## Hardness amplification proofs require majority (2008)

### Cached

### Download Links

Venue: | In Proceedings of the 40th Annual ACM Symposium on the Theory of Computing (STOC |

Citations: | 11 - 4 self |

### BibTeX

@INPROCEEDINGS{Shaltiel08hardnessamplification,

author = {Ronen Shaltiel and Emanuele Viola},

title = {Hardness amplification proofs require majority},

booktitle = {In Proceedings of the 40th Annual ACM Symposium on the Theory of Computing (STOC},

year = {2008}

}

### OpenURL

### Abstract

Hardness amplification is the fundamental task of converting a δ-hard function f: {0, 1} n → {0, 1} into a (1/2 − ɛ)-hard function Amp(f), where f is γ-hard if small circuits fail to compute f on at least a γ fraction of the inputs. Typically, ɛ, δ are small (and δ = 2 −k captures the case where f is worst-case hard). Achieving ɛ = 1/n ω(1) is a prerequisite for cryptography and most pseudorandom-generator constructions. In this paper we study the complexity of black-box proofs of hardness amplification. A class of circuits D proves a hardness amplification result if for any function h that agrees with Amp(f) on a 1/2 + ɛ fraction of the inputs there exists an oracle circuit D ∈ D such that D h agrees with f on a 1 − δ fraction of the inputs. We focus on the case where every D ∈ D makes non-adaptive queries to h. This setting captures most hardness amplification techniques. We prove two main results: 1. The circuits in D “can be used ” to compute the majority function on 1/ɛ bits. In particular, these circuits have large depth when ɛ ≤ 1/poly log n. 2. The circuits in D must make Ω � log(1/δ)/ɛ 2 � oracle queries. Both our bounds on the depth and on the number of queries are tight up to constant factors.

### Citations

405 | Non-Deterministic Exponential Time has TwoProver Interactive Protocols
- Babai, Fortnow, et al.
- 1990
(Show Context)
Citation Context ...g line of research has focused on hardness amplification. This is the task of transforming worst-case hard functions (or sometimes mildly average-case hard functions) into average-case hard functions =-=[65, 40, 8, 6, 7, 32, 20, 15, 35, 36, 12, 55, 59, 51, 56, 47, 60, 58, 29, 52, 24, 33, 34, 22]-=-. This research was largely successful in its goal. In particular, it provided worst-case to average-case connections within many complexity classes. Many of these connections give strongly average-ca... |

381 | A hard-core predicate for all one-way functions
- Goldreich, Levin
- 1989
(Show Context)
Citation Context ... by proving the direct product first and then transforming the amplified function f ◦t into a Boolean function (see, e.g., [20, 35, 55, 47, 56, 29]), often using the remarkable GoldreichLevin Theorem =-=[19]-=-. The converse, proving a direct product lemma from an XOR lemma, is much easier [64]. By contrast, our results show that Yao’s XOR lemma and the direct product lemma are qualitatively different. The ... |

319 | A parallel repetition theorem
- Raz
- 1998
(Show Context)
Citation Context ..., . . . , Viq ) “does not change significantly” when conditioning on E. This lemma can be viewed as a generalization of a Lemma by Raz (in which q = 1) that is used in his parallel repetition theorem =-=[48]-=-. We have recently found out that this lemma follows easily from the results in [14, Section 4]. We apply the lemma to the random variables {BN(y)} y∈{0,1} n. We conclude that there exists a large set... |

288 | Algebraic methods in the theory of lower bounds for Boolean circuit complexity
- Smolensky
- 1987
(Show Context)
Citation Context ...ity theory has produced many exciting and useful lower bounds for restricted computational models, most notably against classes of circuits with unbounded fan-in and constant depth with various gates =-=[16, 66, 27, 49, 54, 28, 25, 5, 26]-=-. In some of these classes we in fact can prove worstcase lower bounds, but cannot prove strongly average-case lower bounds (e.g. [49, 54, 5]). Several such examples are surveyed in the full version o... |

282 | Hardness vs. randomness
- Nisan, Wigderson
- 1994
(Show Context)
Citation Context ...cuits fail to compute on many inputs. Average-case hard functions are especially important as they are a prerequisite for most modern cryptography and can be used to construct pseudorandom generators =-=[46]-=- which in turn have a striking variety of applications (see, e.g., the books by Goldreich [17, 18]). We stress that both these applications require strongly average-case hard functions. That is functi... |

247 | Elements of Information Theory. Wiley Series in Telecommunications - Cover, Thomas - 1991 |

221 |
Parity, circuits and the polynomial-time hierarchy
- Furst, Saxe, et al.
- 1984
(Show Context)
Citation Context ...ity theory has produced many exciting and useful lower bounds for restricted computational models, most notably against classes of circuits with unbounded fan-in and constant depth with various gates =-=[16, 66, 27, 49, 54, 28, 25, 5, 26]-=-. In some of these classes we in fact can prove worstcase lower bounds, but cannot prove strongly average-case lower bounds (e.g. [49, 54, 5]). Several such examples are surveyed in the full version o... |

185 |
Computational limitations of small-depth circuits
- H˚astad
- 1987
(Show Context)
Citation Context ...ity theory has produced many exciting and useful lower bounds for restricted computational models, most notably against classes of circuits with unbounded fan-in and constant depth with various gates =-=[16, 66, 27, 49, 54, 28, 25, 5, 26]-=-. In some of these classes we in fact can prove worstcase lower bounds, but cannot prove strongly average-case lower bounds (e.g. [49, 54, 5]). Several such examples are surveyed in the full version o... |

185 |
Separating the polynomial-time hierarchy by oracles
- Yao
- 1985
(Show Context)
Citation Context |

179 |
P=BPP if E requires exponential circuits: Derandomizing the XOR Lemma
- Impagliazzo, Wigderson
- 1997
(Show Context)
Citation Context ...g line of research has focused on hardness amplification. This is the task of transforming worst-case hard functions (or sometimes mildly average-case hard functions) into average-case hard functions =-=[65, 40, 8, 6, 7, 32, 20, 15, 35, 36, 12, 55, 59, 51, 56, 47, 60, 58, 29, 52, 24, 33, 34, 22]-=-. This research was largely successful in its goal. In particular, it provided worst-case to average-case connections within many complexity classes. Many of these connections give strongly average-ca... |

176 | Natural Proofs
- Razborov, Rudich
- 1997
(Show Context)
Citation Context ...nt-depth circuits, black-box, natural proofs. 1. INTRODUCTION Proving circuit lower bounds is a major goal of Complexity Theory. However, the celebrated “Natural Proofs” result by Razborov and Rudich =-=[50]-=-, coupled with the pseudorandom functions by Naor and Reingold [45], marks the class of polynomial-size constant-depth circuits with majority gates (TC 0 ) as a fundamental limit for most currently av... |

157 |
Hiding instances in multioracle queries
- BEAVER, FEIGENBAUM
- 1990
(Show Context)
Citation Context ...g line of research has focused on hardness amplification. This is the task of transforming worst-case hard functions (or sometimes mildly average-case hard functions) into average-case hard functions =-=[65, 40, 8, 6, 7, 32, 20, 15, 35, 36, 12, 55, 59, 51, 56, 47, 60, 58, 29, 52, 24, 33, 34, 22]-=-. This research was largely successful in its goal. In particular, it provided worst-case to average-case connections within many complexity classes. Many of these connections give strongly average-ca... |

152 | Number-theoretic constructions of efficient pseudo-random functions
- Naor, Reingold
- 1997
(Show Context)
Citation Context ...ng circuit lower bounds is a major goal of Complexity Theory. However, the celebrated “Natural Proofs” result by Razborov and Rudich [50], coupled with the pseudorandom functions by Naor and Reingold =-=[45]-=-, marks the class of polynomial-size constant-depth circuits with majority gates (TC 0 ) as a fundamental limit for most currently available lower bounding techniques. This limitation already applies ... |

128 | Pseudorandom generators without the XOR lemma
- Sudan, Trevisan, et al.
- 1999
(Show Context)
Citation Context |

127 |
Threshold circuits of bounded depth
- Hajnal, Maass, et al.
- 1993
(Show Context)
Citation Context |

125 | Modern Cryptography, Probabilistic Proofs and Pseudorandomness
- Goldreich
- 1998
(Show Context)
Citation Context ...they are a prerequisite for most modern cryptography and can be used to construct pseudorandom generators [46] which in turn have a striking variety of applications (see, e.g., the books by Goldreich =-=[17, 18]-=-). We stress that both these applications require strongly average-case hard functions. That is functions that small circuits cannot computewith even a small advantage over random guessing, for a ran... |

114 | BPP has subexponential time simulations unless EXPTIME has publishable proofs
- Babai, Fortnow, et al.
- 1993
(Show Context)
Citation Context |

112 | Hard-core distributions for somewhat hard problems
- Impagliazzo
- 1995
(Show Context)
Citation Context |

110 |
Σ 1 1-formulae on finite structures
- Ajtai
- 1983
(Show Context)
Citation Context ...putes majority. For this we also exploit that the relevant probabilities in the above reduction are sufficiently bounded away that they can be amplified using circuits of constant-depth by the result =-=[Ajt1]-=- (see also [Ajt2, Vio5]). 2.1 Proving the Zoom Theorem when D contains a single circuit The proof of the Zoom Theorem is the main technical contribution of this paper. What makes this problem challeng... |

108 | Simple Extractors for All Min-Entropies and a New Pseudo-Random Generator
- Shaltiel, Umans
- 2001
(Show Context)
Citation Context |

103 | On the power of small-depth threshold circuits
- H̊astad, Goldmann
- 1991
(Show Context)
Citation Context |

102 | Multiparty protocols, pseudorandom generators for logspace, and time-space trade-offs - Babai, Nisan, et al. - 1992 |

102 |
New directions in testing
- LIPTON
- 1991
(Show Context)
Citation Context |

101 |
The Foundations of Cryptography, volume 1 (Basic Tools
- Goldreich
- 2001
(Show Context)
Citation Context ...they are a prerequisite for most modern cryptography and can be used to construct pseudorandom generators [46] which in turn have a striking variety of applications (see, e.g., the books by Goldreich =-=[17, 18]-=-). We stress that both these applications require strongly average-case hard functions. That is functions that small circuits cannot computewith even a small advantage over random guessing, for a ran... |

100 |
Theory and applications of trapdoor functions (extended abstract
- Yao
- 1982
(Show Context)
Citation Context |

91 | The expressive power of voting polynomials
- Aspnes, Beigel, et al.
- 1994
(Show Context)
Citation Context ...[Kli] stands out. Exploiting precisely the fact that computing majority is all that is needed for hardness amplification, Klivans uses a lower bound for constant-depth circuits with one majority gate =-=[ABFR]-=- to give an alternative proof of the strong average-case hardness of parity for constant-depth circuits without majority gates. We remark that [Kli] does not contradict the results in this paper, but ... |

85 | 1 -formulae on finite structures - Ajtai - 1983 |

73 | Randomness vs Time: Derandomization under a Uniform Assumption
- Impagliazzo, Wigderson
- 2001
(Show Context)
Citation Context |

69 | The polynomial method in circuit complexity - Beigel - 1993 |

57 | On Yao’s XOR-lemma
- Goldreich, Nisan, et al.
- 1995
(Show Context)
Citation Context |

55 | Pseudorandomness and average-case complexity via uniform reductions
- Trevisan, Vadhan
- 2002
(Show Context)
Citation Context |

51 | On worst-case to average-case reductions for NP problems
- Bogdanov, Trevisan
- 2003
(Show Context)
Citation Context ...ication: There are other works that study different, less demanding models of hardness amplification which are tailored to important questions such as worst-case to average-case connections within NP =-=[10, 9, 61]-=-. These works are incomparable with ours, one key difference being that they impose computational restrictions on the starting function f and the amplified function Amp(f), whereas our results do not.... |

50 | Some applications of coding theory in computational complexity
- Trevisan
(Show Context)
Citation Context ...hin many complexity classes. Many of these connections give strongly average-case hard functions. This research also spurred fruitful interaction with coding theory (see, e.g., the survey by Trevisan =-=[57]-=-). Complexity theory has produced many exciting and useful lower bounds for restricted computational models, most notably against classes of circuits with unbounded fan-in and constant depth with vari... |

47 | A theorem on probabilistic constant depth computations - Ajtai, Ben-Or - 1984 |

45 | Hardness amplification within NP
- O’Donnell
- 2002
(Show Context)
Citation Context |

38 | Boosting and hard-core sets
- Klivans, Servedio
(Show Context)
Citation Context ...between the circuit sizes that come up in the assumption and conclusion of the hardness amplification theorem. The question of how much loss is necessary has been raised a number of times (see, e.g., =-=[20, 37]-=-) but was never answered in generality until this paper. Additional motivation is given in the full version of this paper. Theorem 1.4 (Decoding requires many queries). There is a universal constant C... |

35 | The Complexity of Constructing Pseudorandom Generators from Hard Functions
- Viola
- 2004
(Show Context)
Citation Context |

34 | On the hardness of computing the permanent of random matrices
- Feige, Lund
- 1992
(Show Context)
Citation Context |

29 | Lower bounds for approximations by low degree polynomials over DFE
- Alon, Beigel
- 2001
(Show Context)
Citation Context ...(f), whereas our results do not. Finally, we would like to mention that there is a long line of research that is devoted to proving average-case hardness results for circuit classes below TC 0 , e.g. =-=[27, 25, 38, 4, 11, 23, 64]-=-. With a few exceptions (discussed below) this research has been independent of hardness amplification, and our results may be interpreted as a partial explanation for this independence. The work by K... |

28 | Using nondeterminism to amplify hardness
- Healy, Vadhan, et al.
- 2004
(Show Context)
Citation Context |

27 | Approximately list-decoding direct product codes and uniform hardness amplification - Impagliazzo, Jaiswal, et al. - 2006 |

26 | On uniform amplification of hardness in NP
- Trevisan
- 2005
(Show Context)
Citation Context |

25 |
Approximate counting with uniform constant-depth circuits
- Ajtai
- 1990
(Show Context)
Citation Context .... For this we also exploit that the relevant probabilities in the above reduction are sufficiently bounded away that they can be amplified using circuits of constant-depth by the result [2] (see also =-=[3, 63]-=-). 2.1 Proving the Zoom Theorem when D contains a single circuit The proof of the Zoom Theorem is the main technical contribution of this paper. What makes this problem challenging is that the class D... |

24 | When do extra majority gates help? polylog(N) majority gates are equivalent to one
- Beigel
- 1994
(Show Context)
Citation Context ... ≥ 1/ɛ by our Theorem 1.4, we obtain that when ɛ ≤ 1/n we must start from a function that is hard for constant-depth circuits with n majority gates, i.e. TC 0 . (Our current lower bounding techniques =-=[Bei2]-=- let us handle circuits with poly log majority gates, and this in turn lets us amplify hardness up to 1/2 − 1/poly log.) A final remark about this class is in order. For the specific hardness amplific... |

24 | Pseudorandomness for approximate counting and sampling
- Shaltiel, Umans
- 2005
(Show Context)
Citation Context |

23 |
Estimates of certain exponential sums arising in complexity theory
- Bourgain
(Show Context)
Citation Context ...(f), whereas our results do not. Finally, we would like to mention that there is a long line of research that is devoted to proving average-case hardness results for circuit classes below TC 0 , e.g. =-=[27, 25, 38, 4, 11, 23, 64]-=-. With a few exceptions (discussed below) this research has been independent of hardness amplification, and our results may be interpreted as a partial explanation for this independence. The work by K... |

22 | Norms, XOR lemmas, and lower bounds for gf(2) polynomials and multiparty protocols
- Viola, Wigderson
(Show Context)
Citation Context ...t into a Boolean function (see, e.g., [20, 35, 55, 47, 56, 29]), often using the remarkable GoldreichLevin Theorem [19]. The converse, proving a direct product lemma from an XOR lemma, is much easier =-=[64]-=-. By contrast, our results show that Yao’s XOR lemma and the direct product lemma are qualitatively different. The main difference is that the proof of Yao’s XOR lemma requires majority, whereas the p... |

21 |
Bounds on an exponential sum arising in boolean circuit complexity
- Green, Roy, et al.
(Show Context)
Citation Context ...(f), whereas our results do not. Finally, we would like to mention that there is a long line of research that is devoted to proving average-case hardness results for circuit classes below TC 0 , e.g. =-=[27, 25, 38, 4, 11, 23, 64]-=-. With a few exceptions (discussed below) this research has been independent of hardness amplification, and our results may be interpreted as a partial explanation for this independence. The work by K... |

19 | On the hardness of the permanent
- Cai, Pavan, et al.
- 1999
(Show Context)
Citation Context |

19 | Lower bounds for sampling algorithms for estimating the average - Canetti, Even, et al. - 1995 |