More Like this

1. Computing Higher Polynomial Discriminants

   https://doi.org/10.1145/3452143.3465543  

   Kaltofen, Erich L. (July 2021, ISSAC '21: Proceedings of the 2021 on International Symposium on Symbolic and Algebraic Computation)

   null (Ed.)

   In https://arxiv.org/abs/1609.00840, D. Wang and J. Yang in 2016 have posed the problem how to compute the ``third'' discriminant of a polynomial f(X) = (X-A_1) ... (X-A_n), delta_3(f) = product_{1 <= i < j < k < l <= n} ( (A_i + A_j - A_k - A_l) (A_i - A_j + A_k - A_l) (A_i - A_j - A_k + A_l) ), from the coefficients of f; note that delta_3 is a symmetric polynomial in the A_i. For complex roots, delta_3(f) = 0 if the mid-point (average) of 2 roots is equal the mid-point of another 2 roots. Iterated resultant computations yield the square of the third discriminant. We apply a symbolic homotopy by Kaltofen and Trager [JSC, vol. 9, nr. 3, pp. 301--320 (1990)] to compute its squareroot. Our algorithm use polynomially many coefficient field operations. 

   more » « less
2. Counting Roots of Polynomials over Z/p2Z

   Hammonds, Trajan; Johnson, Jeremy; Patini, Angela; Walker, Robert M. (January 2018, Houston journal of mathematics)

   Until recently, the only known method of finding the roots of polynomials over prime power rings, other than fields, was brute force. One reason for this is the lack of a division algorithm, obstructing the use of greatest common divisors. Fix a prime integer p and f in (Z/pnZ)[x] any nonzero polynomial of degree d whose coefficients are not all divisible by p. For the case n=2, we prove a new efficient algorithm to count the roots of f in Z/p2Z within time (d+size(f)+log p),2+o(1), based on a formula conjectured by Cheng, Gao, Rojas, and Wan. 

   more » « less
3. On Multilinear Forms: Bias, Correlation, and Tensor Rank

   Bhrushundi, Abhishek; Harsha, Prahladh; Hatami, Pooya; Kopparty, Swastik; Kumar, Mrinal (August 2020, Leibniz international proceedings in informatics)

   Byrka, Jaroslaw; Meka, Raghu (Ed.)

   In this work, we prove new relations between the bias of multilinear forms, the correlation between multilinear forms and lower degree polynomials, and the rank of tensors over F₂. We show the following results for multilinear forms and tensors. Correlation bounds. We show that a random d-linear form has exponentially low correlation with low-degree polynomials. More precisely, for d = 2^{o(k)}, we show that a random d-linear form f(X₁,X₂, … , X_d) : (F₂^{k}) ^d → F₂ has correlation 2^{-k(1-o(1))} with any polynomial of degree at most d/2 with high probability. This result is proved by giving near-optimal bounds on the bias of a random d-linear form, which is in turn proved by giving near-optimal bounds on the probability that a sum of t random d-dimensional rank-1 tensors is identically zero. Tensor rank vs Bias. We show that if a 3-dimensional tensor has small rank then its bias, when viewed as a 3-linear form, is large. More precisely, given any 3-dimensional tensor T: [k]³ → F₂ of rank at most t, the bias of the 3-linear form f_T(X₁, X₂, X₃) : = ∑_{(i₁, i₂, i₃) ∈ [k]³} T(i₁, i₂, i₃)⋅ X_{1,i₁}⋅ X_{2,i₂}⋅ X_{3,i₃} is at least (3/4)^t. This bias vs tensor-rank connection suggests a natural approach to proving nontrivial tensor-rank lower bounds. In particular, we use this approach to give a new proof that the finite field multiplication tensor has tensor rank at least 3.52 k, which is the best known rank lower bound for any explicit tensor in three dimensions over F₂. Moreover, this relation between bias and tensor rank holds for d-dimensional tensors for any fixed d. 

   more » « less
4. Learning in the Presence of Low-dimensional Structure: A Spiked Random Matrix Perspective

   Ba, Jimmy; Erdogdu, Murat A; Suzuki, Taiji; Wang, Zhichao; Wu, Denny (December 2023, Advances in Neural Information Processing Systems 36 (NeurIPS 2023))

   Oh, A; Naumann, T; Globerson, A; Saenko, K; Hardt, M; Levine, S (Ed.)

   We consider the problem of learning a single-index target function f∗ : Rd → R under the spiked covariance data: f∗(x) = σ∗   √ 1 1+θ ⟨x,μ⟩   , x ∼ N(0, Id + θμμ⊤), θ ≍ dβ for β ∈ [0, 1), where the link function σ∗ : R → R is a degree-p polynomial with information exponent k (defined as the lowest degree in the Hermite expansion of σ∗), and it depends on the projection of input x onto the spike (signal) direction μ ∈ Rd. In the proportional asymptotic limit where the number of training examples n and the dimensionality d jointly diverge: n, d → ∞, n/d → ψ ∈ (0,∞), we ask the following question: how large should the spike magnitude θ be, in order for (i) kernel methods, (ii) neural networks optimized by gradient descent, to learn f∗? We show that for kernel ridge regression, β ≥ 1 − 1 p is both sufficient and necessary. Whereas for two-layer neural networks trained with gradient descent, β > 1 − 1 k suffices. Our results demonstrate that both kernel methods and neural networks benefit from low-dimensional structures in the data. Further, since k ≤ p by definition, neural networks can adapt to such structures more effectively. 

   more » « less
5. Semi-Algebraic Off-Line Range Searching and Biclique Partitions in the Plane

   https://doi.org/10.4230/LIPICS.SOCG.2024.4  

   Agarwal, Pankaj K; Ezra, Esther; Sharir, Micha (January 2024, Schloss Dagstuhl – Leibniz-Zentrum für Informatik)

   Mulzer, Wolfgang; Phillips, Jeff M (Ed.)

   Let P be a set of m points in ℝ², let Σ be a set of n semi-algebraic sets of constant complexity in ℝ², let (S,+) be a semigroup, and let w: P → S be a weight function on the points of P. We describe a randomized algorithm for computing w(P∩σ) for every σ ∈ Σ in overall expected time O^*(m^{2s/(5s-4)}n^{(5s-6)/(5s-4)} + m^{2/3}n^{2/3} + m + n), where s > 0 is a constant that bounds the maximum complexity of the regions of Σ, and where the O^*(⋅) notation hides subpolynomial factors. For s ≥ 3, surprisingly, this bound is smaller than the best-known bound for answering m such queries in an on-line manner. The latter takes O^*(m^{s/(2s-1)}n^{(2s-2)/(2s-1)} + m + n) time. Let Φ: Σ × P → {0,1} be the Boolean predicate (of constant complexity) such that Φ(σ,p) = 1 if p ∈ σ and 0 otherwise, and let Σ_Φ P = {(σ,p) ∈ Σ× P ∣ Φ(σ,p) = 1}. Our algorithm actually computes a partition ℬ_Φ of Σ_Φ P into bipartite cliques (bicliques) of size (i.e., sum of the sizes of the vertex sets of its bicliques) O^*(m^{2s/(5s-4)}n^{(5s-6)/(5s-4)} + m^{2/3}n^{2/3} + m + n). It is straightforward to compute w(P∩σ) for all σ ∈ Σ from ℬ_Φ. Similarly, if η: Σ → S is a weight function on the regions of Σ, ∑_{σ ∈ Σ: p ∈ σ} η(σ), for every point p ∈ P, can be computed from ℬ_Φ in a straightforward manner. We also mention a few other applications of computing ℬ_Φ. 

   more » « less