arXiv subjects

Noga Alon

Publications and source records attributed to Noga Alon.

At least 109 records · Page 6Linked to original sources

arXiv2019

Limits of Private Learning with Access to Public Data

We consider learning problems where the training set consists of two types of examples: private and public. The goal is to design a learning algorithm that satisfies differential privacy only with respect to the private examples. This setting interpolates between private learning (where all examples are private) and classical learning (where all examples are public). We study the limits of learning in this setting in terms of private and public sample complexities. We show that any hypothesis class of VC-dimension $d$ can be agnostically learned up to an excess error of $α$ using only (roughly) $d/α$ public examples and $d/α^2$ private labeled examples. This result holds even when the public examples are unlabeled. This gives a quadratic improvement over the standard $d/α^2$ upper bound on the public sample complexity (where private examples can be ignored altogether if the public examples are labeled). Furthermore, we give a nearly matching lower bound, which we prove via a generic reduction from this setting to the one of private learning without public data.

cs.LG↗

arXiv2019

High-girth near-Ramanujan graphs with localized eigenvectors

We show that for every prime $d$ and $α\in (0,1/6)$, there is an infinite sequence of $(d+1)$-regular graphs $G=(V,E)$ with girth at least $2α\log_{d}(|V|)(1-o_d(1))$, second adjacency matrix eigenvalue bounded by $(3/\sqrt{2})\sqrt{d}$, and many eigenvectors fully localized on small sets of size $O(|V|^α)$. This strengthens the results of Ganguly-Srivastava, who constructed high girth (but not expanding) graphs with similar properties, and may be viewed as a discrete analogue of the "scarring" phenomenon observed in the study of quantum ergodicity on manifolds. Key ingredients in the proof are a technique of Kahale for bounding the growth rate of eigenfunctions of graphs, discovered in the context of vertex expansion and a method of Erdős and Sachs for constructing high girth regular graphs.

math.CO↗

arXiv2019

Deterministic Combinatorial Replacement Paths and Distance Sensitivity Oracles

In this work we derandomize two central results in graph algorithms, replacement paths and distance sensitivity oracles (DSOs) matching in both cases the running time of the randomized algorithms. For the replacement paths problem, let G = (V,E) be a directed unweighted graph with n vertices and m edges and let P be a shortest path from s to t in G. The {\sl replacement paths} problem is to find for every edge e \in P the shortest path from s to t avoiding e. Roditty and Zwick [ICALP 2005] obtained a randomized algorithm with running time of ~O(m \sqrt{n}). Here we provide the first deterministic algorithm for this problem, with the same ~O(m \sqrt{n}) time. For the problem of distance sensitivity oracles, let G = (V,E) be a directed graph with real-edge weights. An f-Sensitivity Distance Oracle (f-DSO) gets as input the graph G=(V,E) and a parameter f, preprocesses it into a data-structure, such that given a query (s,t,F) with s,t \in V and F \subseteq E \cup V, |F| \le f being a set of at most f edges or vertices (failures), the query algorithm efficiently computes the distance from s to t in the graph G \setminus F ({\sl i.e.}, the distance from s to t in the graph G after removing from it the failing edges and vertices F). For weighted graphs with real edge weights, Weimann and Yuster [FOCS 2010] presented a combinatorial randomized f-DSO with ~O(mn^{4-α}) preprocessing time and subquadratic ~O(n^{2-2(1-α)/f}) query time for every value of 0 < α< 1. We derandomize this result and present a combinatorial deterministic f-DSO with the same asymptotic preprocessing and query time.

cs.DS↗

arXiv2019

Efficient arithmetic regularity and removal lemmas for induced bipartite patterns

Let $G$ be an abelian group of bounded exponent and $A \subseteq G$. We show that if the collection of translates of $A$ has VC dimension at most $d$, then for every $ε>0$ there is a subgroup $H$ of $G$ of index at most $ε^{-d-o(1)}$ such that one can add or delete at most $ε|G|$ elements to/from $A$ to make it a union of $H$-cosets. We also establish a removal lemma with polynomial bounds, with applications to property testing, for induced bipartite patterns in a finite abelian group with bounded exponent.

math.CO↗

arXiv2019

Isoperimetry, Stability, and Irredundance in Direct Products

The direct product of graphs $G_1,\ldots,G_n$ is the graph with vertex set $V(G_1)\times\cdots\times V(G_n)$ in which two vertices $(g_1,\ldots,g_n)$ and $(g_1',\ldots,g_n')$ are adjacent if and only if $g_i$ is adjacent to $g_i'$ in $G_i$ for all $i$. Building off of the recent work of Brakensiek, we prove an optimal vertex isoperimetric inequality for direct products of complete multipartite graphs. Applying this inequality, we derive a stability result for independent sets in direct products of balanced complete multipartite graphs, showing that every large independent set must be close to the maximal independent set determined by setting one of the coordinates to be constant. Armed with these isoperimetry and stability results, we prove that the upper irredundance number of a direct product of balanced complete multipartite graphs is equal to its independence number in all but at most $37$ cases. This proves most of a conjecture of Burcroff that arose as a strengthening of a conjecture of the second author and Iyer. We also propose a further strengthening of Burcroff's conjecture.

math.CO↗

arXiv2019

Private PAC learning implies finite Littlestone dimension

We show that every approximately differentially private learning algorithm (possibly improper) for a class $H$ with Littlestone dimension~$d$ requires $Ω\bigl(\log^*(d)\bigr)$ examples. As a corollary it follows that the class of thresholds over $\mathbb{N}$ can not be learned in a private manner; this resolves open question due to [Bun et al., 2015, Feldman and Xiao, 2015]. We leave as an open question whether every class with a finite Littlestone dimension can be learned by an approximately differentially private algorithm.

cs.LG↗

arXiv2019

The Minrank of Random Graphs over Arbitrary Fields

The minrank of a graph $G$ on the set of vertices $[n]$ over a field $\mathbb{F}$ is the minimum possible rank of a matrix $M\in\mathbb{F}^{n\times n}$ with nonzero diagonal entries such that $M_{i,j}=0$ whenever $i$ and $j$ are distinct nonadjacent vertices of $G$. This notion, over the real field, arises in the study of the Lovász theta function of a graph. We obtain tight bounds for the typical minrank of the binomial random graph $G(n,p)$ over any finite or infinite field, showing that for every field $\mathbb{F}=\mathbb F(n)$ and every $p=p(n)$ satisfying $n^{-1} \leq p \leq 1-n^{-0.99}$, the minrank of $G=G(n,p)$ over $\mathbb{F}$ is $Θ(\frac{n \log (1/p)}{\log n})$ with high probability. The result for the real field settles a problem raised by Knuth in 1994. The proof combines a recent argument of Golovnev, Regev, and Weinstein, who proved the above result for finite fields of size at most $n^{O(1)}$, with tools from linear algebra, including an estimate of Rónyai, Babai, and Ganapathy for the number of zero-patterns of a sequence of polynomials.

math.CO↗

arXiv2018

Additive Approximation of Generalized Turán Questions

For graphs $G$ and $T$, and a family of graphs $\mathcal{F}$ let $\mathrm{ex}(G,T,\mathcal{F})$ denote the maximum possible number of copies of $T$ in an $\mathcal{F}$-free subgraph of $G$. We investigate the algorithmic aspects of calculating and estimating this function. We show that for every graph $T$, finite family $\mathcal{F}$ and constant $ε>0$ there is a polynomial time algorithm that approximates $\mathrm{ex}(G,T,\mathcal{F})$ for an input graph $G$ on $n$ vertices up to an additive error of $εn^{v(T)}$. We also consider the possibility of a better approximation, proving several positive and negative results, and suggesting a conjecture on the exact relation between $T$ and $\mathcal{F}$ for which no significantly better approximation can be found in polynomial time unless $P=NP$.

math.CO↗

arXiv2018

Addressing Johnson graphs, complete multipartite graphs, odd cycles and other graphs

Graham and Pollak showed that the vertices of any graph $G$ can be addressed with $N$-tuples of three symbols, such that the distance between any two vertices may be easily determined from their addresses. An addressing is optimal if its length $N$ is minimum possible. In this paper, we determine an addressing of length $k(n-k)$ for the Johnson graphs $J(n,k)$ and we show that our addressing is optimal when $k=1$ or when $k=2, n=4,5,6$, but not when $n=6$ and $k=3$. We study the addressing problem as well as a variation of it in which the alphabet used has more than three symbols, for other graphs such as complete multipartite graphs and odd cycles. We also present computations describing the distribution of the minimum length of addressings for connected graphs with up to $10$ vertices. Motivated by these computations we settle a problem of Graham, showing that most graphs on $n$ vertices have an addressing of length at most $n-(2-o(1))\log_2 n$.

math.CO↗

arXiv2018

Traces of Hypergraphs

Let $\text{Tr}(n,m,k)$ denote the largest number of distinct projections onto $k$ coordinates guaranteed in any family of $m$ binary vectors of length $n$. The classical Sauer-Perles-Shelah Lemma implies that $\text{Tr}(n, n^r, k) = 2^k$ for $k \le r$. While determining $\text{Tr}(n,n^r,k)$ precisely for general $k$ seems hopeless even for constant $r$, estimating it, and more generally estimating the function $\text{Tr}(n,m,k)$ for all range of the parameters, remains a widely open problem with connections to important questions in computer science and combinatorics. Here we essentially resolve this problem when $k$ is linear and $m=n^r$ where $r$ is constant, proving that, for any constant $α>0$, $\text{Tr}(n,n^r,αn) = \tildeΘ(n^C)$ with $C=C(r,α)=\frac{r+1-\log(1+α)}{2-\log(1+α)}$. For the proof we establish a "sparse" version of another classical result, the Kruskal-Katona Theorem, which gives a stronger guarantee when the hypergraph does not induce dense sub-hypergraphs. Furthermore, we prove that the parameters in our sparse Kruskal-Katona theorem are essentially best possible. Finally, we mention two simple applications which may be of independent interest.

math.CO↗

arXiv2018

Multitasking Capacity: Hardness Results and Improved Constructions

We consider the problem of determining the maximal $α\in (0,1]$ such that every matching $M$ of size $k$ (or at most $k$) in a bipartite graph $G$ contains an induced matching of size at least $α|M|$. This measure was recently introduced in Alon et al. (NIPS 2018) and is motivated by connectionist models of cognition as well as modeling interference in wireless and communication networks. We prove various hardness results for computing $α$ either exactly or approximately. En route to our results, we also consider the maximum connected matching problem: determining the largest matching $N$ in a graph $G$ such that every two edges in $N$ are connected by an edge. We prove a nearly optimal $n^{1-ε}$ hardness of approximation result (under randomized reductions) for connected matching in bipartite graphs (with both sides of cardinality $n$). Towards this end we define bipartite half-covers: A new combinatorial object that may be of independent interest. To the best of our knowledge, the best previous hardness result for the connected matching problem was some constant $β>1$. Finally, we demonstrate the existence of bipartite graphs with $n$ vertices on each side of average degree $d$, that achieve $α=1/2-ε$ for matchings of size sufficiently smaller than $n/poly(d)$. This nearly matches the trivial upper bound of $1/2$ on $α$ which holds for any graph containing a path of length 3.

cs.DS↗

arXiv2018

List-decodable zero-rate codes

We consider list-decoding in the zero-rate regime for two cases: the binary alphabet and the spherical codes in Euclidean space. Specifically, we study the maximal $τ\in [0,1]$ for which there exists an arrangement of $M$ balls of relative Hamming radius $τ$ in the binary hypercube (of arbitrary dimension) with the property that no point of the latter is covered by $L$ or more of them. As $M\to \infty$ the maximal $τ$ decreases to a well-known critical value $τ_L$. In this work, we prove several results on the rate of this convergence. For the binary case, we show that the rate is $Θ(M^{-1})$ when $L$ is even, thus extending the classical results of Plotkin and Levenshtein for $L=2$. For $L=3$ the rate is shown to be $Θ(M^{-\tfrac{2}{3}})$. For the similar question about spherical codes, we prove the rate is $Ω(M^{-1})$ and $O(M^{-\tfrac{2L}{L^2-L+2}})$.

cs.IT↗

arXiv2018

Ramsey-nice families of graphs

For a finite family $\mathcal{F}$ of fixed graphs let $R_k(\mathcal{F})$ be the smallest integer $n$ for which every $k$-coloring of the edges of the complete graph $K_n$ yields a monochromatic copy of some $F\in\mathcal{F}$. We say that $\mathcal{F}$ is $k$-nice if for every graph $G$ with $χ(G)=R_k(\mathcal{F})$ and for every $k$-coloring of $E(G)$ there exists a monochromatic copy of some $F\in\mathcal{F}$. It is easy to see that if $\mathcal{F}$ contains no forest, then it is not $k$-nice for any $k$. It seems plausible to conjecture that a (weak) converse holds, namely, for any finite family of graphs $\mathcal{F}$ that contains at least one forest, and for all $k\geq k_0(\mathcal{F})$ (or at least for infinitely many values of $k$), $\mathcal{F}$ is $k$-nice. We prove several (modest) results in support of this conjecture, showing, in particular, that it holds for each of the three families consisting of two connected graphs with 3 edges each and observing that it holds for any family $\mathcal{F}$ containing a forest with at most 2 edges. We also study some related problems and disprove a conjecture by Aharoni, Charbit and Howard regarding the size of matchings in regular 3-partite 3-uniform hypergraphs.

math.CO↗

arXiv2018

Sums, products and ratios along the edges of a graph

In their seminal paper Erdős and Szemerédi formulated conjectures on the size of sumset and product set of integers. The strongest form of their conjecture is about sums and products along the edges of a graph. In this paper we show that this strong form of the Erdős-Szemerédi conjecture does not hold. We give upper and lower bounds on the cardinalities of sumsets, product sets and ratio sets along the edges of graphs.

math.CO↗

arXiv2017

Out-colourings of Digraphs

We study vertex colourings of digraphs so that no out-neighbourhood is monochromatic and call such a colouring an {\bf out-colouring}. The problem of deciding whether a given digraph has an out-colouring with only two colours (called a 2-out-colouring) is ${\cal NP}$-complete. We show that for every choice of positive integers $r,k$ there exists a $k$-strong bipartite tournament which needs at least $r$ colours in every out-colouring. Our main results are on tournaments and semicomplete digraphs. We prove that, except for the Paley tournament $P_7$, every strong semicomplete digraph of minimum out-degree at least 3 has a 2-out-colouring. Furthermore, we show that every semicomplete digraph on at least 7 vertices has a 2-out-colouring if and only if it has a {\bf balanced} such colouring, that is, the difference between the number of vertices that receive colour 1 and colour 2 is at most one. In the second half of the paper we consider the generalization of 2-out-colourings to vertex partitions $(V_1,V_2)$ of a digraph $D$ so that each of the three digraphs induced by respectively, the vertices of $V_1$, the vertices of $V_2$ and all arcs between $V_1$ and $V_2$ have minimum out-degree $k$ for a prescribed integer $k\geq 1$. Using probabilistic arguments we prove that there exists an absolute positive constant $c$ so that every semicomplete digraph of minimum out-degree at least $2k+c\sqrt{k}$ has such a partition. This is tight up to the value of $c$.

cs.DM↗

arXiv2017

Many cliques in $H$-free subgraphs of random graphs

For two fixed graphs $T$ and $H$ let $ex(G(n,p),T,H)$ be the random variable counting the maximum number of copies of $T$ in an $H$-free subgraph of the random graph $G(n,p)$. We show that for the case $T=K_m$ and $χ(H)> m$ the behavior of $ex(G(n,p),K_m,H)$ depends strongly on the relation between $p$ and $m_2(H)=\max_{H'\subset H, |V(H')|'\geq 3}\left\{ \frac{e(H')-1}{v(H')-2} \right\}$. When $m_2(H)> m_2(K_m)$ we prove that with high probability, depending on the value of $p$, either one can maintain almost all copies of $K_m$, or it is asymptotically best to take a $χ(H)-1$ partite subgraph of $G(n,p)$. The transition between these two behaviors occurs at $p=n^{-1/m_2(H)}$. When $m_2(H)< m_2(K_m)$ we show that the above cases still exist, however for $δ>0$ small at $p=n^{-1/m_2(H)+δ}$ one can typically still keep most of the copies of $K_m$ in an $H$-free subgraph of $G(n,p)$. Thus, the transition between the two behaviors in this case occurs at some $p$ significantly bigger than $n^{-1/m_2(H)}$. To show that the second case is not redundant we present a construction which may be of independent interest. For each $k \geq 4$ we construct a family of $k$ chromatic graphs $G(k,ε_i)$ where $m_2(G(k,ε_i))$ tends to $\frac{(k+1)(k-2)}{2(k-1)} (< m_2(K_{k-1}))$ as $i$ tends to infinity. This is tight for all values of $k$

math.CO↗

arXiv2017

Submultiplicative Glivenko-Cantelli and Uniform Convergence of Revenues

In this work we derive a variant of the classic Glivenko-Cantelli Theorem, which asserts uniform convergence of the empirical Cumulative Distribution Function (CDF) to the CDF of the underlying distribution. Our variant allows for tighter convergence bounds for extreme values of the CDF. We apply our bound in the context of revenue learning, which is a well-studied problem in economics and algorithmic game theory. We derive sample-complexity bounds on the uniform convergence rate of the empirical revenues to the true revenues, assuming a bound on the $k$th moment of the valuations, for any (possibly fractional) $k>1$. For uniform convergence in the limit, we give a complete characterization and a zero-one law: if the first moment of the valuations is finite, then uniform convergence almost surely occurs; conversely, if the first moment is infinite, then uniform convergence almost never occurs.

cs.LG↗

arXiv2017

Clique coloring of dense random graphs

The clique chromatic number of a graph G=(V,E) is the minimum number of colors in a vertex coloring so that no maximal (with respect to containment) clique is monochromatic. We prove that the clique chromatic number of the binomial random graph G=G(n,1/2) is, with high probability, Ω(log n). This settles a problem of McDiarmid, Mitsche and Pralat who proved that it is O(log n) with high probability.

math.CO↗