Olivier Sigaud. September 21, 2012

Size: px
Start display at page:

Download "Olivier Sigaud. September 21, 2012"

Transcription

1 Supervised and Reinforcement Learning Tools for Motor Learning Models Olivier Sigaud Université Pierre et Marie Curie - Paris 6 September 21, / 64

2 Introduction Who is speaking? 2 / 64

3 Introduction Two main loops Interaction between cortico-basal loop and cortico-cerebellar loop 3 / 64

4 Introduction Messages of the tutorial A tutorial for a life sciences audience (Not so) dierent learning mechanisms Global view of Reinforcement Learning approaches Understanding recent Actor-Critic advances As few equations as possible......at least in the beginning! If you are interested, you should read : Reinforcement Learning : an Introduction, Sutton and Barto (1998) MIT Press. 4 / 64

5 Introduction Plan Introduction (Not so) dierent learning mechanisms Dynamic programming Reinforcement learning Direct (Model-free) methods: TD learning Action-Value Function Approaches Model-based reinforcement learning Actor-Critic approaches Naive gradient method Modern Actor-Critic Natural gradient NAC versus INAC 5 / 64

6 (Not so) dierent learning mechanisms Associative learning Finding correlations between data The agent builds a model based on data Independent from performance Typical mechanism : Hebb's rule 6 / 64

7 (Not so) dierent learning mechanisms Supervised learning The supervisor indicates to the agent the expected answer The agent corrects a model based on the answer Typical mechanism : gradient backpropagation, RLS Applications : classication, regression, function approximation 7 / 64

8 (Not so) dierent learning mechanisms Self-supervised learning When an agent learns to predict, it proposes its prediction The environment provides the correct answer : next state Supervised learning without supervisor Dicult to distinguish from associative learning 8 / 64

9 (Not so) dierent learning mechanisms Equivalence associative / self-supervised learning Example : LS, RLS Hotly debated issue : does the cerebellum perform supervised learning? Can the Inferior Olive convey an error signal? From outside, both associative and self-supervised learning can explain the same facts With population coding, both mechanisms could be undistinguishable... 9 / 64

10 (Not so) dierent learning mechanisms Reinforcement learning The environment provides the value of action (reward, penalty) Application : behaviour optimisation 10 / 64

11 (Not so) dierent learning mechanisms Reinforcement learning In practice, the value signal is given as a scalar How good is ? Necessity of exploration 11 / 64

12 (Not so) dierent learning mechanisms The exploration/exploitation trade-o Exploring can be (very) harmful Shall I exploit what I know or look for a better policy? Am I optimal? Shall I keep exploring or stop? Decrease the rate of exploration along time ɛ-greedy : take the best action most of the time, and a random action from time to time 12 / 64

13 (Not so) dierent learning mechanisms Temporal Dierence learning in dopamine neurons Udapte of ring rate depending on received reward If the reward is bigger than expected, dopamine neurons re more, If the reward is well predicted, standard level If the reward is smaller than expected, dopamine neurons re less That's not so dierent from a self-supervised learning correction!?!? Not so easy to distinguish learning processes Not so easy to understand the roles of Cerebellum/BG Is there any general underlying (associative) mechanism? 13 / 64

14 Dynamic programming Markov Decision Processes S : states space A : action space T : S A Π(S) : transition function r : S A IR : reward function An MDP denes s t+1 and r t+1 as f (s t, a t) It describes a problem, not a solution 14 / 64

15 Dynamic programming Partial observability/perceptual aliazing Markov property : probability of next state does not depend on history Seldom veried in practice But algorithms tend to be robust 15 / 64

16 Dynamic programming Policy and value functions Goal of planning (and learning) : nd a policy π : S A maximising the agregation of reward on the long run The value function V π : S IR records the agregation of reward on the long run for each state (following policy π) The action value function Q π : S A IR records the agregation of reward on the long run for doing each action in each state (and then following policy π) The computation of value functions assumes the choice of an agregation criterion (discounted, average, etc.) 16 / 64

17 Dynamic programming Bellman equation Given the discounted reward agregation criterion : V (s 0) = r 0 + γv (s 1) Deterministic π : V π (s) = r(s, π(s)) + γ s p(s s, π(s))v π (s ) (1) 17 / 64

18 Dynamic programming Dynamic Programming Looking for an optimal policy boils down to nding an optimal (action-)value function Done through Dynamic Programming (DP) methods : Value Iteration and Policy Iteration They consist in propagating values over the state space So as to determine an optimal policy The optimal (highest) value function corresponds to the optimal policy 18 / 64

19 Dynamic programming Value Iteration in practice R 19 / 64

20 Dynamic programming Value Iteration in practice R 19 / 64

21 Dynamic programming Value Iteration in practice R 19 / 64

22 Dynamic programming Value Iteration in practice R 19 / 64

23 Dynamic programming Value Iteration in practice 19 / 64

24 Dynamic programming Policy Iteration in practice 20 / 64

25 Dynamic programming Policy Iteration in practice 20 / 64

26 Dynamic programming Policy Iteration in practice 20 / 64

27 Dynamic programming Policy Iteration in practice 20 / 64

28 Dynamic programming Policy Iteration in practice 20 / 64

29 Dynamic programming Policy Iteration in practice 20 / 64

30 Dynamic programming Policy Iteration in practice 20 / 64

31 Dynamic programming Policy Iteration in practice 20 / 64

32 Reinforcement learning Reinforcement learning In DP (planning), T and r are given Reinforcement learning goal : build π without knowing T and r Model-free approach : build π without estimating T nor r Actor-critic approach : special case of model-free Model-based approach : build a model of T and r and apply DP 21 / 64

33 Reinforcement learning Direct (Model-free) methods : TD learning Incremental estimation Estimating the average immediate (stochastic) reward in a state s E k (s) = (r 1 + r r k )/k E k+1 (s) = (r 1 + r r k + r k+1 )/(k + 1) Thus E k+1 (s) = k/(k + 1)E k (s) + r k+1 /(k + 1) Or E k+1 (s) = (k + 1)/(k + 1)E k (s) E k (s)/(k + 1) + r k+1 /(k + 1) Or E k+1 (s) = E k (s) + 1/(k + 1)[r k+1 E k (s)] Still needs to store k. Can be approximated as E k+1 (s) = E k (s) + α[r k+1 E k (s)] Will converge to the true average (slower or faster depending on α) without storing anything The formula is everywhere in reinforcement learning 22 / 64

34 Reinforcement learning Direct (Model-free) methods : TD learning Monte Carlo methods Much used in games (Go...) to evaluate a state Generate a lot of trajectories : s 0, s 1,..., s N with observed rewards r 0, r 1,..., r N Update state values V (s k ), k = 0,..., N 1 with : V (s k ) V (s k ) + α(s k )(r k + r k r N V (s k )) It uses the average estimation method from previous slide 23 / 64

35 Reinforcement learning Direct (Model-free) methods : TD learning Temporal Dierence error The goal of TD methods is to estimate the value function V (s) If estimations V (s t) and V (s t+1) were exact, we would get : V (s t) = r t + γr t+1 + γ 2 r t+2 + γ 3 r t V (s t+1) = r t+1 + γ(r t+2 + γ 2 r t Thus V (s t) = r t + γv (s t+1) δ k = r k + γv (s k+1 ) V (s k ) : measures the error between current values of V and the values they should have 24 / 64

36 Reinforcement learning Direct (Model-free) methods : TD learning Temporal Dierence learning in dopamine neurons Udapte rule of the V depending on received reward If r k is bigger than expected, δ > 0 increase V If r k is well predicted, δ = 0 If r k is smaller than expected, δ < 0 decrease V Temporal Dierence method : correct the error by tuning V (s t) with a Widrow-Ho equation : V k (s) V k (s) + α k δ k Bootstrap method : a better estimate from a previous one 25 / 64

37 Reinforcement learning Direct (Model-free) methods : TD learning Temporal Dierence (TD) Methods Temporal Dierence (TD) methods combine the properties of DP methods and Monte Carlo methods : in Monte Carlo, T and r are unknown, but the value update is global, a trajectory is needed in DP, T and r are known, but the value update is local TD : as in DP, V (s t) is updated locally given an estimate of V (s t+1) and T and r are unknown. Note : Monte Carlo can be reformulated incrementally using the temporal dierence δ k udpate 26 / 64

38 Reinforcement learning Direct (Model-free) methods : TD learning TD(0) V (s t) V (s t) + α[r t+1 + γv (s t+1) V (s t)] Combines the temporal dierence udpate (propagation from V (s t+1) to V (s t)) from DP and the incremental estimation method from Monte Carlo. Updates are local from s t, s t+1 and r t+1. Proof of convergence : [?] 27 / 64

39 Reinforcement learning Direct (Model-free) methods : TD learning TD(0) : limitation TD(0) evaluates V (s) One cannot infer π(s) from V (s) without knowing T : one must know which a leads to the best V (s ). Three solutions : Work with Q(s, a) rather than V (s). Learn a model of T : model-based (or indirect) reinforcement learning Actor-critic methods (simultaneously learn V and update π) 28 / 64

40 Reinforcement learning Action-Value Function Approaches Sarsa Action-Value Function Q π : S A IR so that, for each (s, a) S A, Q π (s, a) = r(s, a) + γ s p(s s, a)v π (s ) We have V (s) = max a A Q (s, a) Perform exploration (e.g. ɛ-greedy) For each observed (s t, a t, r t+1, s t+1, a t+1) : Q(s t, a t) Q(s t, a t) + α[r t+1 + γq(s t+1, a t+1) Q(s t, a t)] One must know the action a t+1, thus constrains exploration On-Policy method : more complex convergence proof [?] 29 / 64

41 Reinforcement learning Action-Value Function Approaches Q-Learning For each observed (s t, a t, r t+1, s t+1) : Q(s t, a t) Q(s t, a t) + α[r t+1 + γ max Q(s t+1, a) Q(s t, a t)] a A max a A Q(s t+1, a) instead of Q(s t+1, a t+1) O-Policy method : no more need to know a t+1 Convergence proved provided innite exploration [?] 30 / 64

42 Reinforcement learning Action-Value Function Approaches Q-Learning in practice (Q-learning video) Build a states actions table (Q-Table, eventually incremental) Initialise it (randomly or with 0 is not a good choice) Apply update equation after each action Problem : it is (very) slow [?] : action-specic reward values in the striatum (Science) 31 / 64

43 Reinforcement learning Model-based reinforcement learning Eligibility traces To improve over Q-learning Naive approach : store all (s, a) pair and back-propagate values Limited to nite horizon trajectories Speed/memory trade-o TD(λ), sarsa (λ) and Q(λ) : more sophisticated approach to deal with innite horizon trajectories A variable e(s) is decayed with a factor λ after s was visited and reinitialized each time s is visited again TD(λ) : V (s) V (s) + αδe(s), (similar for sarsa (λ) and Q(λ)), If λ = 0, e(s) goes to 0 immediately, thus we get TD(0), sarsa or Q-learning TD(1) = Monte-Carlo / 64

44 Reinforcement learning Model-based reinforcement learning Dyna architecture : ideas (Dyna-like video) General idea : planning with a learnt model is performing back-ups in the agent's head ([?,?]) Learn a model of T and r Use Policy Iteration (Dyna-PI) or Q-learning (Dyna-Q) to get V or Q There is also Dyna-AC 33 / 64

45 Reinforcement learning Model-based reinforcement learning Dyna architecture : biological background Learning T and r is self-supervised learning In the continuous case, need for function approximator LWPR, gaussian processes, RBF... [?] : combining model-free and model-based 34 / 64

46 Actor-Critic approaches Continuous action problems Until (very) recently, there was very few work on continuous action problems, though reinforcement learning (partially) comes from optimal control Problem : in standard TD methods, one must perform max over actions (several times), (optimisation problem when action is continuous) Gradient methods and Actor-Critic methods reduce the problem, they look for a local optimum (Pontryagine method rather than Bellman method) They get very popular nowadays 35 / 64

47 Actor-Critic approaches Convergence problems Continuous states and actions cannot be enumerated function approximation Combining function approximation and greedy methods based on action-value can diverge easily ([?,?]...) [?] : algorithms based on the advantage function A π (s, a) = Q π (s, a) V π (s) converge more robustly Since V π (s) = max aq π (s, a), A π (s, a) is a negative term equivalent to a regret 36 / 64

48 Actor-Critic approaches Biological background Basal ganglia might be organised into an Actor-Critic architecture 37 / 64

49 Actor-Critic approaches From Q(s, a) to Actor-Critic (1) state / action a 0 a 1 a 2 a 3 e e e e e e In Q learning, given a Q Table, one must determine the max at each step This becomes expensive if there are numerous actions 38 / 64

50 Actor-Critic approaches From Q(s, a) to Actor-Critic (2) state / action a 0 a 1 a 2 a 3 e * e * 0.43 e * 0.73 e * 0.81 e * e * One can store the best value for each state Then one can update the max by just comparing the changed value and the max No more maximum over actions (only in one case) 39 / 64

51 Actor-Critic approaches From Q(s, a) to Actor-Critic (3) state / action a 0 a 1 a 2 a 3 e * e * 0.43 e * 0.73 e * 0.81 e * e * state chosen action e 0 a 1 e 1 a 2 e 2 a 2 e 3 a 2 e 4 a 1 e 5 a 1 Store the max is equivalent to store the policy by the side of the Q Table Update the policy as a function of value updates Basic actor-critic scheme 40 / 64

52 Actor-Critic approaches Dynamic Programming and Actor-Critic (1) In both PI and AC, the architecture contains a representation of the value function (the critic) and the policy (the actor) In PI, the environment (T and r) is known PI alternates two stages : 1. Policy evaluation : update (V (s)) or (Q(s, a)) given the current policy 2. Policy improvement : follow the value gradient 41 / 64

53 Actor-Critic approaches Dynamic Programming and Actor-Critic (2) In AC, T and r are unknown, but not represented (model-free) Information from the environment generates update in the critic, then in the actor 42 / 64

54 Actor-Critic approaches Successive contributions 43 / 64

55 Actor-Critic approaches Naive design Discrete states and actions, stochastic policy An update in the critic generates a local update in the actor Critic : compute δ and update V (s) with V k (s) V k (s) + α k δ k Actor : P π (a s) = P π (a s) + α k δ k NB : no need for a max over actions NB2 : one must then know how to draw an action from a probabilistic policy (not obvious for continuous actions) 44 / 64

56 Actor-Critic approaches Naive gradient method Gradient methods : Parametrized policies We get stochastic policies parametrized by a vector θ, noted π θ (a s) Look for the gradient θ of the long term performance J(π θ ) J is written dierently depending on the criterion : average reward, discounted, etc. The average reward criterion gives rise to clearest prooves nowadays We want to maximize J(π θ ) with a gradient ascent The gradient is estimated with samples collected along trajectories We note θ = J(π θ(a s)) θ 45 / 64

57 Actor-Critic approaches Naive gradient method Naive gradient computation The long term performance J θ in any state is unkown Naive approach : in a state, perform trajectories from that state with dierent θ, estimate performance, then its gradient wrt θ (REINFORCE) Very expensive and slow Given that V π θ (or Q π θ ) summarizes performance of π θ in all states Can we use estimates of V π θ or Q π θ to compute the gradient? [?] shows that yes, we can : value-based methods can be as good as REINFORCE with much fewer samples 46 / 64

58 Actor-Critic approaches Modern Actor-Critic An estimation problem Estimating : approximating a function thanks to a model, some samples, and a method Variance : variability of estimation as a function of samples Bias : error on estimate after convergence due to method (sampling, etc.) due to the model (structure or parameters) 47 / 64

59 Actor-Critic approaches Modern Actor-Critic Bias/variance trade-o Poor model : low variance, large bias Rich model : low bias, large variance Bias/variance trade-o : tune parameters : more parameters = less bias, more variance If we get few samples, better estimation accepting more bias to prevent variance To prove convergence : ensure null bias, then minimise variance 48 / 64

60 Actor-Critic approaches Modern Actor-Critic Gradient computation with value function (1) For a stochastic policy in the discrete case, one shows ([?], theorem 1) that the average reward gradient is written : θ J(π θ ) = s S d π θ (s) a A θ π θ (a s)q π θ (s, a) Where d π θ (s) represents the probability density of being in s following policy π θ in stationary mode d π θ (s) corresponds to sampling peformed by the policy Similar formula for discounted reward case 49 / 64

61 Actor-Critic approaches Modern Actor-Critic Gradient computation with value function (2) We have θ J(π θ ) = s S d π θ (s) a A θπ θ (a s)q π θ (s, a) Thus the gradient can be computed as a function of Q π θ (s, a) Problem : the exact Q π θ (s, a) is unknown, it must be estimated with a parametric representation We set f ω(s, a) = ˆQπ θ ω (s, a) Question : under which condition on f ω(s, a) as a function of π θ (s, a) do we get an unbiased estimation of Q π θ (s, a)? 50 / 64

62 Actor-Critic approaches Modern Actor-Critic Compatible functions It is shown that ([?], theorem 2) such a condition is fω(s,a) = 1 ω π θ (a s) θπ θ (a s) = θ logπ θ (a s) The gradient is given by : θ J(π θ ) = s S d π (s) a A fω(s, a) θ π θ (a s) ω 51 / 64

63 Actor-Critic approaches Modern Actor-Critic Compatibility condition For a policy π θ (s, a) The gradient is given by : θ J(π θ ) = s S d π θ (s) a A fω(s, a) θ π θ (a s) ω is unbiased if fω(s,a) ω = 1 π θ (a s) θπ θ (a s) = θ logπ θ (a s) Thus if f ω(s, a) = ω T θ logπ θ (a s) Proof in [?] The relation must be ensured between the basis functions used in the actor and the critic 52 / 64

64 Actor-Critic approaches Modern Actor-Critic Baseline We have θ J(π θ ) = d π (s) s S a A θπ θ (a s)q π θ (s, a) Introduce a baseline b(s) in the computation of θ J(π θ ) It is proven that, for any b(s), θ J(π θ ) = d π θ (s) θ π θ (a s)(q π θ (s, a) b(s)) s S a A Look for b(s) that most reduces the variance 53 / 64

65 Actor-Critic approaches Modern Actor-Critic Optimal baseline The baseline with the least variance is V π θ (s) ([?], lemma 2) Thus one should compute θ J(π θ ) = d π θ (s) θ π θ (a s)(q π θ (s, a) V π θ (s)) s S a A Thus the best critic (Q compatible approximator) estimates the advantage function from [?] : A π (s, a) = Q π (s, a) V π (s) 54 / 64

66 Actor-Critic approaches Modern Actor-Critic Reminder : Successive contributions 55 / 64

67 Actor-Critic approaches Natural gradient Natural gradient denition A particular parametrisation induces a particular way to explore a sub-space of stochastic policies A parametrisation induces a variety in the space of stochastic policies If the variety is not euclidean, the gradient may not follow the steepest descent direction (another parametrisation would induce another direction) The natural gradient ensures that the steepest descent direction is followed whatever the parametrisation 56 / 64

68 Actor-Critic approaches Natural gradient Natural gradient computation ˆ θ J θ (π θ ) = F 1 (θ) θ J θ (π θ ) F is the Fischer Information matrix Besides, θ J θ (π θ ) = E[ θ logπ θ logπ T ]ω = F (θ)ω Thus ˆ θ J θ (π θ ) = F 1 (θ)f (θ)ω = ω Thus, to update the policy gradient, one can use θ t+1 = θ t + α tω t where 57 / 64

69 Actor-Critic approaches NAC versus INAC NAC based on (LSTD-Q(λ)) ω t is an estimator of A π [?] : one cannot estimate A π thanks to a bootstrap method Use V π (s) = v T f (s) where f (s) are state features Estimate jointly ω t and v t with a least square method Add features f (s) on state s 58 / 64

70 Actor-Critic approaches NAC versus INAC enac (simplication) NAC needs n trajectories to solve a n linear equations system Use of a forgetting factor not to reinitialize everything at each step You don't need to dene value function features Problem : not incremental, not biologically plausible 59 / 64

71 Actor-Critic approaches NAC versus INAC INAC [?] found a way to estimate ω t and v t with a bootstrap method As in NAC, we use V π (s) = v T f (s) where f (s) are state features We use δ t = r t+1 Ĵ t+1 + ˆVt+1 ˆVt Estimated with δ t = r t+1 Ĵ t+1 + v T t f (s t+1) v T t f (s t) v t is updated with temporal dierence : v t+1 = v t + α tδ tf (s t) 60 / 64

72 Actor-Critic approaches NAC versus INAC Natural gradient INAC There are four INAC update equations One is based on the natural gradient, requires less computation Critic update : 1. v t+1 = v t + α tδ t f (s t) 2. ω t+1 = [I α tψ(s t, a t)ψ(s t, a t) T ]ω t + α tδ tψ(s t, a t) Actor update : θ t+1 = θ t + β tω t+1 NB : β must converge to 0 faster than α Diculties : 1. Finding basis functions ensuring compatibility condition for continuous action problems 2. Tuning the parameters is horrible Late breaking news : enac is much more easy to tune 61 / 64

73 Actor-Critic approaches NAC versus INAC Model-based Incremental Natural Actor-Critic 62 / 64

74 Actor-Critic approaches NAC versus INAC Messages Distinguishing learning mechanisms may be misleading Actor-critic methods are ecient for motor control modelling They seem to be biologically plausible They are not model-based Much progress in maths since naive early proposal (NAC and INAC) That progress did not propagate to biological modelling There is room for model-based actor-critic architectures 63 / 64

75 Actor-Critic approaches NAC versus INAC Future directions Look for neurophysiological counterparts of recent actor-critic mechanisms See how this actor-critic architecture can be embedded into a wider motor learning model (attend my CMCW talk on thursday) Towards a Kalman lter based actor-critic architecture (Geist et al. 2009) 64 / 64

Reinforcement Learning: the basics

Reinforcement Learning: the basics Reinforcement Learning: the basics Olivier Sigaud Université Pierre et Marie Curie, PARIS 6 http://people.isir.upmc.fr/sigaud August 6, 2012 1 / 46 Introduction Action selection/planning Learning by trial-and-error

More information

Lecture 8: Policy Gradient

Lecture 8: Policy Gradient Lecture 8: Policy Gradient Hado van Hasselt Outline 1 Introduction 2 Finite Difference Policy Gradient 3 Monte-Carlo Policy Gradient 4 Actor-Critic Policy Gradient Introduction Vapnik s rule Never solve

More information

15-889e Policy Search: Gradient Methods Emma Brunskill. All slides from David Silver (with EB adding minor modificafons), unless otherwise noted

15-889e Policy Search: Gradient Methods Emma Brunskill. All slides from David Silver (with EB adding minor modificafons), unless otherwise noted 15-889e Policy Search: Gradient Methods Emma Brunskill All slides from David Silver (with EB adding minor modificafons), unless otherwise noted Outline 1 Introduction 2 Finite Difference Policy Gradient

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

Actor-critic methods. Dialogue Systems Group, Cambridge University Engineering Department. February 21, 2017

Actor-critic methods. Dialogue Systems Group, Cambridge University Engineering Department. February 21, 2017 Actor-critic methods Milica Gašić Dialogue Systems Group, Cambridge University Engineering Department February 21, 2017 1 / 21 In this lecture... The actor-critic architecture Least-Squares Policy Iteration

More information

Lecture 9: Policy Gradient II 1

Lecture 9: Policy Gradient II 1 Lecture 9: Policy Gradient II 1 Emma Brunskill CS234 Reinforcement Learning. Winter 2019 Additional reading: Sutton and Barto 2018 Chp. 13 1 With many slides from or derived from David Silver and John

More information

Reinforcement Learning. Spring 2018 Defining MDPs, Planning

Reinforcement Learning. Spring 2018 Defining MDPs, Planning Reinforcement Learning Spring 2018 Defining MDPs, Planning understandability 0 Slide 10 time You are here Markov Process Where you will go depends only on where you are Markov Process: Information state

More information

Temporal difference learning

Temporal difference learning Temporal difference learning AI & Agents for IET Lecturer: S Luz http://www.scss.tcd.ie/~luzs/t/cs7032/ February 4, 2014 Recall background & assumptions Environment is a finite MDP (i.e. A and S are finite).

More information

Reinforcement Learning with Function Approximation. Joseph Christian G. Noel

Reinforcement Learning with Function Approximation. Joseph Christian G. Noel Reinforcement Learning with Function Approximation Joseph Christian G. Noel November 2011 Abstract Reinforcement learning (RL) is a key problem in the field of Artificial Intelligence. The main goal is

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

Lecture 9: Policy Gradient II (Post lecture) 2

Lecture 9: Policy Gradient II (Post lecture) 2 Lecture 9: Policy Gradient II (Post lecture) 2 Emma Brunskill CS234 Reinforcement Learning. Winter 2018 Additional reading: Sutton and Barto 2018 Chp. 13 2 With many slides from or derived from David Silver

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

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

Reinforcement Learning and NLP

Reinforcement Learning and NLP 1 Reinforcement Learning and NLP Kapil Thadani kapil@cs.columbia.edu RESEARCH Outline 2 Model-free RL Markov decision processes (MDPs) Derivative-free optimization Policy gradients Variance reduction Value

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Policy Search: Actor-Critic and Gradient Policy search Mario Martin CS-UPC May 28, 2018 Mario Martin (CS-UPC) Reinforcement Learning May 28, 2018 / 63 Goal of this lecture So far

More information

Lecture 23: Reinforcement Learning

Lecture 23: Reinforcement Learning Lecture 23: Reinforcement Learning MDPs revisited Model-based learning Monte Carlo value function estimation Temporal-difference (TD) learning Exploration November 23, 2006 1 COMP-424 Lecture 23 Recall:

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Function approximation Mario Martin CS-UPC May 18, 2018 Mario Martin (CS-UPC) Reinforcement Learning May 18, 2018 / 65 Recap Algorithms: MonteCarlo methods for Policy Evaluation

More information

Machine Learning I Continuous Reinforcement Learning

Machine Learning I Continuous Reinforcement Learning Machine Learning I Continuous Reinforcement Learning Thomas Rückstieß Technische Universität München January 7/8, 2010 RL Problem Statement (reminder) state s t+1 ENVIRONMENT reward r t+1 new step r t

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

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

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 Marc Toussaint University of

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

Reinforcement Learning

Reinforcement Learning Reinforcement Learning RL in continuous MDPs March April, 2015 Large/Continuous MDPs Large/Continuous state space Tabular representation cannot be used Large/Continuous action space Maximization over action

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Policy gradients Daniel Hennes 26.06.2017 University Stuttgart - IPVS - Machine Learning & Robotics 1 Policy based reinforcement learning So far we approximated the action value

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

, 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

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

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

CS599 Lecture 1 Introduction To RL

CS599 Lecture 1 Introduction To RL CS599 Lecture 1 Introduction To RL Reinforcement Learning Introduction Learning from rewards Policies Value Functions Rewards Models of the Environment Exploitation vs. Exploration Dynamic Programming

More information

Lecture 7: Value Function Approximation

Lecture 7: Value Function Approximation Lecture 7: Value Function Approximation Joseph Modayil Outline 1 Introduction 2 3 Batch Methods Introduction Large-Scale Reinforcement Learning Reinforcement learning can be used to solve large problems,

More information

Reinforcement learning

Reinforcement learning Reinforcement learning Based on [Kaelbling et al., 1996, Bertsekas, 2000] Bert Kappen Reinforcement learning Reinforcement learning is the problem faced by an agent that must learn behavior through trial-and-error

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

Reinforcement Learning Part 2

Reinforcement Learning Part 2 Reinforcement Learning Part 2 Dipendra Misra Cornell University dkm@cs.cornell.edu https://dipendramisra.wordpress.com/ From previous tutorial Reinforcement Learning Exploration No supervision Agent-Reward-Environment

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

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Function approximation Daniel Hennes 19.06.2017 University Stuttgart - IPVS - Machine Learning & Robotics 1 Today Eligibility traces n-step TD returns Forward and backward view Function

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

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

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

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

Off-Policy Actor-Critic

Off-Policy Actor-Critic Off-Policy Actor-Critic Ludovic Trottier Laval University July 25 2012 Ludovic Trottier (DAMAS Laboratory) Off-Policy Actor-Critic July 25 2012 1 / 34 Table of Contents 1 Reinforcement Learning Theory

More information

Chapter 7: Eligibility Traces. R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction 1

Chapter 7: Eligibility Traces. R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction 1 Chapter 7: Eligibility Traces R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction 1 Midterm Mean = 77.33 Median = 82 R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction

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

6 Reinforcement Learning

6 Reinforcement Learning 6 Reinforcement Learning As discussed above, a basic form of supervised learning is function approximation, relating input vectors to output vectors, or, more generally, finding density functions p(y,

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

Temporal Difference Learning & Policy Iteration

Temporal Difference Learning & Policy Iteration Temporal Difference Learning & Policy Iteration Advanced Topics in Reinforcement Learning Seminar WS 15/16 ±0 ±0 +1 by Tobias Joppen 03.11.2015 Fachbereich Informatik Knowledge Engineering Group Prof.

More information

On and Off-Policy Relational Reinforcement Learning

On and Off-Policy Relational Reinforcement Learning On and Off-Policy Relational Reinforcement Learning Christophe Rodrigues, Pierre Gérard, and Céline Rouveirol LIPN, UMR CNRS 73, Institut Galilée - Université Paris-Nord first.last@lipn.univ-paris13.fr

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

arxiv: v1 [cs.ai] 5 Nov 2017

arxiv: v1 [cs.ai] 5 Nov 2017 arxiv:1711.01569v1 [cs.ai] 5 Nov 2017 Markus Dumke Department of Statistics Ludwig-Maximilians-Universität München markus.dumke@campus.lmu.de Abstract Temporal-difference (TD) learning is an important

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Dipendra Misra Cornell University dkm@cs.cornell.edu https://dipendramisra.wordpress.com/ Task Grasp the green cup. Output: Sequence of controller actions Setup from Lenz et. al.

More information

Reinforcement Learning via Policy Optimization

Reinforcement Learning via Policy Optimization Reinforcement Learning via Policy Optimization Hanxiao Liu November 22, 2017 1 / 27 Reinforcement Learning Policy a π(s) 2 / 27 Example - Mario 3 / 27 Example - ChatBot 4 / 27 Applications - Video Games

More information

Reinforcement Learning using Continuous Actions. Hado van Hasselt

Reinforcement Learning using Continuous Actions. Hado van Hasselt Reinforcement Learning using Continuous Actions Hado van Hasselt 2005 Concluding thesis for Cognitive Artificial Intelligence University of Utrecht First supervisor: Dr. Marco A. Wiering, University of

More information

CS 287: Advanced Robotics Fall Lecture 14: Reinforcement Learning with Function Approximation and TD Gammon case study

CS 287: Advanced Robotics Fall Lecture 14: Reinforcement Learning with Function Approximation and TD Gammon case study CS 287: Advanced Robotics Fall 2009 Lecture 14: Reinforcement Learning with Function Approximation and TD Gammon case study Pieter Abbeel UC Berkeley EECS Assignment #1 Roll-out: nice example paper: X.

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

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

INF 5860 Machine learning for image classification. Lecture 14: Reinforcement learning May 9, 2018

INF 5860 Machine learning for image classification. Lecture 14: Reinforcement learning May 9, 2018 Machine learning for image classification Lecture 14: Reinforcement learning May 9, 2018 Page 3 Outline Motivation Introduction to reinforcement learning (RL) Value function based methods (Q-learning)

More information

REINFORCE Framework for Stochastic Policy Optimization and its use in Deep Learning

REINFORCE Framework for Stochastic Policy Optimization and its use in Deep Learning REINFORCE Framework for Stochastic Policy Optimization and its use in Deep Learning Ronen Tamari The Hebrew University of Jerusalem Advanced Seminar in Deep Learning (#67679) February 28, 2016 Ronen Tamari

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

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

Variance Reduction for Policy Gradient Methods. March 13, 2017

Variance Reduction for Policy Gradient Methods. March 13, 2017 Variance Reduction for Policy Gradient Methods March 13, 2017 Reward Shaping Reward Shaping Reward Shaping Reward shaping: r(s, a, s ) = r(s, a, s ) + γφ(s ) Φ(s) for arbitrary potential Φ Theorem: r admits

More information

Lecture 1: March 7, 2018

Lecture 1: March 7, 2018 Reinforcement Learning Spring Semester, 2017/8 Lecture 1: March 7, 2018 Lecturer: Yishay Mansour Scribe: ym DISCLAIMER: Based on Learning and Planning in Dynamical Systems by Shie Mannor c, all rights

More information

Lecture 8: Policy Gradient I 2

Lecture 8: Policy Gradient I 2 Lecture 8: Policy Gradient I 2 Emma Brunskill CS234 Reinforcement Learning. Winter 2018 Additional reading: Sutton and Barto 2018 Chp. 13 2 With many slides from or derived from David Silver and John Schulman

More information

REINFORCEMENT LEARNING

REINFORCEMENT LEARNING REINFORCEMENT LEARNING Larry Page: Where s Google going next? DeepMind's DQN playing Breakout Contents Introduction to Reinforcement Learning Deep Q-Learning INTRODUCTION TO REINFORCEMENT LEARNING Contents

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

Policy Gradient Methods. February 13, 2017

Policy Gradient Methods. February 13, 2017 Policy Gradient Methods February 13, 2017 Policy Optimization Problems maximize E π [expression] π Fixed-horizon episodic: T 1 Average-cost: lim T 1 T r t T 1 r t Infinite-horizon discounted: γt r t Variable-length

More information

(Deep) Reinforcement Learning

(Deep) Reinforcement Learning Martin Matyášek Artificial Intelligence Center Czech Technical University in Prague October 27, 2016 Martin Matyášek VPD, 2016 1 / 17 Reinforcement Learning in a picture R. S. Sutton and A. G. Barto 2015

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

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

Bias-Variance Error Bounds for Temporal Difference Updates

Bias-Variance Error Bounds for Temporal Difference Updates Bias-Variance Bounds for Temporal Difference Updates Michael Kearns AT&T Labs mkearns@research.att.com Satinder Singh AT&T Labs baveja@research.att.com Abstract We give the first rigorous upper bounds

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

Lecture 3: Markov Decision Processes

Lecture 3: Markov Decision Processes Lecture 3: Markov Decision Processes Joseph Modayil 1 Markov Processes 2 Markov Reward Processes 3 Markov Decision Processes 4 Extensions to MDPs Markov Processes Introduction Introduction to MDPs Markov

More information

Trust Region Policy Optimization

Trust Region Policy Optimization Trust Region Policy Optimization Yixin Lin Duke University yixin.lin@duke.edu March 28, 2017 Yixin Lin (Duke) TRPO March 28, 2017 1 / 21 Overview 1 Preliminaries Markov Decision Processes Policy iteration

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

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

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

16.410/413 Principles of Autonomy and Decision Making

16.410/413 Principles of Autonomy and Decision Making 16.410/413 Principles of Autonomy and Decision Making Lecture 23: Markov Decision Processes Policy Iteration Emilio Frazzoli Aeronautics and Astronautics Massachusetts Institute of Technology December

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

Reinforcement Learning for NLP

Reinforcement Learning for NLP Reinforcement Learning for NLP Advanced Machine Learning for NLP Jordan Boyd-Graber REINFORCEMENT OVERVIEW, POLICY GRADIENT Adapted from slides by David Silver, Pieter Abbeel, and John Schulman Advanced

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

Learning Control for Air Hockey Striking using Deep Reinforcement Learning

Learning Control for Air Hockey Striking using Deep Reinforcement Learning Learning Control for Air Hockey Striking using Deep Reinforcement Learning Ayal Taitler, Nahum Shimkin Faculty of Electrical Engineering Technion - Israel Institute of Technology May 8, 2017 A. Taitler,

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

Reinforcement Learning and Deep Reinforcement Learning

Reinforcement Learning and Deep Reinforcement Learning Reinforcement Learning and Deep Reinforcement Learning Ashis Kumer Biswas, Ph.D. ashis.biswas@ucdenver.edu Deep Learning November 5, 2018 1 / 64 Outlines 1 Principles of Reinforcement Learning 2 The Q

More information

Sequential Decision Problems

Sequential Decision Problems Sequential Decision Problems Michael A. Goodrich November 10, 2006 If I make changes to these notes after they are posted and if these changes are important (beyond cosmetic), the changes will highlighted

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

Deep Reinforcement Learning SISL. Jeremy Morton (jmorton2) November 7, Stanford Intelligent Systems Laboratory

Deep Reinforcement Learning SISL. Jeremy Morton (jmorton2) November 7, Stanford Intelligent Systems Laboratory Deep Reinforcement Learning Jeremy Morton (jmorton2) November 7, 2016 SISL Stanford Intelligent Systems Laboratory Overview 2 1 Motivation 2 Neural Networks 3 Deep Reinforcement Learning 4 Deep Learning

More information

Linear Least-squares Dyna-style Planning

Linear Least-squares Dyna-style Planning Linear Least-squares Dyna-style Planning Hengshuai Yao Department of Computing Science University of Alberta Edmonton, AB, Canada T6G2E8 hengshua@cs.ualberta.ca Abstract World model is very important for

More information

CSC321 Lecture 22: Q-Learning

CSC321 Lecture 22: Q-Learning CSC321 Lecture 22: Q-Learning Roger Grosse Roger Grosse CSC321 Lecture 22: Q-Learning 1 / 21 Overview Second of 3 lectures on reinforcement learning Last time: policy gradient (e.g. REINFORCE) Optimize

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

Tutorial on Policy Gradient Methods. Jan Peters

Tutorial on Policy Gradient Methods. Jan Peters Tutorial on Policy Gradient Methods Jan Peters Outline 1. Reinforcement Learning 2. Finite Difference vs Likelihood-Ratio Policy Gradients 3. Likelihood-Ratio Policy Gradients 4. Conclusion General Setup

More information

Dialogue management: Parametric approaches to policy optimisation. Dialogue Systems Group, Cambridge University Engineering Department

Dialogue management: Parametric approaches to policy optimisation. Dialogue Systems Group, Cambridge University Engineering Department Dialogue management: Parametric approaches to policy optimisation Milica Gašić Dialogue Systems Group, Cambridge University Engineering Department 1 / 30 Dialogue optimisation as a reinforcement learning

More information

Reinforcement Learning. Summer 2017 Defining MDPs, Planning

Reinforcement Learning. Summer 2017 Defining MDPs, Planning Reinforcement Learning Summer 2017 Defining MDPs, Planning understandability 0 Slide 10 time You are here Markov Process Where you will go depends only on where you are Markov Process: Information state

More information

Reinforcement Learning II. George Konidaris

Reinforcement Learning II. George Konidaris Reinforcement Learning II George Konidaris gdk@cs.brown.edu Fall 2017 Reinforcement Learning π : S A max R = t=0 t r t MDPs Agent interacts with an environment At each time t: Receives sensor signal Executes

More information

Zero-Sum Stochastic Games An algorithmic review

Zero-Sum Stochastic Games An algorithmic review Zero-Sum Stochastic Games An algorithmic review Emmanuel Hyon LIP6/Paris Nanterre with N Yemele and L Perrotin Rosario November 2017 Final Meeting Dygame Dygame Project Amstic Outline 1 Introduction Static

More information

15-780: ReinforcementLearning

15-780: ReinforcementLearning 15-780: ReinforcementLearning J. Zico Kolter March 2, 2016 1 Outline Challenge of RL Model-based methods Model-free methods Exploration and exploitation 2 Outline Challenge of RL Model-based methods Model-free

More information

Reinforcement Learning as Classification Leveraging Modern Classifiers

Reinforcement Learning as Classification Leveraging Modern Classifiers Reinforcement Learning as Classification Leveraging Modern Classifiers Michail G. Lagoudakis and Ronald Parr Department of Computer Science Duke University Durham, NC 27708 Machine Learning Reductions

More information

Deep reinforcement learning. Dialogue Systems Group, Cambridge University Engineering Department

Deep reinforcement learning. Dialogue Systems Group, Cambridge University Engineering Department Deep reinforcement learning Milica Gašić Dialogue Systems Group, Cambridge University Engineering Department 1 / 25 In this lecture... Introduction to deep reinforcement learning Value-based Deep RL Deep

More information

Animal learning theory

Animal learning theory Animal learning theory Based on [Sutton and Barto, 1990, Dayan and Abbott, 2001] Bert Kappen [Sutton and Barto, 1990] Classical conditioning: - A conditioned stimulus (CS) and unconditioned stimulus (US)

More information

Dueling Network Architectures for Deep Reinforcement Learning (ICML 2016)

Dueling Network Architectures for Deep Reinforcement Learning (ICML 2016) Dueling Network Architectures for Deep Reinforcement Learning (ICML 2016) Yoonho Lee Department of Computer Science and Engineering Pohang University of Science and Technology October 11, 2016 Outline

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Ron Parr CompSci 7 Department of Computer Science Duke University With thanks to Kris Hauser for some content RL Highlights Everybody likes to learn from experience Use ML techniques

More information

Reinforcement Learning II. George Konidaris

Reinforcement Learning II. George Konidaris Reinforcement Learning II George Konidaris gdk@cs.brown.edu Fall 2018 Reinforcement Learning π : S A max R = t=0 t r t MDPs Agent interacts with an environment At each time t: Receives sensor signal Executes

More information

Elements of Reinforcement Learning

Elements of Reinforcement Learning Elements of Reinforcement Learning Policy: way learning algorithm behaves (mapping from state to action) Reward function: Mapping of state action pair to reward or cost Value function: long term reward,

More information