learning curriculum policies for reinforcement learning
play

Learning(Curriculum(Policies(for( Reinforcement(Learning - PowerPoint PPT Presentation

Mar 13, 2024 •154 likes •459 views

Learning(Curriculum(Policies(for( Reinforcement(Learning Sanmit'Narvekar and$Peter$Stone Department$of$Computer$Science University$of$Texas$at$Austin {sanmit,$pstone}$@cs.utexas.edu Successes(of(Reinforcement(Learning

  1. Learning(Curriculum(Policies(for( Reinforcement(Learning Sanmit'Narvekar and$Peter$Stone Department$of$Computer$Science University$of$Texas$at$Austin {sanmit,$pstone}$@cs.utexas.edu

  2. Successes(of(Reinforcement(Learning Approaching$or$passing$human$level$performance BUT Can$take$ millions of$episodes!$People$learn$this$MUCH faster University$of$Texas$at$Austin Sanmit$Narvekar 2

  3. People(Learn(via(Curricula People$are$able$to$learn$a$lot$of$complex$tasks$very$efficiently$ University$of$Texas$at$Austin Sanmit$Narvekar 3

  4. Example:(Quick(Chess • Quickly$learn$the$ fundamentals$of$chess • 5$x$6$board$ • Fewer$pieces$per$type • No$castling • No$enOpassant$ University$of$Texas$at$Austin Sanmit$Narvekar 4

  5. Example:(Quick(Chess .$.$.$.$.$. University$of$Texas$at$Austin Sanmit$Narvekar 5

  6. Task(Space Pawns$+$King Pawns$only Target$task Empty$task One$piece$per$type • Quick$Chess$is$a$curriculum$designed$for$people • We$want$to$do$something$similar$automatically for$autonomous$agents University$of$Texas$at$Austin Sanmit$Narvekar 6

  7. Curriculum(Learning Task$=$MDP Environment State Action Reward Agent Task'Creation Assume$Given Transfer'Learning Sequencing This$work:$2$types • Curriculum$learning$is$a$complex$problem$that$ties$task$creation,$sequencing,$ and$transfer$learning University$of$Texas$at$Austin Sanmit$Narvekar 7

  8. Value(Function(Transfer • Initialize Q$function$in$target$task$using$values$learned$in$a$ source task Q source (s,a) • Assumptions: • Tasks$have$overlapping state$and$action$spaces$ • OR$an$interOtask$mapping is$provided • Existing$related$work$on$learning$mappings Image$credit:$Taylor$and$Stone,$JMLR$2009 University$of$Texas$at$Austin Sanmit$Narvekar 8

  9. Reward(Shaping(Transfer • Reward$function$in$target$task$augmented with$a$shaping$reward$ f : New$Reward Old$Reward Shaping$Reward • PotentialObased$advice$restricts$f$to$be$difference$of$potential$ functions: • Use$the$value$function$of$the$source as$the$potential$function: University$of$Texas$at$Austin Sanmit$Narvekar 9

  10. The(Problem:(Autonomous(Sequencing • Existing'work'heuristic=based ,$such$as$examining$performance$on$the$ target$task,$and$using$heuristics$to$select$next$task$ • In$this$work,$we$use$ learning'to'do'sequencing University$of$Texas$at$Austin Sanmit$Narvekar 10

  11. Sequencing(as(an(MDP Curriculum$Task Curriculum Agent Curriculum$Action Task$2 Task$1 Task$N Environment Environment Environment State State State Action Action Action Reward Reward Reward RL$ RL$ RL$ Agent Agent Agent Curriculum$State Curriculum$Reward Curriculum Agent University$of$Texas$at$Austin Sanmit$Narvekar 11

  12. Sequencing(as(an(MDP M 3 ! 1 R 1,3 M 1 ! 4 M 4 R 0,1 R 4,4 M 4 M 3 ! f ! 2 R 2,4 ! 0 R 0,3 M 4 R 5,4 M 2 ! 5 M 3 R 0,2 R 3,3 ! 3 • State'space' S C :$All$policies$ ! i an$agent$can$represent • Action'space' A C :$Different$tasks$ M j an$agent$can$train$on • Transition'function' p C (s C ,a C ) :$Learning$task$ a C transforms$an$agent’s$policy$ s C • Reward'function' r C (s C ,a C ) :$Cost$in$time$steps$to$learn$task$ a C given$policy$ s C University$of$Texas$at$Austin Sanmit$Narvekar 12

  13. Sequencing(as(an(MDP M 3 ! 1 R 1,3 M 1 ! 4 M 4 R 0,1 R 4,4 M 4 M 3 ! f ! 2 R 2,4 ! 0 R 0,3 M 4 R 5,4 M 2 ! 5 M 3 R 0,2 R 3,3 ! 3 • A$policy ! C :$S C ! A C on$this$curriculum$MDP$(CMDP)$specifies$which$task$to$ train$on$given$learning$agent$policy$ ! i • Essentially$training$a$teacher • How$to$do$learning$over$CMDP? • How$does$CMDP$change$when$transfer$method$changes? University$of$Texas$at$Austin Sanmit$Narvekar 13

  14. Learning(in(Curriculum(MDPs [1,3,4,…0] [1,2,3,…0] M 3 ! 1 R 1,3 M 1 ! 4 M 4 R 0,1 [1,2,3,…9] R 4,4 M 4 M 3 ! f ! 2 R 2,4 ! 0 R 0,3 M 4 R 5,4 M 2 ! 5 [0,0,0,…0] M 3 R 0,2 R 3,3 ! 3 Extract$Raw$CMDP$ Function$Approximation$ Extract$Features State$Variables and$Learning • Express$raw$CMDP$state$using$the$weights$of$base$agent’s$VF/policy • Extract$features so$that$similar$policies$(CMDP$states)$are$“close”$in$feature$ space University$of$Texas$at$Austin Sanmit$Narvekar 14

  15. Example:(Discrete(Representations State' State'2 State'3 State'4 1 CMDP'State'1 CMDP'State'2 CMDP'State'3 Left Right Policy Left Right Policy Left Right Policy State$1 0.3 0.7 ! State$1 0.2 0.8 ! State$1 0.7 0.3 " State$2 0.1 0.9 State$2 0.2 0.8 State$2 0.9 0.1 ! ! " State$3 0.4 0.6 State$3 0.2 0.8 State$3 0.6 0.4 ! ! " State$4 0.0 1.0$ State$4 0.3 0.7 State$4 0.0 1.0$ ! ! ! • CMDP$states$1$and$2$encode$very$similar$policies,$and$should$be$close$in$ CMDP$representation$space$$

  16. Example:(Discrete(Representations State'2 State'1 Normalized Normalized Q(State52,5Left) Q(State51,5Left) Normalized Normalized Q(State52,5Right) Q(State51,5Right) • One$approach:$use$tile$coding$ • Create$a$separate$tiling$on$a$stateObyOstate$level • When$comparing$CMDP$states,$the$more$similar$the$policies are$in$a$ primitive$state,$the$more$common$tiles$will$be$activated • Each$primitive$state$contributes$equally$towards$the$similarity$of$the$ CMDP$state University$of$Texas$at$Austin Sanmit$Narvekar 16

  17. Continuous(CMDP(Representations • In$continuous$domains,$weights$ are$not$local$to$a$state • Needs$to$be$done$separately$for$ each$domain • Neural$networks • Tile$coding • Etc… • If$the$base$agent$uses$a$linear$ function$approximator,$one$can$ use$tile$coding$over$the$ parameters$as$before University$of$Texas$at$Austin Sanmit$Narvekar 17

  18. Changes(in(Transfer(Algorithm M 3 " 1 R 1,3 M 1 " 4 M 4 R 0,1 R 4,4 M 4 M 3 " f " 2 R 2,4 " 0 R 0,3 M 4 R 5,4 M 2 " 5 M 3 R 0,2 R 3,3 " 3 • Transfer$method$directly$affects$CMDP$state$representation$and$transition$ function • CMDP$states$represent$“states$of$knowledge,”$where$knowledge$represented$as$ VF,$shaping$reward,$etc.$ • Similar$process$can$be$done$if$knowledge$parameterizable University$of$Texas$at$Austin Sanmit$Narvekar 18

  19. Experimental(Results • Evaluate$whether$curriculum$ policies$can$be$learned • Grid$world • Multiple$base$agents • Multiple$CMDP$state$ representations • Pacman • Multiple$transfer$learning$ algorithms • How$long$to$train$on$sources? University$of$Texas$at$Austin Sanmit$Narvekar 19

  20. Grid(world(Setup Agent'Types • Basic$Agent • State:$Sensors$on$4$sides$that$measure$distance$to$keys,$locks,$etc. • Actions:$Move$in$4$directions,$pickup$key,$unlock$lock • ActionOdependent$Agent$ • State$difference:$weights on$features$are$shared over$4$directions • Rope$Agent • Action$difference:$Like$basic,$but$can$use$rope$action$to$negate$a$pit CMDP'Representations • Finite$State$Representation • For$discrete$domains,$groups$and$normalizes$raw$weights$stateObyOstate$to$form$CMDP$features • Continuous$State$Representation • Directly$uses$raw$weights$of$learning$agent$as$features$for$CMDP$agent University$of$Texas$at$Austin Sanmit$Narvekar 20

  21. Basic(Agent(Results University$of$Texas$at$Austin Sanmit$Narvekar 21

  22. ActionIDependent(Agent(Results University$of$Texas$at$Austin Sanmit$Narvekar 22

  23. Rope(Agent(Results University$of$Texas$at$Austin Sanmit$Narvekar 23

  24. Pacman(Setup Agent'Representation • ActionOdependent$egocentric$features CMDP'Representation • Continuous$State$Representation • Directly$uses$raw$weights$of$learning$agent$as$features$for$CMDP$agent Transfer'Methods • Value$Function$Transfer • Reward$Shaping$Transfer How'long'to'train'on'a'source'task? University$of$Texas$at$Austin Sanmit$Narvekar 24

  25. Pacman(Value(Function(Transfer − 50000 Cost to Learn Target Task − 100000 − 150000 no curriculum − 200000 continuous state representation naive length 1 representation naive length 2 representation − 250000 0 100 200 300 400 500 600 700 CMDP Episodes University$of$Texas$at$Austin Sanmit$Narvekar 25

  26. Pacman(Reward(Shaping(Transfer 0 − 500 Cost to Learn Target Task − 1000 − 1500 − 2000 − 2500 no curriculum Svetlik et al. (2017) − 3000 continuous state representation naive length 2 representation − 3500 0 100 200 300 400 500 600 700 CMDP Episodes University$of$Texas$at$Austin Sanmit$Narvekar 26

  27. How(long(to(train? 0 Cost to Learn Target Task -100000 -200000 -300000 reward shaping (return-based) reward shaping (small fixed) -400000 value function (return-based) value function (small fixed) -500000 0 100 200 300 400 500 CMDP Episodes University$of$Texas$at$Austin Sanmit$Narvekar 27

  28. Related(Work Restrictions'on'source'tasks • Florensa et$al.$2018,$Riedmiller et$al.$2018,$Sukhbaatar et$al.$2017 Heuristic'based'sequencing • Da$Silva$et$al.$2018,$Svetlik et$al.$2017 MDP/POMDP'based'sequencing • Matiisen et$al.$2017,$Narvekar$et$al.$2017$ CL'for'supervised'learning • Bengio et$al.$2009,$Fan$et$al.$2018,$Graves$et$al.$2017 University$of$Texas$at$Austin Sanmit$Narvekar 28

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Automated Curriculum Learning for Reinforcement Learning Feryal Behbahani Jeju Deep Learning

Automated Curriculum Learning for Reinforcement Learning Feryal Behbahani Jeju Deep Learning

Automated Curriculum Learning for Reinforcement Learning Feryal Behbahani Jeju Deep Learning Camp 2018 Shape sorter? Simple children toy: put shapes in the correct holes Trivial for adults Yet children cannot fully solve until 2

606 views • 31 slides
CS885 Reinforcement Learning Module 2: June 6, 2020 Maximum Entropy Reinforcement Learning

CS885 Reinforcement Learning Module 2: June 6, 2020 Maximum Entropy Reinforcement Learning

CS885 Reinforcement Learning Module 2: June 6, 2020 Maximum Entropy Reinforcement Learning Haarnoja, Tang et al. (2017) Reinforcement Learning with Deep Energy Based Policies, ICML . Haarnoja, Zhou et al. (2018) Soft Actor-Critic: Off-Policy

682 views • 24 slides
Reinforcement Learning Framework Reinforcement Learning Rewards, Returns Lectures 4 and 5

Reinforcement Learning Framework Reinforcement Learning Rewards, Returns Lectures 4 and 5

1 Reinforcement Learning Framework Reinforcement Learning Rewards, Returns Lectures 4 and 5 Environment Dynamics Gillian Hayes Components of a Problem 18th January 2007 Values and Action Values, V and Q Optimal Policies

109 views • 7 slides
RL Overview of topics About Reinforcement Learning The Reinforcement Learning Problem

RL Overview of topics About Reinforcement Learning The Reinforcement Learning Problem

Introduction to Reinforcement Learning RL Overview of topics About Reinforcement Learning The Reinforcement Learning Problem Inside an RL agent Temporal difference learning Many faces of Reinforcement Learning What is

551 views • 35 slides
Learning Fair Policies in Multiobjective (Deep) Reinforcement Learning with Average and Discounted

Learning Fair Policies in Multiobjective (Deep) Reinforcement Learning with Average and Discounted

Learning Fair Policies in Multiobjective (Deep) Reinforcement Learning with Average and Discounted Rewards Umer Siddique , Paul Weng, and Matthieu Zimmer University of Michigan-Shanghai Jiao Tong University Joint Institute ICML 2020 U. Siddique,

516 views • 28 slides
Reinforcement Learning Reinforcement Learning Reinforcement Learning in a nutshell g Imagine

Reinforcement Learning Reinforcement Learning Reinforcement Learning in a nutshell g Imagine

Reinforcement Learning Reinforcement Learning Reinforcement Learning in a nutshell g Imagine playing a new game whose rules you dont know; after a hundred or so moves your don t know; after a hundred or so moves, your opponent announces, You

511 views • 30 slides
Reinforcement Learning You can think of supervised learning as the teacher providing answers

Reinforcement Learning You can think of supervised learning as the teacher providing answers

Reinforcement Learning You can think of supervised learning as the teacher providing answers (the class labels) Reinforcement learning: In reinforcement learning, the agent learns Markov Decision Processes based on a punishment/reward

89 views • 6 slides
Deep he(a)p, big feat arXiv:1707.06887 A Distributional Perspective on Reinforcement Learning

Deep he(a)p, big feat arXiv:1707.06887 A Distributional Perspective on Reinforcement Learning

Deep he(a)p, big feat arXiv:1707.06887 A Distributional Perspective on Reinforcement Learning arXiv:1702.08165 Reinforcement Learning with Deep Energy-Based Policies 1 / 25 Reinforcement Learning Environment Action Reward Interpreter State

531 views • 25 slides
Examples of Reinforcement Learning Robocup Soccer Teams Stone & Veloso, Reidmiller et al.

Examples of Reinforcement Learning Robocup Soccer Teams Stone & Veloso, Reidmiller et al.

Reinforcement Learning Slides by Rich Sutton Mods by Dan Lizotte Refer to Reinforcement Learning: An Introduction by Sutton and Barto Alpaydin Chapter 16 Up until now we have been Supervised Learning Classifying, mostly

891 views • 31 slides
Module 11 Introduction to Reinforcement Learning CS 886 Sequential Decision Making and

Module 11 Introduction to Reinforcement Learning CS 886 Sequential Decision Making and

Module 11 Introduction to Reinforcement Learning CS 886 Sequential Decision Making and Reinforcement Learning University of Waterloo Machine Learning Supervised Learning Teacher tells learner what to remember Reinforcement Learning

401 views • 29 slides
Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree

Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree

Deep Reinforcement Learning [Mastering the Game of Go with Deep Reinforcement Learning and Tree Search, Nature 2016] CS 486/686 University of Waterloo Lecture 21: July 12, 2017 Outline AlphaGo Supervised Learning of Policy Networks

540 views • 15 slides
Reinforcement Learning Reinforcement Learning Now that you know a little about Optimal Control

Reinforcement Learning Reinforcement Learning Now that you know a little about Optimal Control

Reinforcement Learning Reinforcement Learning Now that you know a little about Optimal Control Theory, you actually have some knowledge in RL. RL shares the overall goal with OCT: solving for a control policy such that the cumulative cost

610 views • 22 slides
Reinforcement Learning (RL) CE-717: Machine Learning Sharif University of Technology M.

Reinforcement Learning (RL) CE-717: Machine Learning Sharif University of Technology M.

Reinforcement Learning (RL) CE-717: Machine Learning Sharif University of Technology M. Soleymani Fall 2018 Most slides have been taken from Klein and Abdeel, CS188, UC Berkeley. Reinforcement Learning (RL) Learning as a result of

1.45k views • 133 slides
Reinforcement Learning Stephen D. Scott (Adapted from Tom Mitchells slides) 1 Outline

Reinforcement Learning Stephen D. Scott (Adapted from Tom Mitchells slides) 1 Outline

Reinforcement Learning Stephen D. Scott (Adapted from Tom Mitchells slides) 1 Outline Control learning Control policies that choose optimal actions Q learning Convergence Temporal difference learning 2 Control Learning

576 views • 20 slides
Reinforcement Learning CS 4100: Artificial Intelligence Reinforcement Learning II Still

Reinforcement Learning CS 4100: Artificial Intelligence Reinforcement Learning II Still

Reinforcement Learning CS 4100: Artificial Intelligence Reinforcement Learning II Still assume a Marko kov decision process (MDP): A se set o of st states s s S A se set o of a actions ( s (per st state) A A mo model

96 views • 5 slides
R i f R i f Reinforcement Learning III Reinforcement Learning III t L t L i i III III Dec

R i f R i f Reinforcement Learning III Reinforcement Learning III t L t L i i III III Dec

R i f R i f Reinforcement Learning III Reinforcement Learning III t L t L i i III III Dec 03 2008 1 Large State Spaces h When a problem has a large state space we can not longer represent the U or Q functions as explicit tables explicit

436 views • 18 slides
Neural Networks for Machine Learning Lecture 7a Modeling sequences: A brief overview Geoffrey

Neural Networks for Machine Learning Lecture 7a Modeling sequences: A brief overview Geoffrey

Neural Networks for Machine Learning Lecture 7a Modeling sequences: A brief overview Geoffrey Hinton Nitish Srivastava, Kevin Swersky Tijmen Tieleman Abdel-rahman Mohamed Getting targets when modeling sequences When applying machine

719 views • 34 slides
1 Contact Analysis Contact Analysis How would you compute a direction of motion for the

1 Contact Analysis Contact Analysis How would you compute a direction of motion for the

Assembly Planning Levels of Problems Parts are assumed free-flying Assembly sequence planning Tools/fixtures are taken into account Entire manipulation system is taken into account Manipulation planning Assembly Sequence

837 views • 6 slides
DNA Mo'f Discovery COMPSCI 260 Spring 2016 DNA motif discovery

DNA Mo'f Discovery COMPSCI 260 Spring 2016 DNA motif discovery

DNA Mo'f Discovery COMPSCI 260 Spring 2016 DNA motif discovery Input: X 1 A set of DNA sequences X 2 bound by the same TF X 3 Upstream regions of co- regulated genes X 4 Sequences bound in a

418 views • 29 slides
Sustainable Use of Risk-Informed Regulation to Improve Plant Safety Nuclear Regulatory Commission

Sustainable Use of Risk-Informed Regulation to Improve Plant Safety Nuclear Regulatory Commission

Sustainable Use of Risk-Informed Regulation to Improve Plant Safety Nuclear Regulatory Commission Briefing May 11, 2017 1 Risk-Informed Regulation in the U.S. Nuclear Industry Purpose of Probabilistic Risk Assessment (PRA) - Provide

484 views • 23 slides
DNA sequencing applica0ons: iden0fying gene0c varia0on Short sequencing

DNA sequencing applica0ons: iden0fying gene0c varia0on Short sequencing

DNA sequencing applica0ons: iden0fying gene0c varia0on Short sequencing reads from an individual Sequence from the reference genome Short read alignment problem INPUT: A few million

343 views • 13 slides
Learning meets Sequencing: a Generality Framework for Read-Sets Filip Zelezn y, Karel

Learning meets Sequencing: a Generality Framework for Read-Sets Filip Zelezn y, Karel

Learning meets Sequencing: a Generality Framework for Read-Sets Filip Zelezn y, Karel Jalovec, Jakub Tolar Czech Technical University in Prague University of Minnesota Zelezn y, Jalovec, Tolar (CTU Prague) Learning Meets

399 views • 26 slides
Chapter 7: Rapid alignment methods: FASTA and BLAST The biological problem l Search strategies

Chapter 7: Rapid alignment methods: FASTA and BLAST The biological problem l Search strategies

Chapter 7: Rapid alignment methods: FASTA and BLAST The biological problem l Search strategies l FASTA l BLAST l Introduction to bioinformatics, Autumn 2007 83 The biological problem Global and local alignment algoritms are slow in

600 views • 34 slides
Sequencing, and I/O Bjrn Lisper School of Innovation, Design, and Engineering Mlardalen

Sequencing, and I/O Bjrn Lisper School of Innovation, Design, and Engineering Mlardalen

Sequencing, and I/O Bjrn Lisper School of Innovation, Design, and Engineering Mlardalen University bjorn.lisper@mdh.se http://www.idt.mdh.se/blr/ Sequencing, and I/O (revised 2019-02-03) Sequencing in F# We said functional programming

778 views • 30 slides

More recommend