Hardness of Reconstructing Multivariate Polynomials over Finite Fields

In this paper, we give the first construction of a pseudorandom generator, with seed length O(log n), for CC0[p], the class of constant-depth circuits with unbounded fan-in MOD p gates, for some prime p. More accurately, the seed length of our generator is O(log n) for any constant error $${\epsilon > 0}$$ . In fact, we obtain our generator by fooling distributions generated by low-degree polynomials, over $${\mathbb{F}_p}$$ , when evaluated on the Boolean cube. This result significantly extends previous constructions that either required a long seed (Luby et al. 1993) or could only fool the distribution generated by linear functions over $${\mathbb{F}_p}$$ , when evaluated on the Boolean cube (Lovett et al. 2009; Meka & Zuckerman 2009). En route of constructing our PRG, we prove two structural results for low-degree polynomials over finite fields that can be of independent interest.

Abstract: The objective of this technical paper was to propose an approach for solving polynomials of degree higher than two. The main concepts were the decomposition of a polynomial of a higher degree to the product of two polynomials of lower degrees and the n-D Newton-Raphson method for a system of nonlinear equations. The coefficient of each term in an original polynomial of order m will be equated to the corresponding term from the collected-expanded product of the two polynomials of the lower degrees based on the concept of undetermined coefficients. Consequently a system of nonlinear equations was formed. Then the unknown coefficients of the decomposed polynomial of the lower degree of the two decomposed polynomials would be eliminated from the system of the nonlinear equations. After that the unknown coefficients in the decomposed polynomial of the higher degree would be obtained by the n-D Newton-Raphson. Finally the unknown coefficients for the decomposed polynomial of the lower degree would be obtained by back substitutions. In this technical paper the formulations for the decomposed polynomials would be derived for the polynomials degree three to nine. Several numerical examples were also given to verify the applicability of the proposed approach.

Characterizations of locally testable linear- and affine-invariant families

We give an explicit construction of pseudorandom generators against low degree polynomials over finite fields. We show that the sum of 2d small-biased generators with error e2O(d) is a pseudorandom generator against degree d polynomials with error e. This gives a generator with seed length 2O(d) log(n/e). Our construction follows the recent breakthrough result of Bogadnov and Viola. Their work shows that the sum of d small-biased generators is a pseudo-random generator against degree d polynomials, assuming the Inverse Gowers Conjecture. However, this conjecture is only proven for d=2,3. The main advantage of our work is that it does not rely on any unproven conjectures.

In this paper we develop a new approach for learning decision trees and multivariate polynomials via interpolation of multivariate polynomials. This new approach yields simple learning algorithms for multivariate polynomials and decision trees over finite fields under any constant bounded product distribution. The output hypothesis is a (single) multivariate polynomial that is an /spl epsiv/-approximation of the target under any constant bounded product distribution. The new approach demonstrates the learnability of many classes under any constant bounded product distribution and using membership queries, such as j-disjoint DNF and multivariate polynomial with bounded degree over any field. The technique shows how to interpolate multivariate polynomials with bounded term size from membership queries only. This in particular gives a learning algorithm for O(log n)-depth decision tree from membership queries only and a new learning algorithm of any multivariate polynomial over sufficiently large fields from membership queries only. We show that our results for learning from membership queries only are the best possible.

Locally decodable codes (LDCs) have played a central role in many recent results in theoretical computer science. The role of finite fields, and in particular, low-degree polynomials over finite fields, in the construction of these objects is well studied. However the role of group homomorphisms in the construction of such codes is not as widely studied. Here we initiate a systematic study of local decoding of codes based on group homomorphisms. We give an efficient list decoder for the class of homomorphisms from any abelian group G to a fixed abelian group H. The running time of this algorithm is bounded by a polynomial in log|G| and an agreement parameter, where the degree of the polynomial depends on H. Central to this algorithmic result is a combinatorial result bounding the number of homomorphisms that have large agreement with any function from G to H. Our results give a new generalization of the classical work of Goldreich and Levin, and give new abstractions of the list decoder of Sudan, Trevisan and Vadhan. As a by-product we also derive a simple(r) proof of the local testability (beyond the Blum-Luby-Rubinfeld bounds) of homomorphisms mapping \({\mathbb{Z}}_p^n\) to ℤ p , first shown by M. Kiwi.KeywordsAbelian GroupGroup HomomorphismOracle AccessDecodable CodeList DecoderThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

A new approach to the construction of optimal designs

A heuristic algorithm for testing absolute irreducibility of multivariate polynomials over arbitrary fields using Newton polytopes was proposed in Gao and Lauder (Discrete Comput. Geom. 26:89–104, [2001]). A preliminary implementation by Gao and Lauder (2003) established a wide range of families of low degree and sparse polynomials for which the algorithm works efficiently and with a high success rate. In this paper, we develop a BSP variant of the absolute irreducibility testing via polytopes, with the aim of producing a more memory and run-time efficient method that can provide wider ranges of applicability, specifically in terms of the degrees of the input polynomials. In the bivariate case, we describe a balanced load scheme and a corresponding data distribution leading to a parallel algorithm whose efficiency can be established under reasonably realistic conditions. This is later incorporated in a doubly parallel algorithm in the multivariate case that achieves similar scalable performance. Both parallel models are analyzed for efficiency, and the theoretical analysis is compared to the performance of our experiments. In the empirical results we report, we achieve absolute irreducibility testing for bivariate and trivariate polynomials of degrees up to 30,000, and for low degree multivariate polynomials with more than 3,000 variables. To the best of our knowledge, this sets a world record in establishing absolute irreducibility of multivariate polynomials.

Factoring multivariate polynomials over finite fields

Identification schemes based on multivariate polynomials have been receiving attraction in different areas due to the quantum secure property. Identification is one of the most important elements for the IoT to achieve communication between objects, gather and share information with each other. Thus, identification schemes which are post-quantum secure are significant for Internet-of-Things (IoT) devices. Various polar forms of multivariate quadratic and cubic polynomial systems have been proposed for these identification schemes. There is a need to define polar form for multivariate dth degree polynomials, where . In this paper, we propose a solution to this need by defining constructions for multivariate polynomials of degree . We give a generic framework to construct the identification scheme for IoT and RFID applications. In addition, we compare identification schemes and curve-based cryptoGPS which is currently used in RFID applications.

We show that if F and G are polynomials defined over a p-adic field with gcd(F, G) = 1, then the problem of finding a nonzero nonsingular zero of F that is not a zero of G is equivalent to the problem of finding a nonsingular zero of the homogenization of F. In addition, we prove the existence of p-adic zeros of some polynomials of low degree that are not necessarily homogeneous. This extends some well-known results on the existence of p-adic zeros of homogeneous polynomials of low degree.

We present a new approach to the problem of computing the zeta function of a hypersurface over a finite field. For a hypersurface defined by a polynomial of degree d in n variables over the field of q elements, one desires an algorithm whose running time is a polynomial function of dn log(q). (Here we assume d ⩾ 2, for otherwise the problem is easy.) The case n = 1 is related to univariate polynomial factorisation and is comparatively straightforward. When n = 2 one is counting points on curves, and the method of Schoof and Pila yields a complexity of log ⁡ ( q ) C d , where the function Cd depends exponentially on d. For arbitrary n, the theorem of the author and Wan gives a complexity which is a polynomial function of (pdn log(q))n, where p is the characteristic of the field. A complexity estimate of this form can also be achieved for smooth hypersurfaces using the method of Kedlaya, although this has only been worked out in full for curves. The new approach we present should yield a complexity which is a small polynomial function of pdn log(q). In this paper, we work this out in full for Artin–Schreier hypersurfaces defined by equations of the form Zp − Z = f, where the polynomial f has a diagonal leading form. The method utilises a relative p-adic cohomology theory for families of hypersurfaces, due in essence to Dwork. As a corollary of our main theorem, we obtain the following curious result. Let f be a multivariate polynomial with integer coefficients whose leading form is diagonal. There exists an explicit deterministic algorithm which takes as input a prime p, outputs the number of solutions to the congruence equation f = 0 op, and runs in {\mathcal O}(p^{2 + \varepsilon}) bit operations, for any ε > 0. This improves upon the elementary estimate of {\mathcal O}(p^{n-1 + \varepsilon}) bit operations, where n is the number of variables, which can be achieved using Berlekamp's root counting algorithm. 2000 Mathematics Subject Classification 11Y99, 11M38, 11T99.

The uniform approximation of polynomials by polynomials of lower degree

A special family of polynomials is investigated by which, on the unit-sphere, the problem of best approximation in the mean to monomials by polynomials of lower degree can be solved. Vice versa, these polynomials are extremal with respect to their leading coefficient. For homogeneous harmonic polynomials, a similar problem can be treated and a bound for the coefficients is obtained, which improves an earlier result of Kellogg essentially.

Построение интегральной характеристики системы по ряду наблюдений, фиксирующих состояние системы с некоторыми ошибками, можно рассматривать как задачу выделения полезного сигнала на фоне шумов. Сигналом в этом случае являются весовые коэффициенты линейной свертки показателей, которые должны отражать постоянную структуру оцениваемой системы в рассматриваемом периоде. Для ряда наблюдений методы многомерного анализа по-разному определяют структуру главных компонент (главных факторов). Даже небольшое возмущение исходных данных может вызвать значительное изменение весовых коэффициентов при использовании методов многомерного анализа. Причиной этого может являться наличие неустранимых погрешностей используемых данных. Поэтому существенным является вопрос об оценке качества используемых данных в задаче вычисления композитного индикатора качества системы по ряду наблюдений, а именно, оценка наличия в регистрируемых измерениях значительных неустранимых ошибок. В статье рассматривается применение метода конечных разностей для оценки неустранимых ошибок регистрации данных по ряду наблюдений. Для применения этой методики необходимо, чтобы данные допускали приближение полиномами более низких степеней, чем число наблюдений минус один. Справедливость предположения проверена эмпирически на конкретном наборе данных. Рассматриваются 37 переменных, характеризующих качество жизни населения России за 2010-2017 годы. Проанализированы зависимости качества аппроксимации данных от степени полиномиальной регрессии. Результаты численного эксперимента позволяют сделать вывод о правомерности оценки неустранимых погрешностей данных с помощью метода конечных разностей. Применение аппарата конечных разностей для анализа выборки показывает наличие неустранимых ошибок в исследуемой выборке от 0,59% до 28,92 %. Следовательно, получение композитных характеристик объектов на основании таких данных должно обязательно учитывать наличие неустранимой ошибки. В частности, число параметров и число наблюдений, характеризующих систему, должно быть достаточно большим, чтобы компенсировать случайные ошибки. The construction of the composite index of a system can be considered as a problem of separating signal from noise. The signal in this case is the weight coefficients of the linear convolution of indicators. The weights to be determined should reflect the structure of the system being evaluated. However, principal component analysis and factor analysis determine the structure of principal components and principal factors differently for different observations. The reason for this may be the presence of inevitable errors in the used data. A solution of the problem requires a detailed understanding of input data errors’ influence on the calculated model’s parameters. The article discusses the use of the finite difference method for evaluating statistical data quality in the problem of calculating the integral characteristic of a system for a number of observations. For this technique to be applicable, the data must be approximated with polynomials of lower degrees than the number of observations minus one. The assumption is tested empirically on a specific data set. 37 variables characterizing the quality of life of the population of Russia for 2010-2017 are considered. The dependencies of the quality of data approximation on the degree of polynomial regression are analyzed. The results of the numerical experiment make it possible to draw a conclusion about the legitimacy of evaluating data errors using the finite difference method. The use of the finite difference apparatus for analyzing the data shows the presence of fatal errors from 0.59% to 28.92%. Therefore, obtaining the composite characteristics of objects on the basis of such data must necessarily take into account the presence of a fatal error. In particular, the number of parameters characterizing the system should be large enough to compensate for random errors with averaging.