Computational Concentration of Measure: Optimal Bounds, Reductions, and More

Abstract

Product measures of dimension n are known to be "concentrated" under Hamming distance. More precisely, for any set S in the product space of probability Pr[S] ≥ ε, a random point in the space, with probability 1 − δ, has a neighbor in S that is different from the original point in only O( n · ln( 1 /εδ)) coordinates (and this is optimal). In this work, we obtain the tight computational (algorithmic) version of this result, showing how given a random point and access to an S-membership query oracle, we can find such a close point of Hamming distance O( n · ln( 1 /εδ)) in time poly(n, 1/ε, 1/δ). This resolves an open question of [MM19] who proved a weaker result (that works only for ε ≫ 1/ √ n). As corollaries, we obtain polynomial-time poisoning and (in certain settings) evasion attacks against learning algorithms when the original vulnerabilities have any cryptographically non-negligible probability.We call our algorithm MUCIO (short for "MUltiplicative Conditional Influence Optimizer") since proceeding through the coordinates of the product space, it decides to change each coordinate of the given point based on a multiplicative version of the influence of a variable, where the influence is computed conditioned on the value of all previously updated coordinates. MUCIO is an online algorithm in that it decides on the i'th coordinate of the output given only the first i coordinates of the input. It also does not make any convexity assumption about the set S.Motivated by obtaining algorithmic variants of measure concentration in other metric probability spaces, we define a new notion of algorithmic reduction between computational concentration of measure in different probability metric spaces. This notion, whose definition has some subtlety, requires two (inverse) algorithmic mappings one of which is an algorithmic Lipschitz mapping and the other one is an algorithmic coupling connecting the two distributions. As an application, we apply this notion of reduction to obtain computational concentration of measure for high-dimensional Gaussian distributions under the ℓ 1 distance.We further prove several extensions to the results above as follows.(1) Generalizing in another dimension, our computational concentration result is also true when the Hamming distance is weighted.(2) As measure concentration is usually proved for concentration around mean, we show how to use our results above to obtain algorithmic concentration for that setting as well. In particular, we prove a computational variant of McDiarmid's inequality, when properly defined. (3) Our result generalizes to discrete random processes (instead of just product distributions), and this generalization leads to new tampering algorithms for collective coin tossing protocols. (4) Finally, we prove exponential lower bounds on the average running time of non-adaptive query algorithms for proving computational concentration for the case of product spaces. Perhaps surprisingly, such lower bound shows any efficient algorithm must query about S-membership of points tha...

Learning and Certification under Instance-targeted Poisoning

“…There are also previous theoretical works studying instance-targeted poisoning attacks that only use clean labels [Mahloujifar and Mahmoody, 2017, Etesami et al, 2020. The work of Shafahi et al [2018] studied such attacks from a more practical and empirical perspective.…”

A Tight Lower Bound on Adaptively Secure Full-Information Coin Flip

“…Tauman Kalai, Komargodski, and Raz [24] fully answered the single-turn case by proving that no singleturn protocol is resilient to Ω( √ n) adaptive corruptions. Lastly, Etesami, Mahloujifar, and Mahmoody [13] presented an efficient and optimal strongly adaptive attack on protocols of certain properties (e.g., public coins). On a related note, Kalai and Komargodski [18] showed that for any ℓ-party n-round coin-flipping protocol there exists a related ℓ-party n-round protocol of the same communication pattern, output distribution and security guarantees, but of message length polylog(ℓ, n).…”

“…However, the above o(1) stands for 1/ loglog(ℓ), and it remains an intriguing question whether it can be pushed to 2 −polylog(ℓ) as can be achieved, for instance, when attacking the ℓ-party majority protocol. Such attacks are known for uniform single-bit single-turn protocols (a secondary result of [24]) and for strongly adaptive attackers against single-turn protocols [13].…”

Computational Hardness of Collective Coin-Tossing Protocols

“…These hardness of computation results for the coin-tossing functionality extend to other general functionalities as well. Furthermore, the research outcomes for this functionality has direct consequences on diverse topics in mathematics and computer science—for example, extremal graph theory [ 3 , 4 , 5 ], extracting randomness from imperfect sources [ 6 , 7 , 8 , 9 ], cryptography [ 10 , 11 , 12 , 13 , 14 , 15 , 16 ], game theory [ 17 , 18 ], circuit representation [ 19 , 20 , 21 ], distributed protocols [ 22 , 23 , 24 ], and poisoning and evasion attacks on learning algorithms [ 25 , 26 , 27 , 28 ].…”