Source author record

Sang-Won Park

Sang-Won Park appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.soft cond-mat.stat-mech Neural and Evolutionary Computing

Catalog footprint

What is connected

2works

3topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2021arXiv

Correspondence between neuroevolution and gradient descent

We show analytically that training a neural network by conditioned stochastic mutation or neuroevolution of its weights is equivalent, in the limit of small mutations, to gradient descent on the loss function in the presence of Gaussian white noise. Averaged over independent realizations of the learning process, neuroevolution is equivalent to gradient descent on the loss function. We use numerical simulation to show that this correspondence can be observed for finite mutations,for shallow and deep neural networks. Our results provide a connection between two families of neural-network training methods that are usually considered to be fundamentally different.

preprint2016arXiv

Heterogeneous dynamics and its length scale in simple ionic liquid models: A computational study

We numerically investigate the dynamic heterogeneity and its length scale found in the coarse-grained ionic liquid model systems. In our ionic liquid model systems, cations are modeled as dimers with positive charge, while anions are modeled as monomers with negative charge, respectively. To study the effect of the charge distributions on the cations, two ionic liquid models with different charge distributions are used and the model with neutral charge is also considered as a counterpart. To reveal the heterogeneous dynamics in the model systems, we examine spatial distributions of displacement and time distributions of exchange and persistence times. All the models show significant increase of the dynamic heterogeneity as the temperature is lowered. The dynamic heterogeneity is quantified via the well-known four-point susceptibility which measures the fluctuation of a time correlation function. The dynamic correlation length is calculated by fitting the dynamic structure factor, S4(k,t), with Ornstein-Zernike form at the time scale at which the dynamic heterogeneity reaches the maximum value. Obtained time and length scales exhibit a power law relation at the low temperatures, similar to various supercooled liquid models. Especially, the charged model systems show unusual crossover behaviors which are not observed in the uncharged model system. We ascribe the crossover behavior to the enhanced cage effect caused by charges on the particles.