CC BY
10182 Section9
Predicting
a
steady
fluid
flow
over
bluff
bodies
for
shape
optimization
using
machine
learning
I. A. Plokhikh1,2,R. I. Mullyadzhanov1,2
1Novosibirsk State University
2Institute of Thermophysics SB RAS
Email: ivan.ploxix@gmail.com
DOI 10.24412/cl.35065.2021.1.02.71
In this study we apply a Machine Learning methods to the problem of the flow over a bluff body which
shapeistobe optimizedusing ConvolutionalNeural Network(CNN)andReinforcementLearning(RL).Inwork
of Viquerat et al. [1] it was shown that neural networks trained with RL algorithm are able to find optimal
shapes for aerodynamics. Due to the high computational costs required by CFD solvers, it was proposedto use
CNN to approximate stationary solutions of the Navier � Stokes equations [2]. The advantage of this solution is
a significant reductionintime requiredto obtainasolution(about2.3orders) in comparison with the direct
calculation by CFD solver at the cost of a small error rates. This acceleration makes it possible to reduce the
computational costs in the problem of finding the optimal hydrodynamic shape using the Reinforcement
Learning algorithm, where it is necessary to obtain a large number of solutions when searching for optimal
parameters of bluff body geometry.
References:
1. Viquerat J., Rabault J., Kuhnle A., Ghraieb H., Larcher A., Hachem E. Direct shape optimization through Deep
Reinforcement Learning//J.of Computational Physics.2021.V.428,A.110080.
2. Ribeiro M. D., Rehman A., Ahmed S., Dengel A. DeepCFD: EfficientSteady.State Laminar Flow Approximation with
Deep Convolutional Neural Networks. [Electron. resource]. URL: https://arxiv.org/abs/2004.08826 (the date of access:
29.12.2020).
Identification
of
argumentative
sentences
in
scientific
and
popular
science
texts
N. V. Salomatina1,I. S. Pimenov2
1Sobolev Instituteof Mathematics
2Novosibirsk State University
Email: salomatina_nv@live.ru
DOI 10.24412/cl.35065.2021.1.02.72
In this study we analyze the applicability of specific machine learning algorithms to the task of detecting
argumentative sentences in Russian text. We employ a collection of scientific and popular science texts with
manually annotated argumentation to evaluate the quality of identifying argumentative sentences in terms of
precision, recall, andF.measure. The experiment involves three algorithms: MNB, SVM, and MLP in Scikit.learn
implementation. The bag of words model is used for representing texts. Lemmas of words in analyzed sentences
serve as features for the classification. We perform the automatic selection of informative features in
accordance with Variance and .2 criteria combined with weight.based filtration of lemmas (via TF*IDF and
EMI). The training set includes around 800 sentences, while the test set contains 180. The MNB algorithm
demonstrates highestF.measure and recall scores on almost all feature, while the MLP algorithm shows the
bestprecisionforaboutahalfof featureselectionvariations.
The study was carried out within the framework of the state contract of the Sobolev Institute of Mathematics (project
no. 0314.2019.0015).
