Compact Representations and Approximations for Compuation in Games


 Leonard Harrell
 5 years ago
 Views:
Transcription
1 Compact Representations and Approximations for Compuation in Games Kevin Swersky April 23, 2008 Abstract Compact representations have recently been developed as a way of both encoding the strategic interactions of a game in an intuitive way. They are also useful in speeding up the computation of important solution concepts such as Nash Equiliria, because they can often be represented in a size that is polynomial in the number of players. Rather than directly encoding a game in a compact form, another approach is to factor the game into smaller subgames, and then perform computation in the factored space. In this paper, I give an overview of one kind of compact representation called a Graphical Game, and I discuss the approach of factoring games. I then use the idea of creating approximate factorings, which preserve ǫequilbria to develop a new algorithm that can be used to simplify computation in a Graphical Game by creating an approximate version. 1 Introduction In the earliest forms of Game Theory, the standard way to represent a game was to use the so called Normal Form. Given n players with action sets A = {A 1,A 2,...,A n } the Normal form is an enumeration of the utilities U = (u 1,u 2,...,u n ) of each player over all possible action profiles a A. While useful for visualizing and analyzing small games, the size of this representation has the unfortunate characteristic of growing exponentially in the number of players. In this form, a calculation that is linear in the size of the game, such as Expected Value, will take exponential time if the number of players is unbounded. In reality, there are many games that exhibit some form of structure which can be exploited. For example, two players in an n player game might have no ability to influence each others utilities, and in this case the Normal Form would store redundant information. Compact representations, such as Graphical Games[4], Action Graph Games[3], Multi Agent Influence Diagrams[5], are able to significantly reduce the size of a game with special structure, allowing them to be represented with storage that grows polynomially with the number of players. In these reduced representations, algorithms have been found that significantly 1
2 speed up the computation of important solution concepts such as Nash Equilibria or Correlated Equilibria. Recently, an alternative approach to representing games in a more compact form has been proposed in [2]. Rather than knowing the structure in advance, this approach takes a Normal Form game and attempts to automatically find a specific kind of independence structure within the game. If such a structure is found, then the idea is to split the game up into a number of smaller subgames, or Factor Games. Within these factor games, it has been shown that their Nash Equilibria and associated strategy profiles can be combined to form a Nash Equilibrium in the original game. Unfortunately, the independence structure that this technique is able to exploit is extremely limited. More interestingly, there is a way to automatically factor games into approximate subgames where ǫequilibria (an equilibrium where no agent can gain more than ǫ by deviating) are preserved. In this paper I give an overview of Graphical Games, including highlighting some algorithms that are able to use their structure to greatly speed up computation of Nash Equilibria. I then describe the Factoring Games algorithm in its exact and approximate forms, and I discuss some issues that will arise when using this approach in practice. Finally, I propose an algorithm that merges the ideas of Factoring Games and Compact Representations together by iteratively removing edges from a graphical game while attempting to compensate for the loss of strategic interaction between players. In this way, fast approximation techniques may be developed for certain classes of games that can be used to speed up computation of ǫequilibria. 2 Compact Games Representations 2.1 Motivation One of the most fundamental problems in Game Theory is the expected utility problem[7]. In a normal form game, computing expected utility for an agent consists of simply looking over each possible outcome, and multiplying the probability that the outcome will occur by the utility the agent will receive for that outcome. While this simple algorithm is linear in the size of the game representation, it will be exponential for Normal Form representations. In general, this algorithm becomes increasingly more difficult when the size of the representation is reduced, and can potentially be exponential if the game is compacted in a naive way. Several representations exist, however, that do not arbitrarily compress the representation size of a game, but rather exploit structure within the game. In many situations, this can lead to polynomial time expected utility algorithms over a representation that is polynomial in the number of players n and actions A. Indeed, it has been shown that in certain compact representations, computing a Nash Equilibrium can be reduced to computing the Nash Equilibrium in a generalsum 2 player Normal Form game, where the representation size only 2
3 larger by a polynomial amount. This is a good result, as the problem of finding a Nash Equilibrium was recently found in [1] to be PPADcomplete, which means that in some cases computing Equilibria can be exponentially faster using a compact representation. 2.2 Graphical Games In a Graphical Game, we can represent n players and A = A 1 A 2...A n actions per player as an undirected graph G = (V,E). In such a game, each vertex V i corresponds to a player i, and each edge E corresponds to links between players. A player j is said to be a neighbour of i if there is an edge (j,i) E. The set of neighbours of Each player is denoted v(i). In these games, i has an action set A i that they can choose from, and their utility is only influenced by their own action, and the actions of their neighbours. Such a representation captures independence between payoffs, as players who are not neighbours in the graph cannot directly influence each others utilities. Since each node essentially stores a normal form representation of the game with only players from the set v(i), the representation size will be exponential in v(i). If v(i) is bounded by a constant, then this means that the corresponding Grapical Game will be polynomial in the number of players n. With these results, there are a number of algorithms that could be applied to computing Nash Equilibria. One such algorithm called TreeProp will compute an ǫequilibrium in polynomial time if the graph has a tree structure. 3 Factoring Games 3.1 Exact Factoring Factoring Games takes a different approach to creating a Compact Representation. The idea is that to automatically find structure in a Normal Form game that allows it to be split into two independent subgames whose overall strategy space is the same as the original game. Using this, standard problems like Nash Equilibria could be computed more efficiently in this smaller representation, and these would correspond to the same equilibria as those in the full game. The following definitions explain how such a factoring is represented Definition 1. The product of two nplayer games G a and G b, G a G b is a new nplayer game G c in which The pure strategy set for player i is the Cartesian product S c i = Sa i Sb i of agent i s pure strategy sets S a i in G a and S b i in G b. There is exactly one pure strategy profile s c S c for every pair of pure strategy profiles s a S a and s b S b For pure strategy profile s a S a, s b S b and s c S c, i agent i s utility is u c i (s c) = u a i (s a) + u b i (s b) 3
4 Definition 2. When we can write an nplayer game G c as G a G b then G a and G b are called the factors of G c Thus, the game factoring algorithm is designed to find factors where the utilities of outcomes in the product game can be written as the sum of the utilities from outcomes in the factor games. In MAID s, games with this sort of structure would be represented as several different connected components, with each connected component representing a factor. As an example of such a game, the authors introduced the Direct Flight Game neither LACH CHNY Both neither 0,0 0,2 0,4 0,6 LACH 2,01,1 2,41,3 CHNY 4,0 4,21,11,1 Both 6,0 3,1 1,12,2 In this game, there are two airlines, and two routes. Each airline can choose to open one route, the other, both, or none. Here, the routes are completely independent of each other, and so the game can be split into two factors. LACH LACH LACH 0,0 0,2 LACH 2,01,1 CHNY CHNY CHNY 0,0 0,4 CHNY 4,01,1 Notice that the sum of the utilities from action profiles taken in the factor games results in the payoffs given in the product game. There are some interesting results that come from considering such factorings, namely that if such a factoring exists, then combining the strategies of Nash Equilibrium strategy profiles in the factor games will result in a strategy profile that is a Nash Equilibrium of the product game. In order to automatically determine such a factoring if one exists, the authors represent the utilities for each player i as a polynomial where each term signifies all possible outcomes given that player i chooses a specific action, and every action that can be taken by player i is represented as a term in this polynomial. In essence, when the game is represented as these polynomials then they can be factored using standard polynomial factoring algorithms. The resulting factors then represent the outcomes of the factor games. The details of this are not particularly illuminating, but the end result is that if one or more factorings of this sort exist, then their algorithm will find it in time polynomial in the size of the representation. Unfortunately, since the representation is Normal Form, this doesn t appear to be very useful in a general case because it will be exponential in the number of players. Additionally, it seems hard to justify whether arbitrary Normal Form games will exhibit the special structure this algorithm requires. 4
5 3.2 Approximate Factoring A more interesting result of the paper is the idea of Approximate Game Factoring. If no independence structure exists, then the idea behind Approximate Game Factoring is to divide a game into two factors such that the product game only approximately represents the original game to within some ǫ. More formally, Definition 3. Given ǫ > 0, an ǫequilibrium of a game G is an strategy profile s where for every player i s s u i (s) + ǫ u i (s ) Definition 4. A collection of factor games F with product G a is an ǫfactoring of a game G b if for every strategy profile s a S a, s b S b ui (s a ) u i (s b ) ǫ for all players i The authors show that a product game under a strategy profile in which each of factor games are in ǫequilibrium is also in ǫequilibrium. In order to illustrate the idea, the authors modify the Direct Flight Game to the Indirect Flight Game. In this case, there is a bonus reward of 2 for servicing both routes because it allows for indirect flights. neither LACH CHNY Both neither 0,0 0,2 0,4 0,8 LACH 2,01,1 2,41,5 CHNY 4,0 4,21,11,3 Both 8,0 5,1 3,11,1 In this case, the two flights are no longer independent. In the context of MAID s they are said to strategically interact, and can no longer be thought of as two separate connected components. This means that no factoring exists, so the algorithm given above will not work. Interestingly enough, if one were to still pursue factoring, and naively assumed that the two flights were independent, then the smallest ǫ achievable becomes 2. With approximate factoring, however, it is possible to do better. Instead of simply factoring the game and ignoring dependencies, approximate factoring attempts to restructure the payoffs available to each player so as to minimize the impact of ignoring the dependence relation. Under the optimal approximate factoring, the new payoffs in the factor games become LACH LACH LACH 1 4, ,23 4 LACH 2 3 4, , 1 4 CHNY CHNY CHNY 1 4, , CHNY 4 3 4, , 1 4 In this game, the largest difference between utilities in the product game and the original game are bounded by ǫ = 1 2. The way this is accomplished is via a linear program. Given a proposed factoring of a game G c into G a an G b, for each player i and pair of distinct pure strategy profiles s,s S c of G c the authors propose the following 5
6 min u a i (s a ),u b i (sb ) u a i (sa ) + u b i (sb ) u c i (s) = ǫ subject to u a i (sa ) + u b i (sb ) u c i (s) = (ua i (s a ) + u b i (s b ) u c i (s )) s a S a,s b S b u a i (s a) + u b i (s b) u c i (s a s b ) u i (s a ) + u i (s b ) s i (s c ) = ǫ s a S a,s b S b u a i (s a)+u b i (s b) u c i (s a s b ) u i (s a )+u i (s b ) s i (s c ) = ǫ The idea here is to minimize the difference between the utilities in the product game, and the utilities in the original game. The first constraint ensures that the difference will be bounded above and below by the same ǫ, a requirement for ǫequilibria. The second and third constraints ensure that no utilities for actions played in the factored games can overestimate or underestimate the utilities in the full game by any more than ǫ. This linear program is then run over all players and pairs of distinct pure strategy profiles, and the resulting maximum ǫ found is taken to be the largest error between utilities in G c and G a G b. The computational complexity of enumerating over all distinct pairs of outcomes for each player is O(n ( ) S 2 ) = O(n S S 1 2 ) While this is polynomial in the size of the representation, in a Normal Form game where n is unbounded S will grow exponentially, and so this algorithm will require exponential time to compute an ǫfactoring. Another drawback to this approach is that it requires that the factoring be given in advance. The authors propose a bruteforce search over possible factorings, but admit that this is infeasible. Still, if such factorings could be found, and the size of the representation bounded, then it would be interesting to see whether this algrorithm could devise good approximations in a practical scenario. 4 Approximation Techniques for Compact Games In the ideal case, any computation performed on games would yield exact solutions. Of course, there are a number of bottlenecks that make such a goal very difficult to achieve. Several were mentioned previously such as the sheer size of the search space, and the complexity of the underlying algorithms. Others are more subtle, such as the restriction to machine precision. Since exact computation is often intractable, it may be useful in practical situations to allow for approximations. Indeed, the authors in [6] argue that in cases of bounded rationality, agents will converge to ǫequilibria. While the authors treat the idea of factoring games as a separate issue from computation in compact representations, I believe that the two ideas are not mutually exclusive. The creators of Factor Games suggest that computing Factors in existing compact representations might be an easier problem. It seems unlikely that nice, clean factors as described above would exist in the majority of games, however there may be uses for approximate factorings to find ǫequilibria. First, notice that restricting 6
7 our attention to compact games allows us to exploit two properties that do not exist in the Normal Form: 1. The algorithms presented above are polynomial in the size of the representation, and the representation in compact games are generally polynomial in the number of players. Thus, the algorithms presented above would be polynomial if applied to a polynomial compact representation. 2. The approximate factoring algorithm requires a factoring as its input. In compact representations, existing decomposition algorithms could be potentially used to generate such factors. I believe that the most important contribution the Factoring Games paper made was the idea of compensating for lost strategic dependencies. By restructuring the utilities of an approximated game such that its solution properties are as close as possible to the original game, we could simplify computation while ensuring that the results remain valid. If we were to apply factoring games directly to Graphical Games, then a potential factoring would need to be devised by severing edges in the game until we were left with two separate, connected components. In order to find the optimal edges to cut, we would need to be aware of the impact this would have on the approximate equilibria of the game, and we would need to find some way of cutting the optimal edges to create two smaller subgames. This would require some notion of the cost of each edge, and we would have to resort to some technique such as Normalized Graph Cuts[6] in order to find the optimal partition. Rather than taking this approach, I suggest a decomposition where edges are removed from the game one by one until the game has been reduced to a more manageable, but not necessarily separated form. Such an algorithm would be used as a preprocessing step in computing approximate equilibria. It may be useful to reduce a game to a form such as a tree which would enable the use of algorithms like TreeProp. Such an approach could also be an interesting way of simulating equilibria with bounded rationality. Consider a Graphical Game where we have 3 players, a b and c. Each player would be represented as a node on the graph. Now consider two directed edges (b,a) and (c,a). This means that a s utility is directly influenced by the actions of b and c. Let the payoffs for agent a be given by the following matrix b 1 b 2 a a b 1 b 2 a a c 1 c 2 Here, a is able to choose the row, b is able to choose the column, and c is able to choose the matrix. Denote a s utility function for action profile (a i,b j,c k ) as u a (i,j,k). Our goal is to find an edge weight ǫ for the directed edge (c,a) such that if we remove this edge, we can restructure the payoffs of a so that the 7
8 difference between a s payoffs in the original utility funcion, and a s payffs in the new utility function (which is independent of c) is at most ǫ. To do this, first notice that removing dependence on c effectively removes a single matrix from the representation above. Thus, we can encode a s payoffs in the new game as 4 variables (u 11,u 12,u 21,u 22 ). If we were to then project this out into the 3Dimensional space occupied by the original game, the resulting payoffs would be b 1 b 2 a 1 u 11 u 12 a 2 u 21 u 22 b 1 b 2 a 1 u 11 u 12 a 2 u 21 u 22 c 1 c 2 The new game now has redundancies because we have removed c s ability to affect a s payoffs. To find the minimum ǫ, we must first decide how we are going to do this projection from the the original game onto the new game. After this, we must decide on a loss function. Here I choose the same loss function as was given in the Factoring Games paper (minimizing the maximum ǫ), however there may be potential to define other loss functions. For simplicity, if we were to project the original game onto a linear function of the u s, then we would be trying to optimize the following linear program subject to u i,j u a (i,j,k) ǫ, i,j,k = 1,2 u i,j u a (i,j,k) ǫ, i,j,k = 1,2 ǫ,u ij 0, i,j = 1,2 min(ǫ) In this program, the first two constraints correspond to the fact that we want to bound the utilities for a in the new game above and below by ǫ. From this, we can get the payoffs in the new game as well as the cost associated with ignoring the interactions between player a and c. By restructuring a s payoffs in terms of the new game, we have effectively removed the directed edge (c,a) while reducing the impact to the strategy space as much as possible. Since the number of edges in the graph is polynomial with respect to n, this algorithm can be applied to all edges in a Graphical Game, and the costs associated with removing each edge can therefore be computed in polynomial time. Using these, we can then find the sequence of edges with minimum impact, and cut those until a maximum ǫ tolerance is reached. It would be important to ensure that if this algorithm were applied to several edges, then we could specify the ǫ error in the entire approximated game with respect to the original game. This algorithm, while representing a potential strategy to reduce computation in Graphical Games, has a pitfall that is also apparent in Factoring Games. While we can find the minimum ǫ possible when removing strategic interactions, 8
9 this ǫ may be arbitrarily bad. Since we are projecting from a higher dimensional action space onto a lower dimensional one, we risk losing important information that could only be carried in the high dimensional space. A user of this algorithm, or approximate factoring, would not be able to find out what this ǫ is before running the approximation. Thus, they risk losing valuable computation if this ǫ is larger than they are willing to tolerate. For example, running this linear program on the example given above yields the following utilities for a, with ǫ = 1.5 b 1 b 2 a a This illustrates a potential weakness of the approximate factoring approach. It may be the case, however, that certain classes of games have nice guaranteed bounds on the maximum ǫ that can be found by removing strategic interactions. It is also possible that within a game there may be very weak interactions that will generate a small ǫ. This algorithm may also be useful in simulating agents with bounded rationality where they are only able to reason about a small portion of a game, and must ignore other parts that may not be independent. Should some classes of games be found where this holds, these kinds of algorithms might represent a new way of approximating games to speed up computation. Certainly in the algorithm presented, different loss functions and basis functions could be used that capture as much information as possible in the reduced dimension. There are many machine learning techniques available that are designed for this specific task, and it would be interesting to see how they fare in a multiagent environment. It is also possible that severing edges in Graphical Games is too restrictive an approach. There may be other ways of applying similar approximation techniques to compact games which preserve more information, and are not subject to arbitrary losses. 5 Conclusion Computing solution concepts in the Normal Form can be a computationally expensive endeavor. When presented with an unbounded number of agents, the possible outcomes can grow exponentially, and thus even algorithms that run linearly in the size of the representation will need exponential time to compute. Compact representations which seek to exploit structure within a game provide a simple, and intuitive way to drastically reduce the amount of computation needed to compute many solution concepts such as Expected Value and Nash Equilibria. As an alternative, factoring games seeks to take a Normal Form representation and break it into subgames which preserve as much information as possible in terms of strategic interactions in the full game. While certainly interesting, this approach has a number of drawbacks mostly related to the fact that they 9
10 need to start with a Normal Form representation. In this context, the algorithms presented will still run in exponential time if the number of players is unbounded. While it may not be the best approach to try to decompose a Normal Form game in this way, it may be worthwhile to apply the ideas developed in Factoring Games to existing compact representations. In this way, existing representations could be decomposed to simplify computation while still maintaining many of the strategic properties of the original game. I proposed a decomposition algorithm for Graphical Games which attempts to remove edges between players, and compensates for the loss of strategic interaction by reformulating the utilities in the new approximate game. I also suggested the possibility that in such approximation schemes, the approximate versions of the original games might be arbitrarily bad. Regardless, there may be interesting results in classes of games where the maximum error could be bounded, and in these cases it would be interesting to see what kind of computational savings could be made by applying approximation techniques. References References [1] C. Daskalakis, P. Goldberg, and C. Papadimitriou. The complexity of computing a Nash equilibrium. Proceedings of the thirtyeighth annual ACM symposium on Theory of computing, pages 71 78, [2] G. Davis, M. Benisch, K. Carley, and N. Sadeh. Factoring games to isolate strategic interactions. Proceedings of the 6th international joint conference on Autonomous agents and multiagent systems, [3] A. Jiang, K. LeytonBrown, and N. Bhat. ActionGraph Games. [4] M. Kearns, M. Littman, and S. Singh. Graphical models for game theory. Proceedings of UAI, 1, [5] D. Koller and B. Milch. Multiagent influence diagrams for representing and solving games. Games and Economic Behavior, 45(1): , [6] J. Shi and J. Malik. Normalized Cuts and Image Segmentation. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTEL LIGENCE, pages , [7] Y. Shoham and K. LeytonBrown. MultiAgent Systems. Cambridge University Press,
Factoring Games to Isolate Strategic Interactions
Factoring Games to Isolate Strategic Interactions George B. Davis, Michael Benisch, Kathleen M. Carley, and Norman M. Sadeh January 2007 CMUISRI06121R a a This report is a revised version of CMUISRI06121
More informationHow To Find Factor Games In Polynomials
Factoring Games to Isolate Strategic Interactions George B. Davis, Michael Benisch, Kathleen M. Carley and Norman M. Sadeh School of Computer Science, Carnegie Mellon University {gbd, mbenisch, carley,
More informationDATA ANALYSIS II. Matrix Algorithms
DATA ANALYSIS II Matrix Algorithms Similarity Matrix Given a dataset D = {x i }, i=1,..,n consisting of n points in R d, let A denote the n n symmetric similarity matrix between the points, given as where
More informationThe MaxDistance Network Creation Game on General Host Graphs
The MaxDistance Network Creation Game on General Host Graphs 13 Luglio 2012 Introduction Network Creation Games are games that model the formation of largescale networks governed by autonomous agents.
More informationApproximation Algorithms
Approximation Algorithms or: How I Learned to Stop Worrying and Deal with NPCompleteness Ong Jit Sheng, Jonathan (A0073924B) March, 2012 Overview Key Results (I) General techniques: Greedy algorithms
More informationLocalEffect Games. Kevin LeytonBrown and Moshe Tennenholtz. Abstract. 1 Introduction. 2 LocalEffect Games
LocalEffect Games Kevin LeytonBrown and Moshe Tennenholtz kevinlb@stanford.edu; Computer Science Department, Stanford University, Stanford CA 430, USA moshet@ie.technion.ac.il; Faculty of Industrial
More informationMicroeconomic Theory Jamison / Kohlberg / Avery Problem Set 4 Solutions Spring 2012. (a) LEFT CENTER RIGHT TOP 8, 5 0, 0 6, 3 BOTTOM 0, 0 7, 6 6, 3
Microeconomic Theory Jamison / Kohlberg / Avery Problem Set 4 Solutions Spring 2012 1. Subgame Perfect Equilibrium and Dominance (a) LEFT CENTER RIGHT TOP 8, 5 0, 0 6, 3 BOTTOM 0, 0 7, 6 6, 3 Highlighting
More informationUSING SPECTRAL RADIUS RATIO FOR NODE DEGREE TO ANALYZE THE EVOLUTION OF SCALE FREE NETWORKS AND SMALLWORLD NETWORKS
USING SPECTRAL RADIUS RATIO FOR NODE DEGREE TO ANALYZE THE EVOLUTION OF SCALE FREE NETWORKS AND SMALLWORLD NETWORKS Natarajan Meghanathan Jackson State University, 1400 Lynch St, Jackson, MS, USA natarajan.meghanathan@jsums.edu
More informationChapter 11. 11.1 Load Balancing. Approximation Algorithms. Load Balancing. Load Balancing on 2 Machines. Load Balancing: Greedy Scheduling
Approximation Algorithms Chapter Approximation Algorithms Q. Suppose I need to solve an NPhard problem. What should I do? A. Theory says you're unlikely to find a polytime algorithm. Must sacrifice one
More informationNEW VERSION OF DECISION SUPPORT SYSTEM FOR EVALUATING TAKEOVER BIDS IN PRIVATIZATION OF THE PUBLIC ENTERPRISES AND SERVICES
NEW VERSION OF DECISION SUPPORT SYSTEM FOR EVALUATING TAKEOVER BIDS IN PRIVATIZATION OF THE PUBLIC ENTERPRISES AND SERVICES Silvija Vlah Kristina Soric Visnja Vojvodic Rosenzweig Department of Mathematics
More informationI d Rather Stay Stupid: The Advantage of Having Low Utility
I d Rather Stay Stupid: The Advantage of Having Low Utility Lior Seeman Department of Computer Science Cornell University lseeman@cs.cornell.edu Abstract Motivated by cost of computation in game theory,
More information! Solve problem to optimality. ! Solve problem in polytime. ! Solve arbitrary instances of the problem. #approximation algorithm.
Approximation Algorithms 11 Approximation Algorithms Q Suppose I need to solve an NPhard problem What should I do? A Theory says you're unlikely to find a polytime algorithm Must sacrifice one of three
More informationComputational Game Theory and Clustering
Computational Game Theory and Clustering Martin Hoefer mhoefer@mpiinf.mpg.de 1 Computational Game Theory? 2 Complexity and Computation of Equilibrium 3 Bounding Inefficiencies 4 Conclusion Computational
More informationBargaining Solutions in a Social Network
Bargaining Solutions in a Social Network Tanmoy Chakraborty and Michael Kearns Department of Computer and Information Science University of Pennsylvania Abstract. We study the concept of bargaining solutions,
More informationFairness in Routing and Load Balancing
Fairness in Routing and Load Balancing Jon Kleinberg Yuval Rabani Éva Tardos Abstract We consider the issue of network routing subject to explicit fairness conditions. The optimization of fairness criteria
More information! Solve problem to optimality. ! Solve problem in polytime. ! Solve arbitrary instances of the problem. !approximation algorithm.
Approximation Algorithms Chapter Approximation Algorithms Q Suppose I need to solve an NPhard problem What should I do? A Theory says you're unlikely to find a polytime algorithm Must sacrifice one of
More informationA GraphTheoretic Network Security Game
A GraphTheoretic Network Security Game Marios Mavronicolas 1, Vicky Papadopoulou 1, Anna Philippou 1, and Paul Spirakis 2 1 Department of Computer Science, University of Cyprus, Nicosia CY1678, Cyprus.
More informationChapter 7. Sealedbid Auctions
Chapter 7 Sealedbid Auctions An auction is a procedure used for selling and buying items by offering them up for bid. Auctions are often used to sell objects that have a variable price (for example oil)
More information1 Introduction. Linear Programming. Questions. A general optimization problem is of the form: choose x to. max f(x) subject to x S. where.
Introduction Linear Programming Neil Laws TT 00 A general optimization problem is of the form: choose x to maximise f(x) subject to x S where x = (x,..., x n ) T, f : R n R is the objective function, S
More information1 Nonzero sum games and Nash equilibria
princeton univ. F 14 cos 521: Advanced Algorithm Design Lecture 19: Equilibria and algorithms Lecturer: Sanjeev Arora Scribe: Economic and gametheoretic reasoning specifically, how agents respond to economic
More information6.207/14.15: Networks Lecture 15: Repeated Games and Cooperation
6.207/14.15: Networks Lecture 15: Repeated Games and Cooperation Daron Acemoglu and Asu Ozdaglar MIT November 2, 2009 1 Introduction Outline The problem of cooperation Finitelyrepeated prisoner s dilemma
More informationApplied Algorithm Design Lecture 5
Applied Algorithm Design Lecture 5 Pietro Michiardi Eurecom Pietro Michiardi (Eurecom) Applied Algorithm Design Lecture 5 1 / 86 Approximation Algorithms Pietro Michiardi (Eurecom) Applied Algorithm Design
More informationNotes on Factoring. MA 206 Kurt Bryan
The General Approach Notes on Factoring MA 26 Kurt Bryan Suppose I hand you n, a 2 digit integer and tell you that n is composite, with smallest prime factor around 5 digits. Finding a nontrivial factor
More informationEquilibrium computation: Part 1
Equilibrium computation: Part 1 Nicola Gatti 1 Troels Bjerre Sorensen 2 1 Politecnico di Milano, Italy 2 Duke University, USA Nicola Gatti and Troels Bjerre Sørensen ( Politecnico di Milano, Italy, Equilibrium
More informationDynamic programming. Doctoral course Optimization on graphs  Lecture 4.1. Giovanni Righini. January 17 th, 2013
Dynamic programming Doctoral course Optimization on graphs  Lecture.1 Giovanni Righini January 1 th, 201 Implicit enumeration Combinatorial optimization problems are in general NPhard and we usually
More informationAnalysis of an Artificial Hormone System (Extended abstract)
c 2013. This is the author s version of the work. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purpose or for creating
More informationDecentralized Utilitybased Sensor Network Design
Decentralized Utilitybased Sensor Network Design Narayanan Sadagopan and Bhaskar Krishnamachari University of Southern California, Los Angeles, CA 900890781, USA narayans@cs.usc.edu, bkrishna@usc.edu
More information5.1 Bipartite Matching
CS787: Advanced Algorithms Lecture 5: Applications of Network Flow In the last lecture, we looked at the problem of finding the maximum flow in a graph, and how it can be efficiently solved using the FordFulkerson
More informationInfinitely Repeated Games with Discounting Ù
Infinitely Repeated Games with Discounting Page 1 Infinitely Repeated Games with Discounting Ù Introduction 1 Discounting the future 2 Interpreting the discount factor 3 The average discounted payoff 4
More informationScheduling Home Health Care with Separating Benders Cuts in Decision Diagrams
Scheduling Home Health Care with Separating Benders Cuts in Decision Diagrams André Ciré University of Toronto John Hooker Carnegie Mellon University INFORMS 2014 Home Health Care Home health care delivery
More informationComputing GameTheoretic Solutions and Applications to Security
Computing GameTheoretic Solutions and Applications to Security Vincent Conitzer Departments of Computer Science and Economics Duke University Durham, NC 27708, USA conitzer@cs.duke.edu Abstract The multiagent
More informationLearning and Predicting Dynamic Networked Behavior with Graphical Multiagent Models
Learning and Predicting Dynamic Networked Behavior with Graphical Multiagent Models Quang Duong Michael P. Wellman Satinder Singh Michael Kearns Computer Science and Engineering, University of Michigan
More informationCourse: Model, Learning, and Inference: Lecture 5
Course: Model, Learning, and Inference: Lecture 5 Alan Yuille Department of Statistics, UCLA Los Angeles, CA 90095 yuille@stat.ucla.edu Abstract Probability distributions on structured representation.
More informationNPHardness Results Related to PPAD
NPHardness Results Related to PPAD Chuangyin Dang Dept. of Manufacturing Engineering & Engineering Management City University of Hong Kong Kowloon, Hong Kong SAR, China EMail: mecdang@cityu.edu.hk Yinyu
More informationOPRE 6201 : 2. Simplex Method
OPRE 6201 : 2. Simplex Method 1 The Graphical Method: An Example Consider the following linear program: Max 4x 1 +3x 2 Subject to: 2x 1 +3x 2 6 (1) 3x 1 +2x 2 3 (2) 2x 2 5 (3) 2x 1 +x 2 4 (4) x 1, x 2
More informationSmall Maximal Independent Sets and Faster Exact Graph Coloring
Small Maximal Independent Sets and Faster Exact Graph Coloring David Eppstein Univ. of California, Irvine Dept. of Information and Computer Science The Exact Graph Coloring Problem: Given an undirected
More informationGame Theory and Algorithms Lecture 10: Extensive Games: Critiques and Extensions
Game Theory and Algorithms Lecture 0: Extensive Games: Critiques and Extensions March 3, 0 Summary: We discuss a game called the centipede game, a simple extensive game where the prediction made by backwards
More informationNetwork (Tree) Topology Inference Based on Prüfer Sequence
Network (Tree) Topology Inference Based on Prüfer Sequence C. Vanniarajan and Kamala Krithivasan Department of Computer Science and Engineering Indian Institute of Technology Madras Chennai 600036 vanniarajanc@hcl.in,
More information8.1 Min Degree Spanning Tree
CS880: Approximations Algorithms Scribe: Siddharth Barman Lecturer: Shuchi Chawla Topic: Min Degree Spanning Tree Date: 02/15/07 In this lecture we give a local search based algorithm for the Min Degree
More informationBonus Maths 2: Variable Bet Sizing in the Simplest Possible Game of Poker (JB)
Bonus Maths 2: Variable Bet Sizing in the Simplest Possible Game of Poker (JB) I recently decided to read Part Three of The Mathematics of Poker (TMOP) more carefully than I did the first time around.
More informationCournot s model of oligopoly
Cournot s model of oligopoly Single good produced by n firms Cost to firm i of producing q i units: C i (q i ), where C i is nonnegative and increasing If firms total output is Q then market price is P(Q),
More information1. Write the number of the lefthand item next to the item on the right that corresponds to it.
1. Write the number of the lefthand item next to the item on the right that corresponds to it. 1. Stanford prison experiment 2. Friendster 3. neuron 4. router 5. tipping 6. small worlds 7. jobhunting
More informationThe Goldberg Rao Algorithm for the Maximum Flow Problem
The Goldberg Rao Algorithm for the Maximum Flow Problem COS 528 class notes October 18, 2006 Scribe: Dávid Papp Main idea: use of the blocking flow paradigm to achieve essentially O(min{m 2/3, n 1/2 }
More informationPriority Based Load Balancing in a SelfInterested P2P Network
Priority Based Load Balancing in a SelfInterested P2P Network Xuan ZHOU and Wolfgang NEJDL L3S Research Center Expo Plaza 1, 30539 Hanover, Germany {zhou,nejdl}@l3s.de Abstract. A fundamental issue in
More informationAlgorithm Design and Analysis
Algorithm Design and Analysis LECTURE 27 Approximation Algorithms Load Balancing Weighted Vertex Cover Reminder: Fill out SRTEs online Don t forget to click submit Sofya Raskhodnikova 12/6/2011 S. Raskhodnikova;
More informationLinear Programming. March 14, 2014
Linear Programming March 1, 01 Parts of this introduction to linear programming were adapted from Chapter 9 of Introduction to Algorithms, Second Edition, by Cormen, Leiserson, Rivest and Stein [1]. 1
More informationOptimization Modeling for Mining Engineers
Optimization Modeling for Mining Engineers Alexandra M. Newman Division of Economics and Business Slide 1 Colorado School of Mines Seminar Outline Linear Programming Integer Linear Programming Slide 2
More informationLinear Programming. Widget Factory Example. Linear Programming: Standard Form. Widget Factory Example: Continued.
Linear Programming Widget Factory Example Learning Goals. Introduce Linear Programming Problems. Widget Example, Graphical Solution. Basic Theory:, Vertices, Existence of Solutions. Equivalent formulations.
More informationWeek 7  Game Theory and Industrial Organisation
Week 7  Game Theory and Industrial Organisation The Cournot and Bertrand models are the two basic templates for models of oligopoly; industry structures with a small number of firms. There are a number
More informationCost Model: Work, Span and Parallelism. 1 The RAM model for sequential computation:
CSE341T 08/31/2015 Lecture 3 Cost Model: Work, Span and Parallelism In this lecture, we will look at how one analyze a parallel program written using Cilk Plus. When we analyze the cost of an algorithm
More informationThe number of marks is given in brackets [ ] at the end of each question or part question. The total number of marks for this paper is 72.
ADVANCED SUBSIDIARY GCE UNIT 4736/01 MATHEMATICS Decision Mathematics 1 THURSDAY 14 JUNE 2007 Afternoon Additional Materials: Answer Booklet (8 pages) List of Formulae (MF1) Time: 1 hour 30 minutes INSTRUCTIONS
More informationIE 680 Special Topics in Production Systems: Networks, Routing and Logistics*
IE 680 Special Topics in Production Systems: Networks, Routing and Logistics* Rakesh Nagi Department of Industrial Engineering University at Buffalo (SUNY) *Lecture notes from Network Flows by Ahuja, Magnanti
More information. P. 4.3 Basic feasible solutions and vertices of polyhedra. x 1. x 2
4. Basic feasible solutions and vertices of polyhedra Due to the fundamental theorem of Linear Programming, to solve any LP it suffices to consider the vertices (finitely many) of the polyhedron P of the
More informationSIMPLIFIED PERFORMANCE MODEL FOR HYBRID WIND DIESEL SYSTEMS. J. F. MANWELL, J. G. McGOWAN and U. ABDULWAHID
SIMPLIFIED PERFORMANCE MODEL FOR HYBRID WIND DIESEL SYSTEMS J. F. MANWELL, J. G. McGOWAN and U. ABDULWAHID Renewable Energy Laboratory Department of Mechanical and Industrial Engineering University of
More informationProcess Modelling from Insurance Event Log
Process Modelling from Insurance Event Log P.V. Kumaraguru Research scholar, Dr.M.G.R Educational and Research Institute University Chennai 600 095 India Dr. S.P. Rajagopalan Professor Emeritus, Dr. M.G.R
More informationThe UnionFind Problem Kruskal s algorithm for finding an MST presented us with a problem in datastructure design. As we looked at each edge,
The UnionFind Problem Kruskal s algorithm for finding an MST presented us with a problem in datastructure design. As we looked at each edge, cheapest first, we had to determine whether its two endpoints
More informationCompression algorithm for Bayesian network modeling of binary systems
Compression algorithm for Bayesian network modeling of binary systems I. Tien & A. Der Kiureghian University of California, Berkeley ABSTRACT: A Bayesian network (BN) is a useful tool for analyzing the
More informationINTEGER PROGRAMMING. Integer Programming. Prototype example. BIP model. BIP models
Integer Programming INTEGER PROGRAMMING In many problems the decision variables must have integer values. Example: assign people, machines, and vehicles to activities in integer quantities. If this is
More informationTransportation Polytopes: a Twenty year Update
Transportation Polytopes: a Twenty year Update Jesús Antonio De Loera University of California, Davis Based on various papers joint with R. Hemmecke, E.Kim, F. Liu, U. Rothblum, F. Santos, S. Onn, R. Yoshida,
More informationCurrent Standard: Mathematical Concepts and Applications Shape, Space, and Measurement Primary
Shape, Space, and Measurement Primary A student shall apply concepts of shape, space, and measurement to solve problems involving two and threedimensional shapes by demonstrating an understanding of:
More information6.254 : Game Theory with Engineering Applications Lecture 2: Strategic Form Games
6.254 : Game Theory with Engineering Applications Lecture 2: Strategic Form Games Asu Ozdaglar MIT February 4, 2009 1 Introduction Outline Decisions, utility maximization Strategic form games Best responses
More informationDistributed Dynamic Load Balancing for IterativeStencil Applications
Distributed Dynamic Load Balancing for IterativeStencil Applications G. Dethier 1, P. Marchot 2 and P.A. de Marneffe 1 1 EECS Department, University of Liege, Belgium 2 Chemical Engineering Department,
More informationA Game Theoretical Framework for Adversarial Learning
A Game Theoretical Framework for Adversarial Learning Murat Kantarcioglu University of Texas at Dallas Richardson, TX 75083, USA muratk@utdallas Chris Clifton Purdue University West Lafayette, IN 47907,
More informationLinear Programming for Optimization. Mark A. Schulze, Ph.D. Perceptive Scientific Instruments, Inc.
1. Introduction Linear Programming for Optimization Mark A. Schulze, Ph.D. Perceptive Scientific Instruments, Inc. 1.1 Definition Linear programming is the name of a branch of applied mathematics that
More informationModeling Internet Security Investments Tackling Topological Information Uncertainty
Modeling Internet Security Investments Tackling Topological Information Uncertainty Ranjan Pal Department of Computer Science University of Southern California, USA Joint Work with Pan Hui Deutsch Telekom
More informationDynamic programming formulation
1.24 Lecture 14 Dynamic programming: Job scheduling Dynamic programming formulation To formulate a problem as a dynamic program: Sort by a criterion that will allow infeasible combinations to be eli minated
More informationScheduling Shop Scheduling. Tim Nieberg
Scheduling Shop Scheduling Tim Nieberg Shop models: General Introduction Remark: Consider non preemptive problems with regular objectives Notation Shop Problems: m machines, n jobs 1,..., n operations
More informationECON 459 Game Theory. Lecture Notes Auctions. Luca Anderlini Spring 2015
ECON 459 Game Theory Lecture Notes Auctions Luca Anderlini Spring 2015 These notes have been used before. If you can still spot any errors or have any suggestions for improvement, please let me know. 1
More informationOn the Multiple Unicast Network Coding Conjecture
On the Multiple Unicast Networ Coding Conjecture Michael Langberg Computer Science Division Open University of Israel Raanana 43107, Israel miel@openu.ac.il Muriel Médard Research Laboratory of Electronics
More informationLaboratory work in AI: First steps in Poker Playing Agents and Opponent Modeling
Laboratory work in AI: First steps in Poker Playing Agents and Opponent Modeling Avram Golbert 01574669 agolbert@gmail.com Abstract: While Artificial Intelligence research has shown great success in deterministic
More informationA GraphTheoretic Network Security Game. Marios Mavronicolas, Vicky G. Papadopoulou, Anna Philippou and Paul G. Spirakis
Int. J. Autonomous and Adaptive Communications Systems, Vol. x, No. x, xx 1 A GraphTheoretic Network Security Game Marios Mavronicolas, Vicky G. Papadopoulou, Anna Philippou and Paul G. Spirakis Department
More informationBestresponse play in partially observable card games
Bestresponse play in partially observable card games Frans Oliehoek Matthijs T. J. Spaan Nikos Vlassis Informatics Institute, Faculty of Science, University of Amsterdam, Kruislaan 43, 98 SJ Amsterdam,
More informationIntroduction to Logic in Computer Science: Autumn 2006
Introduction to Logic in Computer Science: Autumn 2006 Ulle Endriss Institute for Logic, Language and Computation University of Amsterdam Ulle Endriss 1 Plan for Today Now that we have a basic understanding
More informationManipulability of the Price Mechanism for Data Centers
Manipulability of the Price Mechanism for Data Centers Greg Bodwin 1, Eric Friedman 2,3,4, and Scott Shenker 3,4 1 Department of Computer Science, Tufts University, Medford, Massachusetts 02155 2 School
More informationON THE COMPLEXITY OF THE GAME OF SET. {kamalika,pbg,dratajcz,hoeteck}@cs.berkeley.edu
ON THE COMPLEXITY OF THE GAME OF SET KAMALIKA CHAUDHURI, BRIGHTEN GODFREY, DAVID RATAJCZAK, AND HOETECK WEE {kamalika,pbg,dratajcz,hoeteck}@cs.berkeley.edu ABSTRACT. Set R is a card game played with a
More informationDynamic Programming. Lecture 11. 11.1 Overview. 11.2 Introduction
Lecture 11 Dynamic Programming 11.1 Overview Dynamic Programming is a powerful technique that allows one to solve many different types of problems in time O(n 2 ) or O(n 3 ) for which a naive approach
More informationExtend Table Lens for HighDimensional Data Visualization and Classification Mining
Extend Table Lens for HighDimensional Data Visualization and Classification Mining CPSC 533c, Information Visualization Course Project, Term 2 2003 Fengdong Du fdu@cs.ubc.ca University of British Columbia
More informationFlorida Math for College Readiness
Core Florida Math for College Readiness Florida Math for College Readiness provides a fourthyear math curriculum focused on developing the mastery of skills identified as critical to postsecondary readiness
More informationCMSC 858T: Randomized Algorithms Spring 2003 Handout 8: The Local Lemma
CMSC 858T: Randomized Algorithms Spring 2003 Handout 8: The Local Lemma Please Note: The references at the end are given for extra reading if you are interested in exploring these ideas further. You are
More informationPersuasion by Cheap Talk  Online Appendix
Persuasion by Cheap Talk  Online Appendix By ARCHISHMAN CHAKRABORTY AND RICK HARBAUGH Online appendix to Persuasion by Cheap Talk, American Economic Review Our results in the main text concern the case
More informationA Constraint Programming based Column Generation Approach to Nurse Rostering Problems
Abstract A Constraint Programming based Column Generation Approach to Nurse Rostering Problems Fang He and Rong Qu The Automated Scheduling, Optimisation and Planning (ASAP) Group School of Computer Science,
More information6.852: Distributed Algorithms Fall, 2009. Class 2
.8: Distributed Algorithms Fall, 009 Class Today s plan Leader election in a synchronous ring: Lower bound for comparisonbased algorithms. Basic computation in general synchronous networks: Leader election
More informationMONITORING AND DIAGNOSIS OF A MULTISTAGE MANUFACTURING PROCESS USING BAYESIAN NETWORKS
MONITORING AND DIAGNOSIS OF A MULTISTAGE MANUFACTURING PROCESS USING BAYESIAN NETWORKS Eric Wolbrecht Bruce D Ambrosio Bob Paasch Oregon State University, Corvallis, OR Doug Kirby Hewlett Packard, Corvallis,
More informationLecture 3: Finding integer solutions to systems of linear equations
Lecture 3: Finding integer solutions to systems of linear equations Algorithmic Number Theory (Fall 2014) Rutgers University Swastik Kopparty Scribe: Abhishek Bhrushundi 1 Overview The goal of this lecture
More informationAlgorithms for Interdependent Security Games
Algorithms for Interdependent Security Games Michael Kearns Luis E. Ortiz Department of Computer and Information Science University of Pennsylvania 1 Introduction Inspired by events ranging from 9/11 to
More informationCluster Analysis: Advanced Concepts
Cluster Analysis: Advanced Concepts and dalgorithms Dr. Hui Xiong Rutgers University Introduction to Data Mining 08/06/2006 1 Introduction to Data Mining 08/06/2006 1 Outline Prototypebased Fuzzy cmeans
More informationCNFSAT: Predictive Models, Dimensional Reduction, and Phase Transition
CNFSAT: Predictive Models, Dimensional Reduction, and Phase Transition Neil P. Slagle College of Computing Georgia Institute of Technology Atlanta, GA npslagle@gatech.edu Abstract CNFSAT embodies the P
More information8 Modeling network traffic using game theory
8 Modeling network traffic using game theory Network represented as a weighted graph; each edge has a designated travel time that may depend on the amount of traffic it contains (some edges sensitive to
More informationA Comparison of General Approaches to Multiprocessor Scheduling
A Comparison of General Approaches to Multiprocessor Scheduling JingChiou Liou AT&T Laboratories Middletown, NJ 0778, USA jing@jolt.mt.att.com Michael A. Palis Department of Computer Science Rutgers University
More informationWhy? A central concept in Computer Science. Algorithms are ubiquitous.
Analysis of Algorithms: A Brief Introduction Why? A central concept in Computer Science. Algorithms are ubiquitous. Using the Internet (sending email, transferring files, use of search engines, online
More information1 Solving LPs: The Simplex Algorithm of George Dantzig
Solving LPs: The Simplex Algorithm of George Dantzig. Simplex Pivoting: Dictionary Format We illustrate a general solution procedure, called the simplex algorithm, by implementing it on a very simple example.
More informationComputational Learning Theory Spring Semester, 2003/4. Lecture 1: March 2
Computational Learning Theory Spring Semester, 2003/4 Lecture 1: March 2 Lecturer: Yishay Mansour Scribe: Gur Yaari, Idan Szpektor 1.1 Introduction Several fields in computer science and economics are
More informationSocial Media Mining. Graph Essentials
Graph Essentials Graph Basics Measures Graph and Essentials Metrics 2 2 Nodes and Edges A network is a graph nodes, actors, or vertices (plural of vertex) Connections, edges or ties Edge Node Measures
More informationLoad Balancing and Rebalancing on Web Based Environment. Yu Zhang
Load Balancing and Rebalancing on Web Based Environment Yu Zhang This report is submitted as partial fulfilment of the requirements for the Honours Programme of the School of Computer Science and Software
More informationChapter 4 DECISION ANALYSIS
ASW/QMBCh.04 3/8/01 10:35 AM Page 96 Chapter 4 DECISION ANALYSIS CONTENTS 4.1 PROBLEM FORMULATION Influence Diagrams Payoff Tables Decision Trees 4.2 DECISION MAKING WITHOUT PROBABILITIES Optimistic Approach
More informationBidding Dynamics of Rational Advertisers in Sponsored Search Auctions on the Web
Proceedings of the International Conference on Advances in Control and Optimization of Dynamical Systems (ACODS2007) Bidding Dynamics of Rational Advertisers in Sponsored Search Auctions on the Web S.
More informationTHREE DIMENSIONAL REPRESENTATION OF AMINO ACID CHARAC TERISTICS
THREE DIMENSIONAL REPRESENTATION OF AMINO ACID CHARAC TERISTICS O.U. Sezerman 1, R. Islamaj 2, E. Alpaydin 2 1 Laborotory of Computational Biology, Sabancı University, Istanbul, Turkey. 2 Computer Engineering
More informationMoral Hazard. Itay Goldstein. Wharton School, University of Pennsylvania
Moral Hazard Itay Goldstein Wharton School, University of Pennsylvania 1 PrincipalAgent Problem Basic problem in corporate finance: separation of ownership and control: o The owners of the firm are typically
More informationGame Theory 1. Introduction
Game Theory 1. Introduction Dmitry Potapov CERN What is Game Theory? Game theory is about interactions among agents that are selfinterested I ll use agent and player synonymously Selfinterested: Each
More informationIndividual security and network design
Individual security and network design Diego Cerdeiro Marcin Dziubiński Sanjeev Goyal FIT 2015 Motivation Networks often face external threats in form of strategic or random attacks The attacks can be
More information