๐Ÿ˜‡
Deep Multi-Agent Reinforcement Learning
  • Deep Multi-Agent Reinforcement Learning
  • Abstract & Contents
    • Abstract
  • 1. Introduction
    • 1. INTRODUCTION
      • 1.1 The Industrial Revolution, Cognition, and Computers
      • 1.2 Deep Multi-Agent Reinforcement-Learning
      • 1.3 Overall Structure
  • 2. Background
    • 2. BACKGROUND
      • 2.1 Reinforcement Learning
      • 2.2 Multi-Agent Settings
      • 2.3 Centralized vs Decentralized Control
      • 2.4 Cooperative, Zero-sum, and General-Sum
      • 2.5 Partial Observability
      • 2.6 Centralized Training, Decentralized Execution
      • 2.7 Value Functions
      • 2.8 Nash Equilibria
      • 2.9 Deep Learning for MARL
      • 2.10 Q-Learning and DQN
      • 2.11 Reinforce and Actor-Critic
  • I Learning to Collaborate
    • 3. Counterfactual Multi-Agent Policy Gradients
      • 3.1 Introduction
      • 3.2 Related Work
      • 3.3 Multi-Agent StarCraft Micromanagement
      • 3.4 Methods
        • 3.4.1 Independent Actor-Critic
        • 3.4.2 Counterfactual Multi-Agent Policy Gradients
        • 3.4.2.1 baseline lemma
        • 3.4.2.2 COMA Algorithm
      • 3.5 Results
      • 3.6 Conclusions & Future Work
    • 4 Multi-Agent Common Knowledge Reinforcement Learning
      • 4.1 Introduction
      • 4.2 Related Work
      • 4.3 Dec-POMDP and Features
      • 4.4 Common Knowledge
      • 4.5 Multi-Agent Common Knowledge Reinforcement Learning
      • 4.6 Pairwise MACKRL
      • 4.7 Experiments and Results
      • 4.8 Conclusion & Future Work
    • 5 Stabilizing Experience Replay
      • 5.1 Introduction
      • 5.2 Related Work
      • 5.3 Methods
        • 5.3.1 Multi-Agent Importance Sampling
        • 5.3.2 Multi-Agent Fingerprints
      • 5.4 Experiments
        • 5.4.1 Architecture
      • 5.5 Results
        • 5.5.1 Importance Sampling
        • 5.5.2 Fingerprints
        • 5.5.3 Informative Trajectories
      • 5.6 Conclusion & Future Work
  • II Learning to Communicate
    • 6. Learning to Communicate with Deep Multi-Agent ReinforcementLearning
      • 6.1 Introduction
      • 6.2 Related Work
      • 6.3 Setting
      • 6.4 Methods
        • 6.4.1 Reinforced Inter-Agent Learning
        • 6.4.2 Differentiable Inter-Agent Learning
      • 6.5 DIAL Details
      • 6.6 Experiments
        • 6.6.1 Model Architecture
        • 6.6.2 Switch Riddle
        • 6.6.3 MNIST Games
        • 6.6.4 Effect of Channel Noise
      • 6.7 Conclusion & Future Work
    • 7. Bayesian Action Decoder
      • 7.1 Introduction
      • 7.2 Setting
      • 7.3 Method
        • 7.3.1 Public belief
        • 7.3.2 Public Belief MDP
        • 7.3.3 Sampling Deterministic Partial Policies
        • 7.3.4 Factorized Belief Updates
        • 7.3.5 Self-Consistent Beliefs
      • 7.4 Experiments and Results
        • 7.4.1 Matrix Game
        • 7.4.2 Hanabi
        • 7.4.3 Observations and Actions
        • 7.4.4 Beliefs in Hanabi
        • 7.4.5 Architecture Details for Baselines and Method
        • 7.4.6 Hyperparamters
        • 7.4.7 Results on Hanabi
      • 7.5 Related Work
        • 7.5.1 Learning to Communicate
        • 7.5.2 Research on Hanabi
        • 7.5.3 Belief State Methods
      • 7.6 Conclusion & Future Work
  • III Learning to Reciprocate
    • 8. Learning with Opponent-Learning Awareness
      • 8.1 Introduction
      • 8.2 Related Work
      • 8.3 Methods
        • 8.3.1 Naive Learner
        • 8.3.2 Learning with Opponent Learning Awareness
        • 8.3.3. Learning via Policy gradient
        • 8.3.4 LOLA with Opponent modeling
        • 8.3.5 Higher-Order LOLA
      • 8.4 Experimental Setup
        • 8.4.1 Iterated Games
        • 8.4.2 Coin Game
        • 8.4.3 Training Details
      • 8.5 Results
        • 8.5.1 Iterated Games
        • 8.5.2 Coin Game
        • 8.5.3 Exploitability of LOLA
      • 8.6 Conclusion & Future Work
    • 9. DiCE: The Infinitely Differentiable Monte Carlo Estimator
      • 9.1 Introduction
      • 9.2 Background
        • 9.2.1 Stochastic Computation Graphs
        • 9.2.2 Surrogate Losses
      • 9.3 Higher Order Gradients
        • 9.3.1 Higher Order Gradient Estimators
        • 9.3.2 Higher Order Surrogate Losses
        • 9.3.3. Simple Failing Example
      • 9.4 Correct Gradient Estimators with DiCE
        • 9.4.1 Implement of DiCE
        • 9.4.2 Casuality
        • 9.4.3 First Order Variance Reduction
        • 9.4.4 Hessian-Vector Product
      • 9.5 Case Studies
        • 9.5.1 Empirical Verification
        • 9.5.2 DiCE For multi-agent RL
      • 9.6 Related Work
      • 9.7 Conclusion & Future Work
  • Reference
    • Reference
  • After
    • ๋ณด์ถฉ
    • ์—ญ์ž ํ›„๊ธฐ
Powered by GitBook
On this page

Was this helpful?

  1. I Learning to Collaborate
  2. 5 Stabilizing Experience Replay

5.1 Introduction

Chapter 3,4๋ฅผ ํ†ตํ•ด on-policy MARL์— ๋Œ€ํ•ด ๋ฐฐ์›Œ๋ณด์•˜์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ RL์ž์ฒด์—์„œ ํ•ด๊ฒฐํ•ด์•ผํ•˜๋Š” ํ•„์—ฐ์ ์ธ ๋ฌธ์ œ์ค‘ ํ•˜๋‚˜๋Š” Sample efficiency์ž…๋‹ˆ๋‹ค. ์ด ๋•Œ, on-policy๋Š” ํ•„์—ฐ์ ์œผ๋กœ ๊ฐ™์€ ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•ด ์—ฌ๋Ÿฌ๋ฒˆ ํ•™์Šตํ•˜๊ฑฐ๋‚˜, ์—ฌ๋Ÿฌ๊ฐ€์ง€ policy๋กœ ๋ถ€ํ„ฐ ๋ฐฐ์šธ ์ˆ˜ ์žˆ๋Š” off-policy๋ณด๋‹ค Sample-efficiency๊ฐ€ ๋‚ฎ์„ ์ˆ˜ ๋ฐ–์— ์—†๊ณ  ์ด๋Š” MARL์—์„œ์˜ off-policy๋ฅผ ํ™œ์šฉํ•˜๋Š” ๋ฐฉ๋ฉด์œผ๋กœ ์—ฐ๊ตฌ๋ฅผ ์ผ์œผํ‚ฌ ์ˆ˜๋ฐ–์— ์—†์—ˆ์Šต๋‹ˆ๋‹ค.

off-policy์˜ ๋Œ€ํ‘œ์ ์ธ algorithm์ธ DQN์˜ MARL์— ๋Œ€ํ•œ ์ ์šฉ์€ IQL์ž…๋‹ˆ๋‹ค. ์ด ๋•Œ, ํ™˜๊ฒฝ์— ์กด์žฌํ•˜๋Š” ๋‹ค๋ฅธ agent๋“ค์„ ๋ชจ๋‘ ์ •์ ์ธ ์กด์žฌ๋กœ ์ทจ๊ธ‰ํ•ด ํ•ด๊ฒฐ์„ ํ•˜๊ธฐ ๋•Œ๋ฌธ์—, ์ˆ˜๋ ด์„ ๋ณด์žฅํ•  ์ˆ˜๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ๋‹คํ–‰ํžˆ๋„ ์‹ค์ „์ ์œผ๋กœ ๋ช‡๊ฐœ์˜ ์‹คํ—˜์— ๋Œ€ํ•ด์„œ๋Š” IQL์ด ๊ฝค ๊ดœ์ฐฎ์€ ์„ฑ๋Šฅ์„ ๋ณด์ž„์„ ๋ณด์•˜์Šต๋‹ˆ๋‹ค.

RL์—์„œ์˜ ํฐ ๋ฐœ์ „์„ ์ด๋Œ์—ˆ๋˜ ์š”์†Œ์ค‘ ํ•˜๋‚˜์— Replay memory๋ฅผ ๋นผ๋†“์„ ์ˆ˜ ์—†๋Š”๋ฐ, ์ด๋Š” data๋ฅผ iid๋กœ ๋งŒ๋“ค์–ด Neural Network์˜ ํ•™์Šต์•ˆ์ •์„ฑ์— ๋„์›€์„ ์ค„ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ efficiency๋„ ๋†’์—ฌ์ค๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ด๋Ÿฌํ•œ Replay Memory๋„ IQL์— ์ ์šฉํ•˜๊ธฐ์—๋Š” ๋ฌธ์ œ์ ์ด ์žˆ์Šต๋‹ˆ๋‹ค. MARL์ƒํ™ฉ์—์„œ์˜ Replay Memory๋‚ด์˜ data๋“ค์€ ํ˜„์žฌํ™˜๊ฒฝ์˜ dynamic์„ ํ‘œํ˜„ํ•˜๊ธฐ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ๋‹ค๋ฅธ agent๋“ค์— ์˜ํ•ด non-stationaryํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ IQL์„ ๋ณด๋ฉด ์ ์ง„์ ์œผ๋กœ ์–ด๋–ป๊ฒŒ๋“  ๋ฐฐ์šฐ๋Š” ๊ฒฝํ–ฅ์€ ์žˆ์œผ๋‚˜, non-stationaryํ•œ data๋ฅผ ๊ณ„์† samplingํ•ด ๊ทธ๋ƒฅ ์—…๋ฐ์ดํŠธํ•˜๋Š” ํ–‰์œ„๋Š” ๊ฒฐ๊ณผ์ ์œผ๋กœ agent์˜ ํ•™์Šต์— ํฐ ์žฅ์• ๋ฌผ์ด ์•„๋‹ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค.

์ด๋Ÿฌํ•œ ๋ฌธ์ œ์ ์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ์ด์ „์—” Replay Memory ํฌ๊ธฐ๋ฅผ ์ž‘๊ฒŒ ์œ ์ง€ํ•ด ์ตœ๊ทผ์˜ ๋ฐ์ดํ„ฐ๋งŒ ์‚ฌ์šฉํ•˜๋Š”๋“ฑ sample efficiency๋ฅผ ๋‚ฎ์ถ”๊ณ , ๊ทผ๋ณธ์ ์œผ๋กœ MARL์˜ stability๋ฅผ ์œ ์ง€ํ•˜๋ฉฐ ํ•™์Šตํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด์„œ๋Š” ์„ค๋ช…ํ•˜์ง€ ๋ชปํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฒฐ๊ตญ IQL์—์„œ์˜ Replay Memory ์ ์šฉ์„ ์–ด๋–ป๊ฒŒ ์‹œํ‚ฌ์ง€๊ฐ€ ๋˜ ํ•ด๊ฒฐํ•ด์•ผํ•  ์–ด๋ ค์šด ๋ฌธ์ œ๋กœ ๋‚จ๊ฒŒ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

์ด chapter์—์„œ๋Š” Replay Memory๋ฅผ MARL์— ์ ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๋‘๊ฐ€์ง€ ํšจ๊ณผ์ ์ธ ๋ฐฉ๋ฒ•์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

  • ์ฒซ์งธ๋กœ, Replay Memory๋‚ด์˜ data๋ฅผ off-environment data๋กœ ์ทจ๊ธ‰ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. off policy์—์„œ๋Š” policy์— ์˜ํ•ด ๋“ฑ์žฅํ•˜๋Š” state distribution์˜ ์ฐจ์ด ๋•Œ๋ฌธ์— Importance Sampling์„ ์‚ฌ์šฉํ–ˆ๋‹ค๋ฉด, ์ด๋ฒˆ์—๋Š” agent ์ž…์žฅ์—์„œ์˜ ๋‹ค๋ฅธ agent๋“ค์˜ joint action์— ๋Œ€ํ•ด distribution์ด ๋‹ฌ๋ผ์ ธ ๊ทธ์— ๋Œ€ํ•œ Importance Sampling์„ ์ง„ํ–‰ํ•ฉ๋‹ˆ๋‹ค.

  • ๋‘˜์งธ๋กœ, Hyper Q-learning์— ์˜ํ•ด ์˜๊ฐ์„ ๋ฐ›์€ ์ ‘๊ทผ๋ฒ•์„ ์†Œ๊ฐœํ•˜๋Š”๋ฐ, ์ด๋Š” ๊ฐ agent๊ฐ€ ๋‹ค๋ฅธ agent์˜ policy๋“ค์„ ๊ด€์ฐฐํ•˜๋ฉฐ ์ถ”์ •ํ•˜์—ฌ non-stationary๋ฅผ ํ”ผํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด์— Q-function์˜ space๊ฐ€ ์ปค์งˆ ๋•Œ ์ด๋ฅผ ๊ฐ๋‹นํ•  ์ˆ˜ ์—†๋Š”๋ฐ, ์—ฌ๊ธฐ์„œ๋Š” ์ž‘์€ ์ฐจ์›์˜ fingerprint๋ฅผ ํ†ตํ•ด ์ด์ „์˜ ํ•œ๊ณ„๋ฅผ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ์ด๋Ÿฌํ•œ ๋ฐฉ๋ฒ•๋“ค์— ๋Œ€ํ•ด Starcraft unit micromanagement ํ™˜๊ฒฝ์—์„œ ์„ฑ๊ณต์ ์ธ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

Previous5 Stabilizing Experience ReplayNext5.2 Related Work

Last updated 4 years ago

Was this helpful?