arXiv subjects

Noah Golowich

Publications and source records attributed to Noah Golowich.

62 records · Page 4Linked to original sources

arXiv2019

Round Complexity of Common Randomness Generation: The Amortized Setting

We study the effect of rounds of interaction on the common randomness generation (CRG) problem. In the CRG problem, two parties, Alice and Bob, receive samples $X_i$ and $Y_i$, respectively, drawn jointly from a source distribution $μ$. The two parties wish to agree on a common random key consisting of many bits of randomness, by exchanging messages that depend on each party's input and the previous messages. In this work we study the amortized version of the problem, i.e., the number of bits of communication needed per random bit output by Alice and Bob, in the limit as the number of bits generated tends to infinity. The amortized version of the CRG problem has been extensively studied, though very little was known about the effect of interaction on this problem. Recently Bafna et al. (SODA 2019) considered the non-amortized version of the problem: they gave a family of sources $μ_{r,n}$ parameterized by $r,n\in\mathbb{N}$, such that with $r+2$ rounds of communication one can generate $n$ bits of common randomness with this source with $O(r\log n)$ communication, whereas with roughly $r/2$ rounds the communication complexity is $Ω(n/{\rm poly}\log n)$. Note that their source is designed with the target number of bits in mind and hence the result does not apply to the amortized setting. In this work we strengthen the work of Bafna et al. in two ways: First we show that the results extend to the amortized setting. We also reduce the gap between the round complexity in the upper and lower bounds to an additive constant. Specifically we show that for every pair $r,n \in \mathbb{N}$ the (amortized) communication complexity to generate $Ω(n)$ bits of common randomness from the source $μ_{r,n}$ using $r+2$ rounds of communication is $O(r\log n)$ whereas the amortized communication required to generate the same amount of randomness from $r$ rounds is $Ω(\sqrt n)$.

cs.IT↗

arXiv2018

The m-Degenerate Chromatic Number of a Digraph

The digraph chromatic number of a directed graph $D$, denoted $χ_A(D)$, is the minimum positive integer $k$ such that there exists a partition of the vertices of $D$ into $k$ disjoint sets, each of which induces an acyclic subgraph. For any $m \geq 1$, a digraph is weakly $m$-degenerate if each of its induced subgraphs has a vertex of in-degree or out-degree less than $m$. We introduce a generalization of the digraph chromatic number, namely $χ_m(D)$, which is the minimum number of sets into which the vertices of a digraph $D$ can be partitioned so that each set induces a weakly $m$-degenerate subgraph. We show that for all digraphs $D$ without directed 2-cycles, $χ_m(D) \leq \frac{2Δ(D)}{4m+1} + O(1)$. Because $χ_1(D) = χ_A(D)$, we obtain as a corollary that $χ_A(D) \leq 2/5 \cdot Δ(D) + O(1)$. We then use this bound to show that $χ_A(D) \leq \sqrt{2/3} \cdot \tildeΔ(D) + O(1)$, substantially improving a bound of Harutyunyan and Mohar that states that $χ_A(D) \leq (1 - e^{-13})\cdot \tildeΔ(D)$ for large enough $\tildeΔ(D)$.

math.CO↗

arXiv2018

Coloring Chains for Compression with Uncertain Priors

Haramaty and Sudan considered the problem of transmitting a message between two people, Alice and Bob, when Alice's and Bob's priors on the message are allowed to differ by at most a given factor. To find a deterministic compression scheme for this problem, they showed that it is sufficient to obtain an upper bound on the chromatic number of a graph, denoted $U(N,s,k)$ for parameters $N,s,k$, whose vertices are nested sequences of subsets and whose edges are between vertices that have similar sequences of sets. In turn, there is a close relationship between the problem of determining the chromatic number of $U(N,s,k)$ and a local graph coloring problem considered by Erdős et al. We generalize the results of Erdős et al. by finding bounds on the chromatic numbers of graphs $H$ and $G$ when there is a homomorphism $ϕ:H\rightarrow G$ that satisfies a nice property. We then use these results to improve upper and lower bounds on $χ(U(N,s,k))$.

math.CO↗

arXiv2018

Communication-Rounds Tradeoffs for Common Randomness and Secret Key Generation

We study the role of interaction in the Common Randomness Generation (CRG) and Secret Key Generation (SKG) problems. In the CRG problem, two players, Alice and Bob, respectively get samples $X_1,X_2,\dots$ and $Y_1,Y_2,\dots$ with the pairs $(X_1,Y_1)$, $(X_2, Y_2)$, $\dots$ being drawn independently from some known probability distribution $μ$. They wish to communicate so as to agree on $L$ bits of randomness. The SKG problem is the restriction of the CRG problem to the case where the key is required to be close to random even to an eavesdropper who can listen to their communication (but does not have access to the inputs of Alice and Bob). In this work, we study the relationship between the amount of communication and the number of rounds of interaction in both the CRG and the SKG problems. Specifically, we construct a family of distributions $μ= μ_{r, n,L}$, parametrized by integers $r$, $n$ and $L$, such that for every $r$ there exists a constant $b = b(r)$ for which CRG (respectively SKG) is feasible when $(X_i,Y_i) \sim μ_{r,n,L}$ with $r+1$ rounds of communication, each consisting of $O(\log n)$ bits, but when restricted to $r/2 - 3$ rounds of interaction, the total communication must exceed $Ω(n/\log^{b}(n))$ bits. Prior to our work no separations were known for $r \geq 2$.

cs.IT↗

arXiv2018

Theory of Deep Learning IIb: Optimization Properties of SGD

In Theory IIb we characterize with a mix of theory and experiments the optimization of deep convolutional networks by Stochastic Gradient Descent. The main new result in this paper is theoretical and experimental evidence for the following conjecture about SGD: SGD concentrates in probability -- like the classical Langevin equation -- on large volume, "flat" minima, selecting flat minimizers which are with very high probability also global minimizers

cs.LG↗

arXiv2014

Acyclic Subgraphs of Planar Digraphs

An acyclic set in a digraph is a set of vertices that induces an acyclic subgraph. In 2011, Harutyunyan conjectured that every planar digraph on $n$ vertices without directed 2-cycles possesses an acyclic set of size at least $3n/5$. We prove this conjecture for digraphs where every directed cycle has length at least 8. More generally, if $g$ is the length of the shortest directed cycle, we show that there exists an acyclic set of size at least $(1 - 3/g)n$.

math.CO↗

arXiv2014

Resolving a Conjecture on Degree of Regularity of Linear Homogeneous Equations

A linear equation is $r$-regular, if, for every $r$-coloring of the positive integers, there exist positive integers of the same color which satisfy the equation. In 2005, Fox and Radoicic conjectured that the equation $x_1 + 2x_2 + \cdots + 2^{n-2}x_{n-1} - 2^{n-1}x_n = 0$, for any $n \geq 2$, has a degree of regularity of $n-1$, which would verify a conjecture of Rado from 1933. Rado's conjecture has since been verified with a different family of equations. In this paper, we show that Fox and Radoicic's family of equations indeed have a degree of regularity of $n-1$. We also provide a few extensions of this result.

math.CO↗

arXiv2014

Degree of Regularity of Linear Homogeneous Equations

We define a linear homogeneous equation to be strongly r-regular if, when a finite number of inequalities is added to the equation, the system of the equation and inequalities is still r-regular. In this paper, we show that, if a linear homogeneous equation is r-regular, then it is strongly r-regular. In 2009, Alexeev and Tsimerman introduced a family of equations, each of which is (n-1)-regular but not n-regular, verifying a conjecture of Rado from 1933. These equations are actually strongly (n-1)-regular as an immediate corollary of our results.

math.CO↗