3D NDM-Net Training Dataset Construction

3D NDM-Net Training Dataset Construction Full article

Conference

The 26th International Conference on Computational Science and Its Applications
30 Jun - 3 Jul 2026 , Braga, University of Minho

Journal

Lecture Notes in Computer Science
ISSN: 0302-9743 , E-ISSN: 1611-3349

Output data

Year: 2026, Volume: 16761, Pages: 70-83 Pages count : 14 DOI: 10.1007/978-3-032-30527-5_5

Authors

Gondyul Elena ¹ , Lisitsa Vadim ¹ , Vishnevsky Dmitry ¹

Affiliations

1	Sobolev Institute of Mathematics SB RAS, 4 Koptug ave., Novosibirsk, 630090, Russia

Funding (1)

Russian Science Foundation

22-11-00004-П

One of the important stages in training neural networks is the construction of a training dataset, especially when it is necessary to minimise its size in order to achieve efficiency. Using the example of a neural network that approximates the fine-grid solution from a coarse-grid solution of a seismic modelling problem, we use a clustering algorithm to form a representative training dataset. In this case, the metric used is the distance between sources. This achieves a solution accuracy that is 10% higher than when using other training dataset construction scenarios.

Gondyul E. , Lisitsa V. , Vishnevsky D.
3D NDM-Net Training Dataset Construction
Lecture Notes in Computer Science. 2026. V.16761. P.70-83. DOI: 10.1007/978-3-032-30527-5_5 OpenAlex

Published online:

Jul 3, 2026

≡ OpenAlex:

W7167074895