arXiv subjects

Ryan Williams

Publications and source records attributed to Ryan Williams.

At least 37 records · Page 2Linked to original sources

arXiv2020

Fast Low-Space Algorithms for Subset Sum

We consider the canonical Subset Sum problem: given a list of positive integers $a_1,\ldots,a_n$ and a target integer $t$ with $t > a_i$ for all $i$, determine if there is an $S \subseteq [n]$ such that $\sum_{i \in S} a_i = t$. The well-known pseudopolynomial-time dynamic programming algorithm [Bellman, 1957] solves Subset Sum in $O(nt)$ time, while requiring $Ω(t)$ space. In this paper we present algorithms for Subset Sum with $\tilde O(nt)$ running time and much lower space requirements than Bellman's algorithm, as well as that of prior work. We show that Subset Sum can be solved in $\tilde O(nt)$ time and $O(\log(nt))$ space with access to $O(\log n \log \log n+\log t)$ random bits. This significantly improves upon the $\tilde O(n t^{1+\varepsilon})$-time, $\tilde O(n\log t)$-space algorithm of Bringmann (SODA 2017). We also give an $\tilde O(n^{1+\varepsilon}t)$-time, $O(\log(nt))$-space randomized algorithm, improving upon previous $(nt)^{O(1)}$-time $O(\log(nt))$-space algorithms by Elberfeld, Jakoby, and Tantau (FOCS 2010), and Kane (2010). In addition, we also give a $\mathrm{poly} \log(nt)$-space, $\tilde O(n^2 t)$-time deterministic algorithm. We also study time-space trade-offs for Subset Sum. For parameter $1\le k\le \min\{n,t\}$, we present a randomized algorithm running in $\tilde O((n+t)\cdot k)$ time and $O((t/k) \mathrm{polylog} (nt))$ space. As an application of our results, we give an $\tilde{O}(\min\{n^2/\varepsilon, n/\varepsilon^2\})$-time and $\mathrm{polylog}(nt)$-space algorithm for "weak" $\varepsilon$-approximations of Subset Sum.

cs.DS↗

arXiv2020

Tight Hardness for Shortest Cycles and Paths in Sparse Graphs

Fine-grained reductions have established equivalences between many core problems with $\tilde{O}(n^3)$-time algorithms on $n$-node weighted graphs, such as Shortest Cycle, All-Pairs Shortest Paths (APSP), Radius, Replacement Paths, Second Shortest Paths, and so on. These problems also have $\tilde{O}(mn)$-time algorithms on $m$-edge $n$-node weighted graphs, and such algorithms have wider applicability. Are these $mn$ bounds optimal when $m \ll n^2$? Starting from the hypothesis that the minimum weight $(2\ell+1)$-Clique problem in edge weighted graphs requires $n^{2\ell+1-o(1)}$ time, we prove that for all sparsities of the form $m = Θ(n^{1+1/\ell})$, there is no $O(n^2 + mn^{1-ε})$ time algorithm for $ε>0$ for \emph{any} of the below problems: Minimum Weight $(2\ell+1)$-Cycle in a directed weighted graph, Shortest Cycle in a directed weighted graph, APSP in a directed or undirected weighted graph, Radius (or Eccentricities) in a directed or undirected weighted graph, Wiener index of a directed or undirected weighted graph, Replacement Paths in a directed weighted graph, Second Shortest Path in a directed weighted graph, Betweenness Centrality of a given node in a directed weighted graph. That is, we prove hardness for a variety of sparse graph problems from the hardness of a dense graph problem. Our results also lead to new conditional lower bounds from several related hypothesis for unweighted sparse graph problems including $k$-cycle, shortest cycle, Radius, Wiener index and APSP.

cs.DS↗

arXiv2020

Lower Bounds Against Sparse Symmetric Functions of ACC Circuits: Expanding the Reach of $\#$SAT Algorithms

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 $\text{Quasi-NP} = \text{NTIME}[n^{(\log n)^{O(1)}}]$ and $\text{NEXP}$ do not have small circuits from various circuit classes ${\cal C}$, by showing that ${\cal C}$ admits non-trivial satisfiability and/or $\#$SAT algorithms which beat exhaustive search by a minor amount. In this paper, we present a new strong lower bound consequence of non-trivial $\#$SAT algorithm for a circuit class ${\mathcal C}$. Say a symmetric Boolean function $f(x_1,\ldots,x_n)$ is sparse if it outputs $1$ on $O(1)$ values of $\sum_i x_i$. We show that for every sparse $f$, and for all "typical" ${\cal C}$, faster $\#$SAT algorithms for ${\cal C}$ circuits actually imply lower bounds against the circuit class $f \circ {\cal C}$, which may be stronger than ${\cal C}$ itself. In particular: $\#$SAT algorithms for $n^k$-size ${\cal C}$-circuits running in $2^n/n^k$ time (for all $k$) imply $\text{NEXP}$ does not have $f \circ {\cal C}$-circuits of polynomial size. $\#$SAT algorithms for $2^{n^ε}$-size ${\cal C}$-circuits running in $2^{n-n^ε}$ time (for some $ε> 0$) imply $\text{Quasi-NP}$ does not have $f \circ {\cal C}$-circuits of polynomial size. Applying $\#$SAT algorithms from the literature, one immediate corollary of our results is that $\text{Quasi-NP}$ does not have $\text{EMAJ} \circ \text{ACC}^0 \circ \text{THR}$ circuits of polynomial size, where $\text{EMAJ}$ is the "exact majority" function, improving previous lower bounds against $\text{ACC}^0$ [Williams JACM'14] and $\text{ACC}^0 \circ \text{THR}$ [Williams STOC'14], [Murray-Williams STOC'18]. This is the first nontrivial lower bound against such a circuit class.

cs.CC↗

arXiv2018

An Equivalence Class for Orthogonal Vectors

The Orthogonal Vectors problem ($\textsf{OV}$) asks: given $n$ vectors in $\{0,1\}^{O(\log n)}$, are two of them orthogonal? $\textsf{OV}$ is easily solved in $O(n^2 \log n)$ time, and it is a central problem in fine-grained complexity: dozens of conditional lower bounds are based on the popular hypothesis that $\textsf{OV}$ cannot be solved in (say) $n^{1.99}$ time. However, unlike the APSP problem, few other problems are known to be non-trivially equivalent to $\textsf{OV}$. We show $\textsf{OV}$ is truly-subquadratic equivalent to several fundamental problems, all of which (a priori) look harder than $\textsf{OV}$. A partial list is given below: ($\textsf{Min-IP}/\textsf{Max-IP}$) Find a red-blue pair of vectors with minimum (respectively, maximum) inner product, among $n$ vectors in $\{0,1\}^{O(\log n)}$. ($\textsf{Exact-IP}$) Find a red-blue pair of vectors with inner product equal to a given target integer, among $n$ vectors in $\{0,1\}^{O(\log n)}$. ($\textsf{Apx-Min-IP}/\textsf{Apx-Max-IP}$) Find a red-blue pair of vectors that is a 100-approximation to the minimum (resp. maximum) inner product, among $n$ vectors in $\{0,1\}^{O(\log n)}$. (Approx. $\textsf{Bichrom.-$\ell_p$-Closest-Pair}$) Compute a $(1 + Ω(1))$-approximation to the $\ell_p$-closest red-blue pair (for a constant $p \in [1,2]$), among $n$ points in $\mathbb{R}^d$, $d \le n^{o(1)}$. (Approx. $\textsf{$\ell_p$-Furthest-Pair}$) Compute a $(1 + Ω(1))$-approximation to the $\ell_p$-furthest pair (for a constant $p \in [1,2]$), among $n$ points in $\mathbb{R}^d$, $d \le n^{o(1)}$. We also show that there is a $\text{poly}(n)$ space, $n^{1-ε}$ query time data structure for Partial Match with vectors from $\{0,1\}^{O(\log n)}$ if and only if such a data structure exists for $1+Ω(1)$ Approximate Nearest Neighbor Search in Euclidean space.

cs.DS↗

arXiv2017

Distributed PCP Theorems for Hardness of Approximation in P

We present a new distributed model of probabilistically checkable proofs (PCP). A satisfying assignment $x \in \{0,1\}^n$ to a CNF formula $φ$ is shared between two parties, where Alice knows $x_1, \dots, x_{n/2}$, Bob knows $x_{n/2+1},\dots,x_n$, and both parties know $φ$. The goal is to have Alice and Bob jointly write a PCP that $x$ satisfies $φ$, while exchanging little or no information. Unfortunately, this model as-is does not allow for nontrivial query complexity. Instead, we focus on a non-deterministic variant, where the players are helped by Merlin, a third party who knows all of $x$. Using our framework, we obtain, for the first time, PCP-like reductions from the Strong Exponential Time Hypothesis (SETH) to approximation problems in P. In particular, under SETH we show that there are no truly-subquadratic approximation algorithms for Bichromatic Maximum Inner Product over {0,1}-vectors, Bichromatic LCS Closest Pair over permutations, Approximate Regular Expression Matching, and Diameter in Product Metric. All our inapproximability factors are nearly-tight. In particular, for the first two problems we obtain nearly-polynomial factors of $2^{(\log n)^{1-o(1)}}$; only $(1+o(1))$-factor lower bounds (under SETH) were known before.

cs.CC↗

arXiv2017

On the Difference Between Closest, Furthest, and Orthogonal Pairs: Nearly-Linear vs Barely-Subquadratic Complexity in Computational Geometry

Point location problems for $n$ points in $d$-dimensional Euclidean space (and $\ell_p$ spaces more generally) have typically had two kinds of running-time solutions: * (Nearly-Linear) less than $d^{poly(d)} \cdot n \log^{O(d)} n$ time, or * (Barely-Subquadratic) $f(d) \cdot n^{2-1/Θ(d)}$ time, for various $f$. For small $d$ and large $n$, "nearly-linear" running times are generally feasible, while "barely-subquadratic" times are generally infeasible. For example, in the Euclidean metric, finding a Closest Pair among $n$ points in ${\mathbb R}^d$ is nearly-linear, solvable in $2^{O(d)} \cdot n \log^{O(1)} n$ time, while known algorithms for Furthest Pair (the diameter of the point set) are only barely-subquadratic, requiring $Ω(n^{2-1/Θ(d)})$ time. Why do these proximity problems have such different time complexities? Is there a barrier to obtaining nearly-linear algorithms for problems which are currently only barely-subquadratic? We give a novel exact and deterministic self-reduction for the Orthogonal Vectors problem on $n$ vectors in $\{0,1\}^d$ to $n$ vectors in ${\mathbb Z}^{ω(\log d)}$ that runs in $2^{o(d)}$ time. As a consequence, barely-subquadratic problems such as Euclidean diameter, Euclidean bichromatic closest pair, ray shooting, and incidence detection do not have $O(n^{2-ε})$ time algorithms (in Turing models of computation) for dimensionality $d = ω(\log \log n)^2$, unless the popular Orthogonal Vectors Conjecture and the Strong Exponential Time Hypothesis are false. That is, while poly-log-log-dimensional Closest Pair is in $n^{1+o(1)}$ time, the analogous case of Furthest Pair can encode larger-dimensional problems conjectured to require $n^{2-o(1)}$ time. We also show that the All-Nearest Neighbors problem in $ω(\log n)$ dimensions requires $n^{2-o(1)}$ time to solve, assuming either of the above conjectures.

cs.CG↗

arXiv2017

The Orthogonal Vectors Conjecture for Branching Programs and Formulas

In the Orthogonal Vectors (OV) problem, we wish to determine if there is an orthogonal pair of vectors among $n$ Boolean vectors in $d$ dimensions. The OV Conjecture (OVC) posits that OV requires $n^{2-o(1)}$ time to solve, for all $d=ω(\log n)$. Assuming the OVC, optimal time lower bounds have been proved for many prominent problems in $P$. We prove that OVC is true in several computational models of interest: * For all sufficiently large $n$ and $d$, OV for $n$ vectors in $\{0,1\}^d$ has branching program complexity $\tildeΘ(n\cdot \min(n,2^d))$. In particular, the lower bounds match the upper bounds up to polylog factors. * OV has Boolean formula complexity $\tildeΘ(n\cdot \min(n,2^d))$, over all complete bases of $O(1)$ fan-in. * OV requires $\tildeΘ(n\cdot \min(n,2^d))$ wires, in formulas comprised of gates computing arbitrary symmetric functions of unbounded fan-in. Our lower bounds basically match the best known (quadratic) lower bounds for any explicit function in those models. Analogous lower bounds hold for many related problems shown to be hard under OVC, such as Batch Partial Match, Batch Subset Queries, and Batch Hamming Nearest Neighbors, all of which have very succinct reductions to OV. The proofs use a certain kind of input restriction that is different from typical random restrictions where variables are assigned independently. We give a sense in which independent random restrictions cannot be used to show hardness, in that OVC is false in the "average case" even for $AC^0$ formulas: * For every fixed $p \in (0,1)$ there is an $ε_p > 0$ such that for every $n$ and $d$, OV instances where input bits are independently set to $1$ with probability $p$ (and $0$ otherwise) can be solved with $AC^0$ formulas of size $O(n^{2-ε_p})$, on all but a $o_n(1)$ fraction of instances.

cs.CC↗

arXiv2017

Probabilistic Rank and Matrix Rigidity

We consider a notion of probabilistic rank and probabilistic sign-rank of a matrix, which measures the extent to which a matrix can be probabilistically represented by low-rank matrices. We demonstrate several connections with matrix rigidity, communication complexity, and circuit lower bounds, including: The Walsh-Hadamard Transform is Not Very Rigid. We give surprising upper bounds on the rigidity of a family of matrices whose rigidity has been extensively studied, and was conjectured to be highly rigid. For the $2^n \times 2^n$ Walsh-Hadamard transform $H_n$ (a.k.a. Sylvester matrices, or the communication matrix of Inner Product mod 2), we show how to modify only $2^{εn}$ entries in each row and make the rank drop below $2^{n(1-Ω(ε^2/\log(1/ε)))}$, for all $ε> 0$, over any field. That is, it is not possible to prove arithmetic circuit lower bounds on Hadamard matrices, via L. Valiant's matrix rigidity approach. We also show non-trivial rigidity upper bounds for $H_n$ with smaller target rank. Matrix Rigidity and Threshold Circuit Lower Bounds. We give new consequences of rigid matrices for Boolean circuit complexity. We show that explicit $n \times n$ Boolean matrices which maintain rank at least $2^{(\log n)^{1-δ}}$ after $n^2/2^{(\log n)^{δ/2}}$ modified entries would yield a function lacking sub-quadratic-size $AC^0$ circuits with two layers of arbitrary linear threshold gates. We also prove that explicit 0/1 matrices over $\mathbb{R}$ which are modestly more rigid than the best known rigidity lower bounds for sign-rank would imply strong lower bounds for the infamously difficult class $THR\circ THR$.

cs.CC↗

arXiv2016

Polynomial Representations of Threshold Functions and Algorithmic Applications

We design new polynomials for representing threshold functions in three different regimes: probabilistic polynomials of low degree, which need far less randomness than previous constructions, polynomial threshold functions (PTFs) with "nice" threshold behavior and degree almost as low as the probabilistic polynomials, and a new notion of probabilistic PTFs where we combine the above techniques to achieve even lower degree with similar "nice" threshold behavior. Utilizing these polynomial constructions, we design faster algorithms for a variety of problems: $\bullet$ Offline Hamming Nearest (and Furthest) Neighbors: Given $n$ red and $n$ blue points in $d$-dimensional Hamming space for $d=c\log n$, we can find an (exact) nearest (or furthest) blue neighbor for every red point in randomized time $n^{2-1/O(\sqrt{c}\log^{2/3}c)}$ or deterministic time $n^{2-1/O(c\log^2c)}$. These also lead to faster MAX-SAT algorithms for sparse CNFs. $\bullet$ Offline Approximate Nearest (and Furthest) Neighbors: Given $n$ red and $n$ blue points in $d$-dimensional $\ell_1$ or Euclidean space, we can find a $(1+ε)$-approximate nearest (or furthest) blue neighbor for each red point in randomized time near $dn+n^{2-Ω(ε^{1/3}/\log(1/ε))}$. $\bullet$ SAT Algorithms and Lower Bounds for Circuits With Linear Threshold Functions: We give a satisfiability algorithm for $AC^0[m]\circ LTF\circ LTF$ circuits with a subquadratic number of linear threshold gates on the bottom layer, and a subexponential number of gates on the other layers, that runs in deterministic $2^{n-n^ε}$ time. This also implies new circuit lower bounds for threshold circuits. We also give a randomized $2^{n-n^ε}$-time SAT algorithm for subexponential-size $MAJ\circ AC^0\circ LTF\circ AC^0\circ LTF$ circuits, where the top $MAJ$ gate and middle $LTF$ gates have $O(n^{6/5-δ})$ fan-in.

cs.DS↗

arXiv2016

Faster Online Matrix-Vector Multiplication

We consider the Online Boolean Matrix-Vector Multiplication (OMV) problem studied by Henzinger et al. [STOC'15]: given an $n \times n$ Boolean matrix $M$, we receive $n$ Boolean vectors $v_1,\ldots,v_n$ one at a time, and are required to output $M v_i$ (over the Boolean semiring) before seeing the vector $v_{i+1}$, for all $i$. Previous known algorithms for this problem are combinatorial, running in $O(n^3/\log^2 n)$ time. Henzinger et al. conjecture there is no $O(n^{3-\varepsilon})$ time algorithm for OMV, for all $\varepsilon > 0$; their OMV conjecture is shown to imply strong hardness results for many basic dynamic problems. We give a substantially faster method for computing OMV, running in $n^3/2^{Ω(\sqrt{\log n})}$ randomized time. In fact, after seeing $2^{ω(\sqrt{\log n})}$ vectors, we already achieve $n^2/2^{Ω(\sqrt{\log n})}$ amortized time for matrix-vector multiplication. Our approach gives a way to reduce matrix-vector multiplication to solving a version of the Orthogonal Vectors problem, which in turn reduces to "small" algebraic matrix-matrix multiplication. Applications include faster independent set detection, partial match retrieval, and 2-CNF evaluation. We also show how a modification of our method gives a cell probe data structure for OMV with worst case $O(n^{7/4}/\sqrt{w})$ time per query vector, where $w$ is the word size. This result rules out an unconditional proof of the OMV conjecture using purely information-theoretic arguments.

cs.DS↗

arXiv2016

Strong ETH Breaks With Merlin and Arthur: Short Non-Interactive Proofs of Batch Evaluation

We present an efficient proof system for Multipoint Arithmetic Circuit Evaluation: for every arithmetic circuit $C(x_1,\ldots,x_n)$ of size $s$ and degree $d$ over a field ${\mathbb F}$, and any inputs $a_1,\ldots,a_K \in {\mathbb F}^n$, $\bullet$ the Prover sends the Verifier the values $C(a_1), \ldots, C(a_K) \in {\mathbb F}$ and a proof of $\tilde{O}(K \cdot d)$ length, and $\bullet$ the Verifier tosses $\textrm{poly}(\log(dK|{\mathbb F}|/\varepsilon))$ coins and can check the proof in about $\tilde{O}(K \cdot(n + d) + s)$ time, with probability of error less than $\varepsilon$. For small degree $d$, this "Merlin-Arthur" proof system (a.k.a. MA-proof system) runs in nearly-linear time, and has many applications. For example, we obtain MA-proof systems that run in $c^{n}$ time (for various $c < 2$) for the Permanent, $\#$Circuit-SAT for all sublinear-depth circuits, counting Hamiltonian cycles, and infeasibility of $0$-$1$ linear programs. In general, the value of any polynomial in Valiant's class ${\sf VP}$ can be certified faster than "exhaustive summation" over all possible assignments. These results strongly refute a Merlin-Arthur Strong ETH and Arthur-Merlin Strong ETH posed by Russell Impagliazzo and others. We also give a three-round (AMA) proof system for quantified Boolean formulas running in $2^{2n/3+o(n)}$ time, nearly-linear time MA-proof systems for counting orthogonal vectors in a collection and finding Closest Pairs in the Hamming metric, and a MA-proof system running in $n^{k/2+O(1)}$-time for counting $k$-cliques in graphs. We point to some potential future directions for refuting the Nondeterministic Strong ETH.

cs.CC↗

arXiv2015

Losing Weight by Gaining Edges

We present a new way to encode weighted sums into unweighted pairwise constraints, obtaining the following results. - Define the k-SUM problem to be: given n integers in [-n^2k, n^2k] are there k which sum to zero? (It is well known that the same problem over arbitrary integers is equivalent to the above definition, by linear-time randomized reductions.) We prove that this definition of k-SUM remains W[1]-hard, and is in fact W[1]-complete: k-SUM can be reduced to f(k) * n^o(1) instances of k-Clique. - The maximum node-weighted k-Clique and node-weighted k-dominating set problems can be reduced to n^o(1) instances of the unweighted k-Clique and k-dominating set problems, respectively. This implies a strong equivalence between the time complexities of the node weighted problems and the unweighted problems: any polynomial improvement on one would imply an improvement for the other. - A triangle of weight 0 in a node weighted graph with m edges can be deterministically found in m^1.41 time.

cs.CC↗

arXiv2015

Super-Linear Gate and Super-Quadratic Wire Lower Bounds for Depth-Two and Depth-Three Threshold Circuits

In order to formally understand the power of neural computing, we first need to crack the frontier of threshold circuits with two and three layers, a regime that has been surprisingly intractable to analyze. We prove the first super-linear gate lower bounds and the first super-quadratic wire lower bounds for depth-two linear threshold circuits with arbitrary weights, and depth-three majority circuits computing an explicit function. $\bullet$ We prove that for all $ε\gg \sqrt{\log(n)/n}$, the linear-time computable Andreev's function cannot be computed on a $(1/2+ε)$-fraction of $n$-bit inputs by depth-two linear threshold circuits of $o(ε^3 n^{3/2}/\log^3 n)$ gates, nor can it be computed with $o(ε^{3} n^{5/2}/\log^{7/2} n)$ wires. This establishes an average-case ``size hierarchy'' for threshold circuits, as Andreev's function is computable by uniform depth-two circuits of $o(n^3)$ linear threshold gates, and by uniform depth-three circuits of $O(n)$ majority gates. $\bullet$ We present a new function in $P$ based on small-biased sets, which we prove cannot be computed by a majority vote of depth-two linear threshold circuits with $o(n^{3/2}/\log^3 n)$ gates, nor with $o(n^{5/2}/\log^{7/2}n)$ wires. $\bullet$ We give tight average-case (gate and wire) complexity results for computing PARITY with depth-two threshold circuits; the answer turns out to be the same as for depth-two majority circuits. The key is a new random restriction lemma for linear threshold functions. Our main analytical tool is the Littlewood-Offord Lemma from additive combinatorics.

cs.CC↗

arXiv2015

Simulating Branching Programs with Edit Distance and Friends or: A Polylog Shaved is a Lower Bound Made

A recent and active line of work achieves tight lower bounds for fundamental problems under the Strong Exponential Time Hypothesis (SETH). A celebrated result of Backurs and Indyk (STOC'15) proves that the Edit Distance of two sequences of length n cannot be computed in strongly subquadratic time under SETH. The result was extended by follow-up works to simpler looking problems like finding the Longest Common Subsequence (LCS). SETH is a very strong assumption, asserting that even linear size CNF formulas cannot be analyzed for satisfiability with an exponential speedup over exhaustive search. We consider much safer assumptions, e.g. that such a speedup is impossible for SAT on much more expressive representations, like NC circuits. Intuitively, this seems much more plausible: NC circuits can implement complex cryptographic primitives, while CNFs cannot even approximately compute an XOR of bits. Our main result is a surprising reduction from SAT on Branching Programs to fundamental problems in P like Edit Distance, LCS, and many others. Truly subquadratic algorithms for these problems therefore have consequences that we consider to be far more remarkable than merely faster CNF SAT algorithms. For example, SAT on arbitrary o(n)-depth bounded fan-in circuits (and therefore also NC-Circuit-SAT) can be solved in (2-eps)^n time. A very interesting feature of our work is that we can prove major consequences even from mildly subquadratic algorithms for Edit Distance or LCS. For example, we show that if we can shave an arbitrarily large polylog factor from n^2 for Edit Distance then NEXP does not have non-uniform NC^1 circuits. A more fine-grained examination shows that even shaving a $\log^c{n}$ factor, for a specific constant $c \approx 10^3$, already implies new circuit lower bounds.

cs.CC↗

arXiv2015

Natural Proofs Versus Derandomization

We study connections between Natural Proofs, derandomization, and the problem of proving "weak" circuit lower bounds such as ${\sf NEXP} \not\subset {\sf TC^0}$. Natural Proofs have three properties: they are constructive (an efficient algorithm $A$ is embedded in them), have largeness ($A$ accepts a large fraction of strings), and are useful ($A$ rejects all strings which are truth tables of small circuits). Strong circuit lower bounds that are "naturalizing" would contradict present cryptographic understanding, yet the vast majority of known circuit lower bound proofs are naturalizing. So it is imperative to understand how to pursue un-Natural Proofs. Some heuristic arguments say constructivity should be circumventable: largeness is inherent in many proof techniques, and it is probably our presently weak techniques that yield constructivity. We prove: $\bullet$ Constructivity is unavoidable, even for $\sf NEXP$ lower bounds. Informally, we prove for all "typical" non-uniform circuit classes ${\cal C}$, ${\sf NEXP} \not\subset {\cal C}$ if and only if there is a polynomial-time algorithm distinguishing some function from all functions computable by ${\cal C}$-circuits. Hence ${\sf NEXP} \not\subset {\cal C}$ is equivalent to exhibiting a constructive property useful against ${\cal C}$. $\bullet$ There are no $\sf P$-natural properties useful against ${\cal C}$ if and only if randomized exponential time can be "derandomized" using truth tables of circuits from ${\cal C}$ as random seeds. Therefore the task of proving there are no $\sf P$-natural properties is inherently a derandomization problem, weaker than but implied by the existence of strong pseudorandom functions. These characterizations are applied to yield several new results, including improved ${\sf ACC}^0$ lower bounds and new unconditional derandomizations.

cs.CC↗

arXiv2015

Probabilistic Polynomials and Hamming Nearest Neighbors

We show how to compute any symmetric Boolean function on $n$ variables over any field (as well as the integers) with a probabilistic polynomial of degree $O(\sqrt{n \log(1/ε)})$ and error at most $ε$. The degree dependence on $n$ and $ε$ is optimal, matching a lower bound of Razborov (1987) and Smolensky (1987) for the MAJORITY function. The proof is constructive: a low-degree polynomial can be efficiently sampled from the distribution. This polynomial construction is combined with other algebraic ideas to give the first subquadratic time algorithm for computing a (worst-case) batch of Hamming distances in superlogarithmic dimensions, exactly. To illustrate, let $c(n) : \mathbb{N} \rightarrow \mathbb{N}$. Suppose we are given a database $D$ of $n$ vectors in $\{0,1\}^{c(n) \log n}$ and a collection of $n$ query vectors $Q$ in the same dimension. For all $u \in Q$, we wish to compute a $v \in D$ with minimum Hamming distance from $u$. We solve this problem in $n^{2-1/O(c(n) \log^2 c(n))}$ randomized time. Hence, the problem is in "truly subquadratic" time for $O(\log n)$ dimensions, and in subquadratic time for $d = o((\log^2 n)/(\log \log n)^2)$. We apply the algorithm to computing pairs with maximum inner product, closest pair in $\ell_1$ for vectors with bounded integer entries, and pairs with maximum Jaccard coefficients.

cs.DS↗

arXiv2014

Faster all-pairs shortest paths via circuit complexity

We present a new randomized method for computing the min-plus product (a.k.a., tropical product) of two $n \times n$ matrices, yielding a faster algorithm for solving the all-pairs shortest path problem (APSP) in dense $n$-node directed graphs with arbitrary edge weights. On the real RAM, where additions and comparisons of reals are unit cost (but all other operations have typical logarithmic cost), the algorithm runs in time \[\frac{n^3}{2^{Ω(\log n)^{1/2}}}\] and is correct with high probability. On the word RAM, the algorithm runs in $n^3/2^{Ω(\log n)^{1/2}} + n^{2+o(1)}\log M$ time for edge weights in $([0,M] \cap {\mathbb Z})\cup\{\infty\}$. Prior algorithms used either $n^3/(\log^c n)$ time for various $c \leq 2$, or $O(M^αn^β)$ time for various $α> 0$ and $β> 2$. The new algorithm applies a tool from circuit complexity, namely the Razborov-Smolensky polynomials for approximately representing ${\sf AC}^0[p]$ circuits, to efficiently reduce a matrix product over the $(\min,+)$ algebra to a relatively small number of rectangular matrix products over ${\mathbb F}_2$, each of which are computable using a particularly efficient method due to Coppersmith. We also give a deterministic version of the algorithm running in $n^3/2^{\log^δ n}$ time for some $δ> 0$, which utilizes the Yao-Beigel-Tarui translation of ${\sf AC}^0[m]$ circuits into "nice" depth-two circuits.

cs.DS↗

arXiv2014

New algorithms and lower bounds for circuits with linear threshold gates

Let $ACC \circ THR$ be the class of constant-depth circuits comprised of AND, OR, and MOD$m$ gates (for some constant $m > 1$), with a bottom layer of gates computing arbitrary linear threshold functions. This class of circuits can be seen as a "midpoint" between $ACC$ (where we know nontrivial lower bounds) and depth-two linear threshold circuits (where nontrivial lower bounds remain open). We give an algorithm for evaluating an arbitrary symmetric function of $2^{n^{o(1)}}$ $ACC \circ THR$ circuits of size $2^{n^{o(1)}}$, on all possible inputs, in $2^n \cdot poly(n)$ time. Several consequences are derived: $\bullet$ The number of satisfying assignments to an $ACC \circ THR$ circuit of subexponential size can be computed in $2^{n-n^{\varepsilon}}$ time (where $\varepsilon > 0$ depends on the depth and modulus of the circuit). $\bullet$ $NEXP$ does not have quasi-polynomial size $ACC \circ THR$ circuits, nor does $NEXP$ have quasi-polynomial size $ACC \circ SYM$ circuits. Nontrivial size lower bounds were not known even for $AND \circ OR \circ THR$ circuits. $\bullet$ Every 0-1 integer linear program with $n$ Boolean variables and $s$ linear constraints is solvable in $2^{n-Ω(n/((\log M)(\log s)^{5}))}\cdot poly(s,n,M)$ time with high probability, where $M$ upper bounds the bit complexity of the coefficients. (For example, 0-1 integer programs with weights in $[-2^{poly(n)},2^{poly(n)}]$ and $poly(n)$ constraints can be solved in $2^{n-Ω(n/\log^6 n)}$ time.) We also present an algorithm for evaluating depth-two linear threshold circuits (a.k.a., $THR \circ THR$) with exponential weights and $2^{n/24}$ size on all $2^n$ input assignments, running in $2^n \cdot poly(n)$ time. This is evidence that non-uniform lower bounds for $THR \circ THR$ are within reach.

cs.CC↗