CSCE475/875 Multiagent Systems

Handout 15: Topics Covered

November 8, 2011

1.    Agents

·         Agent

·         Intelligent Agent

·         The five characteristics of environments for agent-based solutions

o   Complete vs. incomplete (fully vs. partially observable)

o   Certain vs. uncertain (deterministic vs. stochastic)

o   Episodic vs. non-episodic

o   Static vs. dynamic

o   Discrete vs. continuous

 

2.    Chapter 1: Distributed Constraint Satisfaction

·         Constraint satisfaction problems

·         Solution approaches

o   Least-commitment approach

o   Backtracking approach

·         Filtering algorithms

·         Variable and value ordering, minimum remaining values heuristic, degree heuristic, least-constraining-value heuristic

·         Min-conflicts algorithm

·         Relation to multiagent systems

 

3.    Chapter 2: Distributed Optimization

·         Different from constraint satisfaction – now looking for optimal solutions

·         Four general family of approaches:

o   Distributed dynamic programming

§  Asyncronous Dynamic Programming

§  Learning Real-Time A*

o   Distributed solutions to Markov Decision Problems (MDPs)

§  Action selection in MDP, using a value iteration algorithm

o   Optimization algorithms with an economic flavor (as applied to matching nad scheduling problems)

§  Contract net and auction – See later Chapter on Auctions

o   Coordination via social laws and conventions

§  Voting, social preferences

 

4.    Chapter 3: Noncooperative Game Theory

·         Self-interested agents

o   Axioms on Completeness, Transitivity, Substitutability, Decomposability, Monotonicity, Continuity, and the von Neumann and Morgenstern Theorem.

·         Games in normal form

o   Prisoner’s dilemma

o   Common-payoff games

o   Constant-sum games

·         Strategies in normal-form games

o   Pure strategy vs. mixed strategy profiles

o   Definitions for Support and Expected utility of a mixed strategy

·         Analyzing games: from optimality to equilibrium

o   The notion of an optimal strategy for a given agent is not meaningful; the best strategy depends on the choices of others

o   Pareto domination

o   Pareto optimality

o   Best response

o   Nash equilibrium

o   Strict Nash

o   Weak Nash

 

5.    Chapter 7: Learning and Teaching

·         The interaction between learning and teaching

o   Stackelberg game

·         What constitutes learning?

·         Two categories of theories of learning in MAS: Descriptive and Prescriptive

·         Descriptive

o   Realism, Convergence

o   Convergence properties

§  Show convergence to stationary strategies which form a Nash equilibrium of the stage game

§  Require that the empirical frequency of play converge to a Nash equilibrium

§  Seek convergence to a correlated equilibrium of the stage game

§  Require that the non-stationary policies converge to an interesting state

·         Prescriptive

o   Strategic normative games – where agents are self-motivated

o   Notion of self-play

o   Safety, Rationality, and No-Regret, informal

·         Fictitious play – an instance of model-=based learning, in which the learner explicitly maintains beliefs about the opponent’s strategy

·         Rational learning

·         Reinforcement learning

o   Q-learning

§  Alpha, beta,

o   Belief-based reinforcement learning

o   Targeted learning, no-regret learning

 

6.    Chapter 9: Aggregating Preferences: Social Choice

·         Plurality voting

o   Condorcet condition

·         Social choice function, Social choice correspondence, Condorcet winner, Smith set, Social welfare function

·         Voting

o   Plurality voting, cumulative voting, approval voting, plurality with elimination, Borda voting, pairwise elimination

o   Voting paradoxes: Condorcet condition not met, sensitivity to a losing candidate, sensitivity to the agenda setter

·         Social welfare functions (ordering!)

o   Pareto efficiency (PE), Independence of irrelevant alternatives (IIA), Nondictatorship

o   Arrow’s Impossibility Theorem

·         Social choice functions (top-ranked outcome!)

o   Weak Pareto efficiency, Monotonicity, Nondictatorship

o   Muller-Satterhwaite’s Impossibility Theorem

·         Ranking system

o   Agents are asked to vote to express their opinions about each other, with the goal of determining a social ranking

§  Agents who are ranked higher by others have more weighted votes.

o   Approval voting satisfies IIA, PE, and nondictatorship

o   Approval voting satisfies Ranked IIA, positive response and anonymity

 

7.    Chapter 10: Protocols for Strategic Agents: Mechanism Design

·         Strategic!  Assume that agents will behave so as to maximize their individual payoffs

·         Why is mechanism design so important to MAS designers?

·         Local decision making vs. global, emergent coherence

o   Autonomy vs. social chaos

·         Bayesian game setting and mechanism

o   Implementation in dominant strategies

o   Implementation in Bayes-Nash equilibrium

·         The truthfulness property

o   The revelation principle

·         Gibbard-Satterthwaite’s Impossibility Theorem

·         A way to get around the impossibility: Quasinlinear Preferences

o   Rewards and payments

o   Risk attitudes: neutral, averse, and seeking

o   Conditional utility independence, valuation, payment

o   Basic constraints: Truthfulness, Efficiency, Budget Balance, Ex interim Individual Rationality, Ex post Individual Rationality, Tractability

o   Optimization properties: Revenue Maximization, Revenue minimization, Maxmin Fairness, and Price-Of-Anarchy Minimization

·         Groves Mechanism

o   Truth telling is a dominant strategy under any Groves mechanism

·         The Vickrey-Clarke-Groves (VCG) Mechanism

o   a.k.a. Pivot Mechanism

o   Clarke tax

o   How does the mechanism work? 

o   Drawbacks of VCG

 

8.    Chapter 11: Protocols for Multiagent Resource Allocation: Auction

·         How can auctions be used to allocate task or resources?

·         Single-good auctions

o   English, Japanese, Dutch, and sealed-bid auctions

§  Open-cry, Open-exit, First price vs. second price, Vickrey

·         Auctions as negotiations

·         Auctions as Bayesian mechanisms

o   Independent private value (IPV) setting (as opposed to common value or interdependent value settings)

·         In a second-price auction where bidders have independent private values, truth telling is a dominant strategy

·         Strategically equivalence, time complexity, communication complexity.

·         Revenue equivalence theorem

o   Risk attitudes

o   Relationships between revenues of various single-good auction protocols

·         Other auctions: Reverse auctions, Double auctions, All-pay (with entry costs) auctions

·         Collusions

o   How does a bidding ring survive?

o   How does revenue equivalence theorem factor into a bidder’s decision to join a ring?

·         Contract net protocol (CNP)

o   Task announcement, Task announcement processing, Bidding, Bid processing, Contract processing, reporting results, and termination, and Negotiation tradeoffs

 

9.    Chapter 12: Teams of Selfish Agents: An Introduction to Coalitional Game Theory

·         How self-interested agents can combine to form effective teams

o   Which coalitions to form?

o   How to distribute payoffs?

·         Coalitional game with transferability utility

o   The payoffs to a coalition may be freely redistributed among its members

·         Examples: voting game, airport game

·         Supperadditive game, Additive game, Constant-Sum game, Convex game, Simple game

·         Analyzing coalitional games in terms of payoffs to members

o   Feasibly payoff, Pre-imputation, Imputation, Individual rationality

·         Payoffs should be divided fairly

o   Axioms: Symmetry, Dummy player, Additivity

o   Given a coalitional game, there is a unique pre-imputation that satisfies the symmetry, dummy player, additivity axioms

o   Shapley value!

§  Average marginal contribution

§  Why is it fair?

o   The Core

§  The stability issue

§  A payoff vector is in the core of a coalitional game iff for each coalition, the sum of all agents’ rewards is greater than the valuation of the coalition.

§  Is the core always nonempty?

§  Characterizing when a coalition game has a nonempty core

·         Balanced Weights and Bondereva-Shapley

·         Veto player’s role in simple game – core?

·         Every convex game has a nonempty core.  The Shapley value is in the core.