Searcharxiv⌕ Search

arXiv subjects

Jinyoung Park

Publications and source records attributed to Jinyoung Park.

64 records · Page 4Linked to original sources
arXiv2020

Deep neural network for solving differential equations motivated by Legendre-Galerkin approximation

Nonlinear differential equations are challenging to solve numerically and are important to understanding the dynamics of many physical systems. Deep neural networks have been applied to help alleviate the computational cost that is associated with solving these systems. We explore the performance and accuracy of various neural architectures on both linear and nonlinear differential equations by creating accurate training sets with the spectral element method. Next, we implement a novel Legendre-Galerkin Deep Neural Network (LGNet) algorithm to predict solutions to differential equations. By constructing a set of a linear combination of the Legendre basis, we predict the corresponding coefficients, $α_i$ which successfully approximate the solution as a sum of smooth basis functions $u \simeq \sum_{i=0}^{N} α_i φ_i$. As a computational example, linear and nonlinear models with Dirichlet or Neumann boundary conditions are considered.

math.NA↗
arXiv2020

Robust Neural Networks inspired by Strong Stability Preserving Runge-Kutta methods

Deep neural networks have achieved state-of-the-art performance in a variety of fields. Recent works observe that a class of widely used neural networks can be viewed as the Euler method of numerical discretization. From the numerical discretization perspective, Strong Stability Preserving (SSP) methods are more advanced techniques than the explicit Euler method that produce both accurate and stable solutions. Motivated by the SSP property and a generalized Runge-Kutta method, we propose Strong Stability Preserving networks (SSP networks) which improve robustness against adversarial attacks. We empirically demonstrate that the proposed networks improve the robustness against adversarial examples without any defensive methods. Further, the SSP networks are complementary with a state-of-the-art adversarial training scheme. Lastly, our experiments show that SSP networks suppress the blow-up of adversarial perturbations. Our results open up a way to study robust architectures of neural networks leveraging rich knowledge from numerical discretization literature.

cs.CV↗
arXiv2020

Tuza's Conjecture for random graphs

A celebrated conjecture of Zs. Tuza says that in any (finite) graph, the minimum size of a cover of triangles by edges is at most twice the maximum size of a set of edge-disjoint triangles. Resolving a recent question of Bennett, Dudek, and Zerbib, we show that this is true for random graphs; more precisely: \[ \mbox{for any $p=p(n)$, $\mathbb P(\mbox{$G_{n,p}$ satisfies Tuza's Conjecture})\rightarrow 1 $ (as $n\rightarrow\infty$).} \]

math.CO↗
arXiv2019

Thresholds versus fractional expectation-thresholds

Proving a conjecture of Talagrand, a fractional version of the 'expectation-threshold' conjecture of Kalai and the second author, we show for any increasing family $F$ on a finite set $X$ that $p_c (F) =O( q_f (F) \log \ell(F))$, where $p_c(F)$ and $q_f(F)$ are the threshold and 'fractional expectation-threshold' of $F$, and $\ell(F)$ is the largest size of a minimal member of $F$. This easily implies several heretofore difficult results and conjectures in probabilistic combinatorics, including thresholds for perfect hypergraph matchings (Johansson--Kahn--Vu), bounded-degree spanning trees (Montgomery), and bounded-degree spanning graphs (new). We also resolve (and vastly extend) the 'axial' version of the random multi-dimensional assignment problem (earlier considered by Martin--Mézard--Rivoire and Frieze--Sorkin). Our approach builds on a recent breakthrough of Alweiss, Lovett, Wu and Zhang on the Erdős--Rado 'Sunflower Conjecture'.

math.CO↗
arXiv2019

An isoperimetric inequality for the Hamming cube and some consequences

Our basic result, an isoperimetric inequality for Hamming cube $Q_n$, can be written: \[ \int h_A^βdμ\ge 2 μ(A)(1-μ(A)). \] Here $μ$ is uniform measure on $V=\{0,1\}^n$ ($=V(Q_n)$); $β=\log_2(3/2)$; and, for $S\subseteq V$ and $x\in V$, \[ h_S(x) = \begin{cases} d_{V \setminus S}(x) &\mbox{ if } x \in S, 0 &\mbox{ if } x \notin S \end{cases} \] (where $d_T(x)$ is the number of neighbors of $x$ in $T$). This implies inequalities involving mixtures of edge and vertex boundaries, with related stability results, and suggests some more general possibilities. One application, a stability result for the set of edges connecting two disjoint subsets of $V$ of size roughly $|V|/2$, is a key step in showing that the number of maximal independent sets in $Q_n$ is $(1+o(1))2n\exp_2[2^{n-2}]$. This asymptotic statement, whose proof will appear separately, was the original motivation for the present work.

math.CO↗
arXiv2019

The number of maximal independent sets in the Hamming cube

Let $Q_n$ be the $n$-dimensional Hamming cube and $N=2^n$. We prove that the number of maximal independent sets in $Q_n$ is asymptotically \[2n2^{N/4},\] as was conjectured by Ilinca and the first author in connection with a question of Duffus, Frankl and Rödl. The value is a natural lower bound derived from a connection between maximal independent sets and induced matchings. The proof that it is also an upper bound draws on various tools, among them "stability" results for maximal independent set counts and old and new results on isoperimetric behavior in $Q_n$.

math.CO↗
arXiv2019

The number of 4-colorings of the Hamming cube

Let $Q_d$ be the $d$-dimensional hypercube and $N=2^d$. We prove that the number of (proper) 4-colorings of $Q_d$ is asymptotically \[6e2^N,\] as was conjectured by Engbers and Galvin in 2012. The proof uses a combination of information theory (entropy) and isoperimetric ideas originating in work of Sapozhenko in the 1980's.

math.CO↗
arXiv2018

Stability for maximal independent sets

Answering questions of Y. Rabinovich, we prove "stability" versions of upper bounds on maximal independent set counts in graphs under various restrictions. Roughly these say that being close to the maximum implies existence of a large induced matching or triangle matching (depending on assumptions).

math.CO↗