- Award ID(s):
- 1946311
- NSF-PAR ID:
- 10420423
- Date Published:
- Journal Name:
- Research in Number Theory
- Volume:
- 9
- Issue:
- 1
- ISSN:
- 2522-0160
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
We study the $\ell_p$ regression problem, which requires finding $\mathbf{x}\in\mathbb R^{d}$ that minimizes $\|\mathbf{A}\mathbf{x}-\mathbf{b}\|_p$ for a matrix $\mathbf{A}\in\mathbb R^{n \times d}$ and response vector $\mathbf{b}\in\mathbb R^{n}$. There has been recent interest in developing subsampling methods for this problem that can outperform standard techniques when $n$ is very large. However, all known subsampling approaches have run time that depends exponentially on $p$, typically, $d^{\mathcal{O}(p)}$, which can be prohibitively expensive. We improve on this work by showing that for a large class of common \emph{structured matrices}, such as combinations of low-rank matrices, sparse matrices, and Vandermonde matrices, there are subsampling based methods for $\ell_p$ regression that depend polynomially on $p$. For example, we give an algorithm for $\ell_p$ regression on Vandermonde matrices that runs in time $\mathcal{O}(n\log^3 n+(dp^2)^{0.5+\omega}\cdot\text{polylog}\,n)$, where $\omega$ is the exponent of matrix multiplication. The polynomial dependence on $p$ crucially allows our algorithms to extend naturally to efficient algorithms for $\ell_\infty$ regression, via approximation of $\ell_\infty$ by $\ell_{\mathcal{O}(\log n)}$. Of practical interest, we also develop a new subsampling algorithm for $\ell_p$ regression for arbitrary matrices, which is simpler than previous approaches for $p \ge 4$.more » « less
-
Let $p$ be an odd prime, $q=p^e$, $e \geq 1$, and $\mathbb{F} = \mathbb{F}_q$ denote the finite field of $q$ elements. Let $f: \mathbb{F}^2\to \mathbb{F}$ and $g: \mathbb{F}^3\to \mathbb{F}$ be functions, and let $P$ and $L$ be two copies of the 3-dimensional vector space $\mathbb{F}^3$. Consider a bipartite graph $\Gamma_\mathbb{F} (f, g)$ with vertex partitions $P$ and $L$ and with edges defined as follows: for every $(p)=(p_1,p_2,p_3)\in P$ and every $[l]= [l_1,l_2,l_3]\in L$, $\{(p), [l]\} = (p)[l]$ is an edge in $\Gamma_\mathbb{F} (f, g)$ if $$p_2+l_2 =f(p_1,l_1) \;\;\;\text{and}\;\;\; p_3 + l_3 = g(p_1,p_2,l_1).$$The following question appeared in Nassau: Given $\Gamma_\mathbb{F} (f, g)$, is it always possible to find a function $h:\mathbb{F}^2\to \mathbb{F}$ such that the graph $\Gamma_\mathbb{F} (f, h)$ with the same vertex set as $\Gamma_\mathbb{F} (f, g)$ and with edges $(p)[l]$ defined in a similar way by the system $$p_2+l_2 =f(p_1,l_1) \;\;\;\text{and}\;\;\; p_3 + l_3 = h(p_1,l_1),$$ is isomorphic to $\Gamma_\mathbb{F} (f, g)$ for infinitely many $q$? In this paper we show that the answer to the question is negative and the graphs $\Gamma_{\mathbb{F}_p}(p_1\ell_1, p_1\ell_1p_2(p_1 + p_2 + p_1p_2))$ provide such an example for $p \equiv 1 \pmod{3}$. Our argument is based on proving that the automorphism group of these graphs has order $p$, which is the smallest possible order of the automorphism group of graphs of the form $\Gamma_{\mathbb{F}}(f, g)$.more » « less
-
Abstract For every integer k there exists a bound $$B=B(k)$$ B = B ( k ) such that if the characteristic polynomial of $$g\in \textrm{SL}_n(q)$$ g ∈ SL n ( q ) is the product of $$\le k$$ ≤ k pairwise distinct monic irreducible polynomials over $$\mathbb {F}_q$$ F q , then every element x of $$\textrm{SL}_n(q)$$ SL n ( q ) of support at least B is the product of two conjugates of g . We prove this and analogous results for the other classical groups over finite fields; in the orthogonal and symplectic cases, the result is slightly weaker. With finitely many exceptions ( p , q ), in the special case that $$n=p$$ n = p is prime, if g has order $$\frac{q^p-1}{q-1}$$ q p - 1 q - 1 , then every non-scalar element $$x \in \textrm{SL}_p(q)$$ x ∈ SL p ( q ) is the product of two conjugates of g . The proofs use the Frobenius formula together with upper bounds for values of unipotent and quadratic unipotent characters in finite classical groups.more » « less
-
Abstract We continue the program of proving circuit lower bounds via circuit satisfiability algorithms. So far, this program has yielded several concrete results, proving that functions in
and other complexity classes do not have small circuits (in the worst case and/or on average) from various circuit classes$\mathsf {Quasi}\text {-}\mathsf {NP} = \mathsf {NTIME}[n^{(\log n)^{O(1)}}]$ , by showing that$\mathcal { C}$ admits non-trivial satisfiability and/or$\mathcal { C}$ # SAT algorithms which beat exhaustive search by a minor amount. In this paper, we present a new strong lower bound consequence of having a non-trivial# SAT algorithm for a circuit class . Say that a symmetric Boolean function${\mathcal C}$ f (x 1,…,x n ) issparse if it outputs 1 onO (1) values of . We show that for every sparse${\sum }_{i} x_{i}$ f , and for all “typical” , faster$\mathcal { C}$ # SAT algorithms for circuits imply lower bounds against the circuit class$\mathcal { C}$ , which may be$f \circ \mathcal { C}$ stronger than itself. In particular:$\mathcal { C}$ # SAT algorithms forn k -size -circuits running in 2$\mathcal { C}$ n /n k time (for allk ) implyN E X P does not have -circuits of polynomial size.$(f \circ \mathcal { C})$ # SAT algorithms for -size$2^{n^{{\varepsilon }}}$ -circuits running in$\mathcal { C}$ time (for some$2^{n-n^{{\varepsilon }}}$ ε > 0) implyQ u a s i -N P does not have -circuits of polynomial size.$(f \circ \mathcal { C})$ Applying
# SAT algorithms from the literature, one immediate corollary of our results is thatQ u a s i -N P does not haveE M A J ∘A C C 0∘T H R circuits of polynomial size, whereE M A J is the “exact majority” function, improving previous lower bounds againstA C C 0[Williams JACM’14] andA C C 0∘T H R [Williams STOC’14], [Murray-Williams STOC’18]. This is the first nontrivial lower bound against such a circuit class. -
Abstract Let us fix a prime
p and a homogeneous system ofm linear equations for$$a_{j,1}x_1+\dots +a_{j,k}x_k=0$$ with coefficients$$j=1,\dots ,m$$ . Suppose that$$a_{j,i}\in \mathbb {F}_p$$ , that$$k\ge 3m$$ for$$a_{j,1}+\dots +a_{j,k}=0$$ and that every$$j=1,\dots ,m$$ minor of the$$m\times m$$ matrix$$m\times k$$ is non-singular. Then we prove that for any (large)$$(a_{j,i})_{j,i}$$ n , any subset of size$$A\subseteq \mathbb {F}_p^n$$ contains a solution$$|A|> C\cdot \Gamma ^n$$ to the given system of equations such that the vectors$$(x_1,\dots ,x_k)\in A^k$$ are all distinct. Here,$$x_1,\dots ,x_k\in A$$ C and are constants only depending on$$\Gamma $$ p ,m andk such that . The crucial point here is the condition for the vectors$$\Gamma in the solution$$x_1,\dots ,x_k$$ to be distinct. If we relax this condition and only demand that$$(x_1,\dots ,x_k)\in A^k$$ are not all equal, then the statement would follow easily from Tao’s slice rank polynomial method. However, handling the distinctness condition is much harder, and requires a new approach. While all previous combinatorial applications of the slice rank polynomial method have relied on the slice rank of diagonal tensors, we use a slice rank argument for a non-diagonal tensor in combination with combinatorial and probabilistic arguments.$$x_1,\dots ,x_k$$