Brown University with Michael Littman and Martin Zinkevich Stony Brook Game Theory Festival July 21, 2005 Agenda Theorem Value iteration converges

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Markov decision process & Dynamic programming Evaluative feedback, value function, Bellman equation, optimality, Markov property, Markov decision process, dynamic programming, value

More information

This question has three parts, each of which can be answered concisely, but be prepared to explain and justify your concise answer.

This question has three parts, each of which can be answered concisely, but be prepared to explain and justify your concise answer. This question has three parts, each of which can be answered concisely, but be prepared to explain and justify your concise answer. 1. Suppose you have a policy and its action-value function, q, then you

More information

Basics of reinforcement learning

Basics of reinforcement learning Basics of reinforcement learning Lucian Buşoniu TMLSS, 20 July 2018 Main idea of reinforcement learning (RL) Learn a sequential decision policy to optimize the cumulative performance of an unknown system

More information

PART A and ONE question from PART B; or ONE question from PART A and TWO questions from PART B.

PART A and ONE question from PART B; or ONE question from PART A and TWO questions from PART B. Advanced Topics in Machine Learning, GI13, 2010/11 Advanced Topics in Machine Learning, GI13, 2010/11 Answer any THREE questions. Each question is worth 20 marks. Use separate answer books Answer any THREE

More information

Reinforcement Learning Active Learning

Reinforcement Learning Active Learning Reinforcement Learning Active Learning Alan Fern * Based in part on slides by Daniel Weld 1 Active Reinforcement Learning So far, we ve assumed agent has a policy We just learned how good it is Now, suppose

More information

Chapter 3: The Reinforcement Learning Problem

Chapter 3: The Reinforcement Learning Problem Chapter 3: The Reinforcement Learning Problem Objectives of this chapter: describe the RL problem we will be studying for the remainder of the course present idealized form of the RL problem for which

More information

Real Time Value Iteration and the State-Action Value Function

Real Time Value Iteration and the State-Action Value Function MS&E338 Reinforcement Learning Lecture 3-4/9/18 Real Time Value Iteration and the State-Action Value Function Lecturer: Ben Van Roy Scribe: Apoorva Sharma and Tong Mu 1 Review Last time we left off discussing

More information

Artificial Intelligence & Sequential Decision Problems

Artificial Intelligence & Sequential Decision Problems Artificial Intelligence & Sequential Decision Problems (CIV6540 - Machine Learning for Civil Engineers) Professor: James-A. Goulet Département des génies civil, géologique et des mines Chapter 15 Goulet

More information

MS&E338 Reinforcement Learning Lecture 1 - April 2, Introduction

MS&E338 Reinforcement Learning Lecture 1 - April 2, Introduction MS&E338 Reinforcement Learning Lecture 1 - April 2, 2018 Introduction Lecturer: Ben Van Roy Scribe: Gabriel Maher 1 Reinforcement Learning Introduction In reinforcement learning (RL) we consider an agent

More information

CS 570: Machine Learning Seminar. Fall 2016

CS 570: Machine Learning Seminar. Fall 2016 CS 570: Machine Learning Seminar Fall 2016 Class Information Class web page: http://web.cecs.pdx.edu/~mm/mlseminar2016-2017/fall2016/ Class mailing list: cs570@cs.pdx.edu My office hours: T,Th, 2-3pm or

More information

CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions

CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions 1. Neural Networks: a. There are 2 2n distinct Boolean functions over n inputs. Thus there are 16 distinct Boolean functions

More information

CS 7180: Behavioral Modeling and Decision- making in AI

CS 7180: Behavioral Modeling and Decision- making in AI CS 7180: Behavioral Modeling and Decision- making in AI Learning Probabilistic Graphical Models Prof. Amy Sliva October 31, 2012 Hidden Markov model Stochastic system represented by three matrices N =

More information

Lecture 3: The Reinforcement Learning Problem

Lecture 3: The Reinforcement Learning Problem Lecture 3: The Reinforcement Learning Problem Objectives of this lecture: describe the RL problem we will be studying for the remainder of the course present idealized form of the RL problem for which

More information

Reinforcement Learning. George Konidaris

Reinforcement Learning. George Konidaris Reinforcement Learning George Konidaris gdk@cs.brown.edu Fall 2017 Machine Learning Subfield of AI concerned with learning from data. Broadly, using: Experience To Improve Performance On Some Task (Tom

More information

Marks. bonus points. } Assignment 1: Should be out this weekend. } Mid-term: Before the last lecture. } Mid-term deferred exam:

Marks. bonus points. } Assignment 1: Should be out this weekend. } Mid-term: Before the last lecture. } Mid-term deferred exam: Marks } Assignment 1: Should be out this weekend } All are marked, I m trying to tally them and perhaps add bonus points } Mid-term: Before the last lecture } Mid-term deferred exam: } This Saturday, 9am-10.30am,

More information

Chapter 3: The Reinforcement Learning Problem

Chapter 3: The Reinforcement Learning Problem Chapter 3: The Reinforcement Learning Problem Objectives of this chapter: describe the RL problem we will be studying for the remainder of the course present idealized form of the RL problem for which

More information

Lecture 3: Policy Evaluation Without Knowing How the World Works / Model Free Policy Evaluation

Lecture 3: Policy Evaluation Without Knowing How the World Works / Model Free Policy Evaluation Lecture 3: Policy Evaluation Without Knowing How the World Works / Model Free Policy Evaluation CS234: RL Emma Brunskill Winter 2018 Material builds on structure from David SIlver s Lecture 4: Model-Free

More information

Machine Learning and Bayesian Inference. Unsupervised learning. Can we find regularity in data without the aid of labels?

Machine Learning and Bayesian Inference. Unsupervised learning. Can we find regularity in data without the aid of labels? Machine Learning and Bayesian Inference Dr Sean Holden Computer Laboratory, Room FC6 Telephone extension 6372 Email: sbh11@cl.cam.ac.uk www.cl.cam.ac.uk/ sbh11/ Unsupervised learning Can we find regularity

More information

Review: TD-Learning. TD (SARSA) Learning for Q-values. Bellman Equations for Q-values. P (s, a, s )[R(s, a, s )+ Q (s, (s ))]

Review: TD-Learning. TD (SARSA) Learning for Q-values. Bellman Equations for Q-values. P (s, a, s )[R(s, a, s )+ Q (s, (s ))] Review: TD-Learning function TD-Learning(mdp) returns a policy Class #: Reinforcement Learning, II 8s S, U(s) =0 set start-state s s 0 choose action a, using -greedy policy based on U(s) U(s) U(s)+ [r

More information

Complexity of stochastic branch and bound methods for belief tree search in Bayesian reinforcement learning

Complexity of stochastic branch and bound methods for belief tree search in Bayesian reinforcement learning Complexity of stochastic branch and bound methods for belief tree search in Bayesian reinforcement learning Christos Dimitrakakis Informatics Institute, University of Amsterdam, Amsterdam, The Netherlands

More information

Reinforcement learning

Reinforcement learning Reinforcement learning Stuart Russell, UC Berkeley Stuart Russell, UC Berkeley 1 Outline Sequential decision making Dynamic programming algorithms Reinforcement learning algorithms temporal difference

More information

COMP3702/7702 Artificial Intelligence Lecture 11: Introduction to Machine Learning and Reinforcement Learning. Hanna Kurniawati

COMP3702/7702 Artificial Intelligence Lecture 11: Introduction to Machine Learning and Reinforcement Learning. Hanna Kurniawati COMP3702/7702 Artificial Intelligence Lecture 11: Introduction to Machine Learning and Reinforcement Learning Hanna Kurniawati Today } What is machine learning? } Where is it used? } Types of machine learning

More information

Discrete planning (an introduction)

Discrete planning (an introduction) Sistemi Intelligenti Corso di Laurea in Informatica, A.A. 2017-2018 Università degli Studi di Milano Discrete planning (an introduction) Nicola Basilico Dipartimento di Informatica Via Comelico 39/41-20135

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Temporal Difference Learning Temporal difference learning, TD prediction, Q-learning, elibigility traces. (many slides from Marc Toussaint) Vien Ngo MLR, University of Stuttgart

More information

AM 121: Intro to Optimization Models and Methods: Fall 2018

AM 121: Intro to Optimization Models and Methods: Fall 2018 AM 11: Intro to Optimization Models and Methods: Fall 018 Lecture 18: Markov Decision Processes Yiling Chen Lesson Plan Markov decision processes Policies and value functions Solving: average reward, discounted

More information

Q-Learning in Continuous State Action Spaces

Q-Learning in Continuous State Action Spaces Q-Learning in Continuous State Action Spaces Alex Irpan alexirpan@berkeley.edu December 5, 2015 Contents 1 Introduction 1 2 Background 1 3 Q-Learning 2 4 Q-Learning In Continuous Spaces 4 5 Experimental

More information

Today s s Lecture. Applicability of Neural Networks. Back-propagation. Review of Neural Networks. Lecture 20: Learning -4. Markov-Decision Processes

Today s s Lecture. Applicability of Neural Networks. Back-propagation. Review of Neural Networks. Lecture 20: Learning -4. Markov-Decision Processes Today s s Lecture Lecture 20: Learning -4 Review of Neural Networks Markov-Decision Processes Victor Lesser CMPSCI 683 Fall 2004 Reinforcement learning 2 Back-propagation Applicability of Neural Networks

More information

Notes on Reinforcement Learning

Notes on Reinforcement Learning 1 Introduction Notes on Reinforcement Learning Paulo Eduardo Rauber 2014 Reinforcement learning is the study of agents that act in an environment with the goal of maximizing cumulative reward signals.

More information

Temporal Difference. Learning KENNETH TRAN. Principal Research Engineer, MSR AI

Temporal Difference. Learning KENNETH TRAN. Principal Research Engineer, MSR AI Temporal Difference Learning KENNETH TRAN Principal Research Engineer, MSR AI Temporal Difference Learning Policy Evaluation Intro to model-free learning Monte Carlo Learning Temporal Difference Learning

More information

PART A and ONE question from PART B; or ONE question from PART A and TWO questions from PART B.

PART A and ONE question from PART B; or ONE question from PART A and TWO questions from PART B. Advanced Topics in Machine Learning, GI13, 2010/11 Advanced Topics in Machine Learning, GI13, 2010/11 Answer any THREE questions. Each question is worth 20 marks. Use separate answer books Answer any THREE

More information