Multi-agent Systems & Reinforcement Learning - PowerPoint PPT Presentation

multi agent systems reinforcement learning n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
Multi-agent Systems & Reinforcement Learning PowerPoint Presentation
Download Presentation
Multi-agent Systems & Reinforcement Learning

play fullscreen
1 / 10
Multi-agent Systems & Reinforcement Learning
203 Views
Download Presentation
wendi
Download Presentation

Multi-agent Systems & Reinforcement Learning

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. Multi-agent Systems &Reinforcement Learning A Presentation

  2. What • Artificial Intelligence -> Distributed Artificial Intelligence • Concerned with information management issues and distributed/parallel problem solving • Distributed Artificial Intelligence -> Multi-agent Systems • Different problem solving agents with their own interests and goals

  3. Why • Some of the trends in computing • Ubiquity, interconnection, intelligence, delegation. • The Internet of Things,self-steering cars, home automation devices. • What advantages does it offer over the alternatives? • In what circumstances is it useful?

  4. Answers • Parallelism • Robustness • Fault-tolerance • Scalability • Simpler programming • Not for situations where parallel action is not possible and there is no action uncertainty.

  5. Multi-Agent Systems • Two main dimensions: • Agent Heterogeneity • Amount of communication among agents • Multi-agent scenarios • Homogeneous non-communicating agents • Heterogeneous non-communicating agents • Homogeneous communicating agents • Heterogeneous communicating agents

  6. The Predator/Prey (“Pursuit”) Domain

  7. Homogeneous Non-Communicating • Issues • Reactive vs. Deliberative agents • Local vs. Global perspective • Modeling other agents’ states • How to affect others • Techniques • Reactive behaviors for formation maintenance • Local knowledge sometimes better • Recursive Modeling Method • Don’t model others – Just pay attention to reward • Stigmergy

  8. Heterogeneous Non-Communicating • Issues • Benevolence vs. Competitiveness • Stable vs. evolving (arms race, credit/blame) • Modeling of others’ goals, actions, and knowledge • Social conventions • Roles • Techniques • Game theory, iterative play • Minimax-Q • Competitive co-evolution • Deduce intentions through observation • Autoepistemic reasoning (ignorance) • Model as a team (individuals follow roles) • Focal points/Emergent conventions • Design agents play different roles

  9. Heterogeneous Communicating • Issues • Understanding each other • Planning communicative acts • Benevolence vs. competitiveness • Commitment/decommitment • Truth in communication • Techniques • Language Protocols: CL, ACL, KQML • Speech acts • Learning social behaviors • Multi-agent Q-learning • Training other agents’ Q-functions • Contract nets for electronic commerce/Market-based systems • Belief/Desire/Intention (BDI) models • Coalitions • Reasoning about truthfulness

  10. The End • Thanks for listening • I skipped a lot of material • Multiagent Systems: A Survey from a Machine Learning Perspective, Peter Stone and Manuela Veloso, December 4, 1997 • No programming segment • Questions time