Agent-Organized Networks for Dynamic Team Formation Multi-Agent Planning and Learning Laboratory (MAPLE) Department of Computer Science and EE University.

Slides:



Advertisements
Similar presentations
Supporting Cooperative Caching in Disruption Tolerant Networks
Advertisements

Evolving Cooperation in the N-player Prisoner's Dilemma: A Social Network Model Dept Computer Science and Software Engineering Golriz Rezaei Michael Kirley.
Distributed Advice-Seeking on an Evolving Social Network Dept Computer Science and Software Engineering The University of Melbourne - Australia Golriz.
Scalable and Dynamic Quorum Systems Moni Naor & Udi Wieder The Weizmann Institute of Science.
Modeling Maze Navigation Consider the case of a stationary robot and a mobile robot moving towards a goal in a maze. We can model the utility of sharing.
Maximum Battery Life Routing to Support Ubiquitous Mobile Computing in Wireless Ad Hoc Networks By C. K. Toh.
Mauro Sozio and Aristides Gionis Presented By:
1 Regret-based Incremental Partial Revelation Mechanism Design Nathanaël Hyafil, Craig Boutilier AAAI 2006 Department of Computer Science University of.
Methods for Coalition Formation in Adaptation-Based Social Networks Levi Barton, Vicki H. Allan Utah State University.
1 Sensor Relocation in Mobile Sensor Networks Guiling Wang, Guohong Cao, Tom La Porta, and Wensheng Zhang Department of Computer Science & Engineering.
Game-Theoretic Approaches to Multi-Agent Systems Bernhard Nebel.
Effective Coordination of Multiple Intelligent Agents for Command and Control The Robotics Institute Carnegie Mellon University PI: Katia Sycara
AAMAS 2009, Budapest1 Analyzing the Performance of Randomized Information Sharing Prasanna Velagapudi, Katia Sycara and Paul Scerri Robotics Institute,
1 Resource Management in IP Telephony Networks Matthew Caesar, Dipak Ghosal, Randy H. Katz {mccaesar,
Localized Techniques for Power Minimization and Information Gathering in Sensor Networks EE249 Final Presentation David Tong Nguyen Abhijit Davare Mentor:
Beneficial Caching in Mobile Ad Hoc Networks Bin Tang, Samir Das, Himanshu Gupta Computer Science Department Stony Brook University.
ICNP'061 Benefit-based Data Caching in Ad Hoc Networks Bin Tang, Himanshu Gupta and Samir Das Department of Computer Science Stony Brook University.
Collaboration: Software Development, Learning James Chisan February, 2003.
Distributed Rational Decision Making Sections By Tibor Moldovan.
Social Networks: Advertising, Pricing and All That Zvi Topol & Itai Yarom.
Online Data Gathering for Maximizing Network Lifetime in Sensor Networks IEEE transactions on Mobile Computing Weifa Liang, YuZhen Liu.
© 2005 Prentice-Hall 8-1 Understanding Work Teams Chapter 8 Essentials of Organizational Behavior, 8/e Stephen P. Robbins.
1 Characterizing Selfishly Constructed Overlay Routing Networks March 11, 2004 Byung-Gon Chun, Rodrigo Fonseca, Ion Stoica, and John Kubiatowicz University.
Agent-Organized Networks for Dynamic Team Formation Gaton, M.E. and desJardins, M., In Proceedings of AAMAS-2005, pp Seo, Young-Woo.
On Distinguishing between Internet Power Law B Bu and Towsley Infocom 2002 Presented by.
Establishing Pairwise Keys in Distributed Sensor Networks Donggang Liu, Peng Ning Jason Buckingham CSCI 7143: Secure Sensor Networks October 12, 2004.
1 Quality of Experience Control Strategies for Scalable Video Processing Wim Verhaegh, Clemens Wüst, Reinder J. Bril, Christian Hentschel, Liesbeth Steffens.
Summary from Previous Lecture Real networks: –AS-level N= 12709, M=27384 (Jan 02 data) route-views.oregon-ix.net, hhtp://abroude.ripe.net/ris/rawdata –
Intelligent Agents: an Overview. 2 Definitions Rational behavior: to achieve a goal minimizing the cost and maximizing the satisfaction. Rational agent:
Algorithms for Self-Organization and Adaptive Service Placement in Dynamic Distributed Systems Artur Andrzejak, Sven Graupner,Vadim Kotov, Holger Trinks.
Cache Updates in a Peer-to-Peer Network of Mobile Agents Elias Leontiadis Vassilios V. Dimakopoulos Evaggelia Pitoura Department of Computer Science University.
Optimization Based Modeling of Social Network Yong-Yeol Ahn, Hawoong Jeong.
(Social) Networks Analysis III Prof. Dr. Daning Hu Department of Informatics University of Zurich Oct 16th, 2012.
Modeling Information Diffusion in Networks with Unobserved Links Quang Duong Michael P. Wellman Satinder Singh Computer Science and Engineering University.
IEEE P2P, Aachen, Germany, September Ad-hoc Limited Scale-Free Models for Unstructured Peer-to-Peer Networks Hasan Guclu
Collectively Cognitive Agents in Cooperative Teams Jacek Brzeziński, Piotr Dunin-Kęplicz Institute of Computer Science, Polish Academy of Sciences Barbara.
Surface Simplification Using Quadric Error Metrics Michael Garland Paul S. Heckbert.
Trust-based Multi-Objective Optimization for Node-to-Task Assignment in Coalition Networks 1 Jin-Hee Cho, Ing-Ray Chen, Yating Wang, and Kevin S. Chan.
NOBEL WP Szept Stockholm Game Theory in Inter-domain Routing LÓJA Krisztina - SZIGETI János - CINKLER Tibor BME TMIT Budapest,
REINFORCEMENT LEARNING LEARNING TO PERFORM BEST ACTIONS BY REWARDS Tayfun Gürel.
Introduction Many decision making problems in real life
UbiStore: Ubiquitous and Opportunistic Backup Architecture. Feiselia Tan, Sebastien Ardon, Max Ott Presented by: Zainab Aljazzaf.
Topology aggregation and Multi-constraint QoS routing Presented by Almas Ansari.
Implicit An Agent-Based Recommendation System for Web Search Presented by Shaun McQuaker Presentation based on paper Implicit:
1 Near-Optimal Play in a Social Learning Game Ryan Carr, Eric Raboin, Austin Parker, and Dana Nau Department of Computer Science, University of Maryland.
A Case Study in Understanding OSPFv2 and BGP4 Interactions Using Efficient Experiment Design David Bauer†, Murat Yuksel‡, Christopher Carothers† and Shivkumar.
CS584 - Software Multiagent Systems Lecture 12 Distributed constraint optimization II: Incomplete algorithms and recent theoretical results.
Social Network Analysis Prof. Dr. Daning Hu Department of Informatics University of Zurich Mar 5th, 2013.
Selfishness, Altruism and Message Spreading in Mobile Social Networks September 2012 In-Seok Kang
Thursday, May 9 Heuristic Search: methods for solving difficult optimization problems Handouts: Lecture Notes See the introduction to the paper.
Learning the Structure of Related Tasks Presented by Lihan He Machine Learning Reading Group Duke University 02/03/2006 A. Niculescu-Mizil, R. Caruana.
Understanding Work Teams
1 12/9/04 – Review TNO/TRAIL project #16 Jonne Zutt Delft University of Technology Information Technology and Systems Collective Agent Based Systems Group.
Algorithmic, Game-theoretic and Logical Foundations
Introduction of Intelligent Agents
KAIS T On the problem of placing Mobility Anchor Points in Wireless Mesh Networks Lei Wu & Bjorn Lanfeldt, Wireless Mesh Community Networks Workshop, 2006.
Networked Games: Coloring, Consensus and Voting Prof. Michael Kearns Networked Life NETS 112 Fall 2013.
A survey of Constraint Handling Techniques in Evolutionary Computation Methods Author: Zbigneiw Michalewicz Presenter: Masoud Mazloom 27 th Oct
Learning for Physically Diverse Robot Teams Robot Teams - Chapter 7 CS8803 Autonomous Multi-Robot Systems 10/3/02.
Multi-Agent Systems: Overview and Research Directions CMSC 471 March 11, 2014 Prof. Marie desJardins.
Evolution of Cooperation in Mobile Ad Hoc Networks Jeff Hudack (working with some Italian guy)
-1/16- Maximum Battery Life Routing to Support Ubiquitous Mobile Computing in Wireless Ad Hoc Networks C.-K. Toh, Georgia Institute of Technology IEEE.
Cmpe 588- Modeling of Internet Emergence of Scale-Free Network with Chaotic Units Pulin Gong, Cees van Leeuwen by Oya Ünlü Instructor: Haluk Bingöl.
Link-Level Internet Structures
Discrete ABC Based on Similarity for GCP
Consistency Methods for Temporal Reasoning
Multi-Agent Exploration
Towards Next Generation Panel at SAINT 2002
CASE − Cognitive Agents for Social Environments
Market-based Dynamic Task Allocation in Mobile Surveillance Systems
Presentation transcript:

Agent-Organized Networks for Dynamic Team Formation Multi-Agent Planning and Learning Laboratory (MAPLE) Department of Computer Science and EE University of Maryland Baltimore County Matt Gaston Marie desJardins

Agent-Organized Networks 2 Overview  Introduction and Motivation  Team Formation  Agent-Organized Networks  Experimental Results  Related Projects:  Connections Model AONs  AONs for Production and Exchange  Stable Team Formation  Future Work and Conclusions

Agent-Organized Networks 3 Introduction and Motivation

Agent-Organized Networks 4 Introduction: Multi-Agent Systems  Agent: Autonomous, intelligent software system.  Physical (robot, autonomous vehicle, mobile sensor) or virtual (search, travel planning, trading / e-commerce, information retrieval)  MAS: “Community” of agents – competitive or cooperative  Connections form a “social network” of agents

Agent-Organized Networks 5 Why Adapt?  Multi-agent systems are growing in popularity and size  Technologies like the Semantic Web support the deployment and evolution of large-scale, dynamic multi- agent systems  Agent cognitive capacities are limited, preventing all agents from knowing/interacting with all other agents  Previous findings suggest that network structure plays an essential role in understanding team formation dynamics in multi-agent systems  Identifying the “best” network structure is difficult or impossible to do a priori  Solution: Agent-Organized Networks

Agent-Organized Networks 6 Team Formation [AAMAS 2005]

Agent-Organized Networks 7 Multi-Agent Team Formation Model  Agents must form teams to complete tasks  Agent states:  Uncommitted  Committed  Active  Tasks are advertised to the network of agents  A valid team:  Connected path in network  Task skill requirements met  Formed within time constraints , 1, 2, 3 2, 2 1, 2, 2, 2, 4 4

Agent-Organized Networks 8 Multi-Agent Team Formation Model  Some model details  Parameters Number of agents: N Skill diversity:  Task introduction interval:  Team/task size: T Advertisement duration:  Task duration:  Network structure  Organizational Efficiency # of tasks successfully completed total # of tasks advertised efficiency =

Agent-Organized Networks 9 Team Joining Strategy With some initiation probability, start a new team if needed: Always join a team if it’s already been started, and it needs your skill. Considering each task in random order...

Agent-Organized Networks 10 Agent-Organized Networks

11 Agent-Organized Networks  Definition: An agent-organized network (AON) is an organizational network structure, or agent-to-agent interaction topology, that is the result of local rewiring decisions made by the individual agents in a networked multi-agent system.  Design considerations:  Local perception of global performance  Adaptation triggers  Rewiring strategies  Evaluation metrics:  Learning rate  Stability  Structural properties of resulting networks

Agent-Organized Networks 12 Structure-Based Adaptation  Adapt based on preferential attachment  Natural network formation process that leads to scale-free networks  Adaptation trigger (random):  Probability of adaptation for each uncommitted agent: 1/N  Rewiring strategy:  Disconnect from a random neighbor  Connect to some neighbor’s neighbor with probability

Agent-Organized Networks 13 Performance-Based Adaptation  Adaptation trigger:  Adapt if performance drops below neighbors’ average performance:  Rewiring strategy:  Drop the lowest-performing neighbor:  Add a connection to the highest-performing neighbor a k of the highest-performing neighbor a l:

Agent-Organized Networks 14 Results

Agent-Organized Networks 15 Experimental Setup  Initial network structure: Random geometric graph  Randomly place agents in a unit square  Connect agents that are closer than d units apart  Use the minimal d that guarantees all neighbors have at least one edge  Run team formation with no adaptation to establish baseline  Run with each adaptation strategy separately  Results are an average of 50 runs

Agent-Organized Networks 16 Results: Summary  Significant performance improvement (over baseline) for both AON methods

Agent-Organized Networks 17 Stability of Networks  Structure-based AONs outperform performance-based AONs, but result in substantially more rewirings  Performance-based AONs are more efficient (“better value” if adaptation cost is in similar units to performance measure)

Agent-Organized Networks 18 Evolution of the Network: Structure-Based  Converges to a network with hub structure and short average path length

Agent-Organized Networks 19 Evolution of the Network: Performance-Based  Convergence to short-average-path-length structure happens more slowly  Qualitatively similar structure to strategy-based (but in this case not by design!)

Agent-Organized Networks 20 Connections Model AONs [AAAI 2005 Workshop on Multi-Agent Learning]

Agent-Organized Networks 21 The (Symmetric) Connections Model  Symmetric when  ij =  and c ij = c for all i and j  0 <  < 1 is the value of a relationship, discounted by distance  c is the cost of a direct connection (Jackson & Wolinsky 1996; Jackson 2002)

Agent-Organized Networks 22 Dynamic Network Formation in SCM Based on pairwise stability (Watts 2001):  At each iteration:  Two agents meet (are selected) at random (synchronous)  If they have a connection, they remove the connection if at least one of them benefits -- unilateral deletion  If the do not have a connection, they add a connection if it is mutually beneficial -- bilateral creation But...

Agent-Organized Networks 23 Experiment: Watts Dynamic Network Formation  = 0.9, c = 0.8, optimal =

Agent-Organized Networks 24 A (Simple) Multi-Agent Learning Approach  Goals:  Eliminate need for “global” knowledge  Eliminate need for “global” computation  Maintain bilateral network formation (agents agree to create link)  Follow dynamic network formation process of Watts  On-line learning  Approach  Stateless Q-Learning (Claus & Boutilier 1998)  A = { add, delete, nothing }  Agents add connection if both have largest Q value for add (bilateral)  Agents remove connection if one has largest Q value for delete (unilateral)  Reinforcement signal comes from omniscient oracle (!) Agent-Organized Networks (AONs)“Distributed Annealing”

Agent-Organized Networks 25 Experiment: Learning to Form Networks Adaptive Learning Rate: Win or Lose Fast (WoLF) (Bowling & Veloso 2002)  = 0.9, c = 0.8, optimal =

Agent-Organized Networks 26 Experiment: Adding an Unselfish Agent  = 0.9, c = 0.8, optimal =

Agent-Organized Networks 27 AONs for Production and Exchange [AAAI 2005]

Agent-Organized Networks 28 A Model of Production and Exchange  n agents in an artificial economy with two goods  Each agent i possesses g 1 i units of good 1 and g 2 i units of good 2  Each agent is a producer of either good 1 or good 2  At each iteration of the model, the agents are selected in random order and choose between initiating trade with another agent or producing their respective good in order to maximize utility  Agent utility: (Wilhite 2001: 2003) fully rational behavior

Agent-Organized Networks 29 Push Referral AON Strategies  Random referral: agent selected randomly from N j (i)  Degree referral:  Production referral: Definition: Assuming that agent i is adapting its connection to agent j, a push referral is a local rewiring by i from j to an agent in N j (i)

Agent-Organized Networks 30 Results production referral degree referral random selection n = 400, q = 30,  = 0.05  =  = 0.1  =  = 0.1 initialized values to 1

Agent-Organized Networks 31 Stable Team Formation [AAAI 2004 Workshop on Team/Coalition Formation]

Agent-Organized Networks 32 Economic Model of Team Formation  Share-based scheme for pay-off distribution  Team’s revenue is stored in “team account”  Team members get shares for joining and working  Share value = team account / # outstanding shares  Agents bound to the team by a contract  Joining Shares, S join : sign-on bonus  Commission, S comm : shares given to the agent for every task completed by the team in which the agent actively participates  Dividend, S div : shares given to the agent for every task completed by the team in which the agent does not participate (Dividend < Commission)  Penalty, p : the amount to be paid to the team when leaving the team

Agent-Organized Networks 33 Results: Effect of Deadlines

Agent-Organized Networks 34 Results: Stable vs. Dynamic Agents

Agent-Organized Networks 35 Conclusions and Future Work Summary:  AONs based only on local knowledge can improve team formation in networked MAS  AON ideas can also be applied to other MAS domains and models  Stability can be achieved through a contractual model of team formation Future Work:  Quantitative analysis of post-adaptation network structures  Learning individual agent team selection strategies  [JAAMAS 2006]  Skill placement and replacement for dynamic team formation

Agent-Organized Networks 36