## Simple Strategies for Large Zero-Sum Games with Applications to Complexity Theory (1994)

Venue: | STOC 94 |

Citations: | 23 - 2 self |

### BibTeX

@MISC{Lipton94simplestrategies,

author = {Richard J. Lipton and Neal E. Young},

title = {Simple Strategies for Large Zero-Sum Games with Applications to Complexity Theory},

year = {1994}

}

### Years of Citing Articles

### OpenURL

### Abstract

Von Neumann’s Min-Max Theorem guarantees that each player of a zero-sum matrix game hss an optimal mixed strategy. We show that each player has a near-optimal mixed strategy that chooses uniformly from a multiset of pure strategies of size logarithmic in the number of pure strategies available to the opponent. Thus, for exponentially large games, for which even representing an optimal mixed strategy can require exponential space, there are nearoptimal, linear-size strategies. These strategies are eaay to play and serve as small witnesses to the approximate value of the game. Because of the fundamental role of games, we expect this theorem to have many applications in complexity theory and cryptography. We use it to strengthen the connection estab-lished by Yao between randomized and distributional complexity and to obtain the following results: (1) Every language has anti-checkers — small hard multisets of inputs certifying that small circuits can’t decide the language. (2) Circuits of a given size can generate random instances that are hard for all circuits of linearly smaller size. (3) Given an oracle M for any exponentially large game, the approximate value of the game and near-optimal strategies for it can be computed in I&‘(M). (4) For any NP-complete lan-guage L, the problems of (a) computing a hard distribution of instances of L and (b) estimating the circuit complexity of L are both in Z;.

### Citations

1486 | Probability inequalities for sums of bounded random variables
- Hoeffding
- 1963
(Show Context)
Citation Context ...y that X i2S 1 jSj M ijsV(M ) + ffl (1) is bounded by e \Gamma2kffl 2 . This is because the left-hand side is the average of k independent random variables in [0; 1] with expected value at most V(M ) =-=[11]-=-. By the choice of k, e \Gamma2kffl 2 ! 1=c. Thus, the expected number of the opponent's c pure strategies that satisfy (1) is less than 1. Since the number of such strategies is an integer, it must b... |

1032 | The Knowledge Complexity of Interactive Proof Systems
- Goldwasser, Micali, et al.
- 1989
(Show Context)
Citation Context ...ction Games play a fundamental role in many parts of theory. For example, cryptographic problems can often be viewed as games between those who wish to keep a secret and those who wish to discover it =-=[8]-=-. Many computational classes can be defined in natural ways as games: for example, PSPACE can be defined in this way [6]. Other times games arise in a slightly more subtle way. For example, questions ... |

665 | The strength of weak learnability
- Schapire
- 1990
(Show Context)
Citation Context ...t applies to uniform complexity classes, not circuits. More precisely, it generates inputs that are hard only for programs that are exponentially smaller than the inputs. Circuit complexity. Schapire =-=[14]-=- shows that his technique for boosting the correctness of PAC-learning strategies can also be applied to boost the correctness of circuits. This implies the existence of distributions from which rando... |

302 | Designing programs that check their work
- Blum, Kannan
- 1995
(Show Context)
Citation Context ...rcuits of size s is a multiset of inputs such that any circuit of size s fails to correctly classify (w.r.t. L) a fraction of the inputs in the multiset. Anti-checkers are similar to program checkers =-=[5]-=- (which verify program correctness on a perinputsbasis) in that anti-checkers allow certification of the complexity of L on a per-circuit basis. We apply Corollary 4 is to show that, provided s is sli... |

115 | A Graph-Theoretic Game and its Application to the k-Server Problem
- Alon, Karp, et al.
- 1995
(Show Context)
Citation Context ... One example of the use of the MinMax Theorem in the stronger direction (to upper bound randomized complexity, measured, in this case, by the competitive ratio) is given by Alon, Karp, Peleg and West =-=[2]-=-. They show the existence of randomized kserver strategies by considering a certain zero-sum matrix game. The competitiveness of the strategy is related to the value of the game, which in turn depends... |

106 | On the theory of average case complexity
- Ben-David, Chor, et al.
- 1989
(Show Context)
Citation Context ...can be used in these characterizations of NEXP and coNEXP. Most research on hard distributions to date concerns uniform complexity. A significant body of work concerns average-case completeness, e.g, =-=[9, 16, 10, 12, 4]-=-. These results are analogous to NP-completeness results, except they concern distributional problems --- hproblem, input distributioni pairs. These results relate the complexities of classes of distr... |

95 |
Two theorems on random polynomial time
- Adleman
- 1978
(Show Context)
Citation Context ...from a multiset of size logarithmic in the number of pure strategies available to the opponent. The proof is a surprisingly simple probabilistic argument similar to circuit derandomization techniques =-=[1, 15]-=-. However, the central nature of games in theory suggests that this simple result may have far-reaching consequences. This result was obtained independently by Althofer [3]. Strengthening the connecti... |

90 |
bounds by probabilistic arguments
- Lower
- 1983
(Show Context)
Citation Context ...games arise in a slightly more subtle way. For example, questions about how hard it is to generate hard instances of some problem can be modeled as a game between the generator and the algorithm. Yao =-=[18, 19]-=- exploits this idea to prove lower bounds on randomized algorithms. The classic result on games is the famous Min-Max Theorem of von Neumann [17], which guarantees that each player of a zero-sum game ... |

62 |
No Better Ways to Generate Hard NP Instances than Picking Uniformly at Random
- Impagliazzo, Levin
(Show Context)
Citation Context ...can be used in these characterizations of NEXP and coNEXP. Most research on hard distributions to date concerns uniform complexity. A significant body of work concerns average-case completeness, e.g, =-=[9, 16, 10, 12, 4]-=-. These results are analogous to NP-completeness results, except they concern distributional problems --- hproblem, input distributioni pairs. These results relate the complexities of classes of distr... |

56 | Learning simple concept under simple distributions
- Li, Vitányi
- 1991
(Show Context)
Citation Context .... These results relate the complexities of classes of distributional problems. Generally, few relations to worst-case complexity are known (see, however, [4]). Ben-David et al. [4] and Li and Vitanyi =-=[13]-=- show the existence of distributions under which the average-case complexity of any program is within a constant (exponential in the size of the program) of the worst-case complexity. Generating rando... |

42 |
Probabilistic complexity classes and lowness
- Schöning
- 1989
(Show Context)
Citation Context ...from a multiset of size logarithmic in the number of pure strategies available to the opponent. The proof is a surprisingly simple probabilistic argument similar to circuit derandomization techniques =-=[1, 15]-=-. However, the central nature of games in theory suggests that this simple result may have far-reaching consequences. This result was obtained independently by Althofer [3]. Strengthening the connecti... |

40 |
Random Instances of a Graph Coloring Problem are
- Venkatesan, Levin
- 1988
(Show Context)
Citation Context ...can be used in these characterizations of NEXP and coNEXP. Most research on hard distributions to date concerns uniform complexity. A significant body of work concerns average-case completeness, e.g, =-=[9, 16, 10, 12, 4]-=-. These results are analogous to NP-completeness results, except they concern distributional problems --- hproblem, input distributioni pairs. These results relate the complexities of classes of distr... |

20 |
Zur Theorie der Gesellschaftspiele
- Neumann
(Show Context)
Citation Context ... game between the generator and the algorithm. Yao [18, 19] exploits this idea to prove lower bounds on randomized algorithms. The classic result on games is the famous Min-Max Theorem of von Neumann =-=[17]-=-, which guarantees that each player of a zero-sum game has an optimal mixed strategy. For exponentially large games, optimal strategies are generally exponentially large. In many cases, we need to kno... |

15 |
Matrix decomposition problem is complete for the average case
- Gurevich
- 1990
(Show Context)
Citation Context |

14 |
Complete and Incomplete Randomized NP Problems
- Gurevich
- 1987
(Show Context)
Citation Context |

7 |
Probabilistic complexity: towards a unified measure of complexity
- Yao
- 1977
(Show Context)
Citation Context ...games arise in a slightly more subtle way. For example, questions about how hard it is to generate hard instances of some problem can be modeled as a game between the generator and the algorithm. Yao =-=[18, 19]-=- exploits this idea to prove lower bounds on randomized algorithms. The classic result on games is the famous Min-Max Theorem of von Neumann [17], which guarantees that each player of a zero-sum game ... |

6 |
private communication
- Feigenbaum
- 1993
(Show Context)
Citation Context ...form complexity. As mentioned previously, the complexity class PSPACE has a natural characterization via games. More recently, the complexity classes NEXP and coNEXP have been similarly characterized =-=[7]-=-. Our variant of von Neumann's Theorem can be used in these characterizations of NEXP and coNEXP. Most research on hard distributions to date concerns uniform complexity. A significant body of work co... |

6 | Zur Theorie der Gesellschaftspiele. Mathematische Annalen 100:295–320 - Neumann - 1928 |

2 | Greedy algorithms by derandomizing unknown distributions
- Young
- 1485
(Show Context)
Citation Context ...riant of von Neumann's MinMax Theorem, was obtained independently by Althofer [3]. He considers applications to other linear programs, large game trees, and uniform sampling spaces. A subsequent work =-=[20]-=- gives simple greedy algorithms that (given the payoff matrix) find the k-uniform strategies shown to exist in Theorems 2 and 3. Uniform complexity. As mentioned previously, the complexity class PSPAC... |

1 |
On sparse approximations to randomized strategies and covex combinations. Linear Algebra and its Applications
- Althofer
- 1994
(Show Context)
Citation Context ...andomization techniques [1, 15]. However, the central nature of games in theory suggests that this simple result may have far-reaching consequences. This result was obtained independently by Althofer =-=[3]-=-. Strengthening the connection between randomized and distributional complexities. This connection was first established by Yao [18, 19]. He considered a game where Min's pure strategies are the deter... |

1 | Random inst antes of a graph coloring problem are hard - Venkatesan, Levin - 1988 |