Skip navigation
Please use this identifier to cite or link to this item: https://libeldoc.bsuir.by/handle/123456789/45891
Title: Developing a Seq2Seq neural network using visual attention to transform mathematical expressions from images to LaTeX
Authors: Vyaznikov, P.
Kotilevets, I.
Keywords: материалы конференций;conference proceedings;neural network;images to LaTeX
Issue Date: 2021
Publisher: БГУИР
Citation: Vyaznikov, P. Developing a Seq2Seq neural network using visual attention to transform mathematical expressions from images to LaTeX / P. Vyaznikov, I. Kotilevets // Nano-Desing, Tehnology, Computer Simulations=Нанопроектирование, технология, компьютерное моделирование (NDTCS-2021) : тезисы докладов XIX Международного симпозиума, Минск, 28-29 октября 2021 года / Белорусский государственный университет информатики и радиоэлектроники ; редкол.: В. А. Богуш [и др.]. – Минск, 2021. – P. 61–63.
Abstract: The presented neural network with the seq2seq architecture and Attention mechanism successfully solves the im2latex problem, which is confirmed by the results of measuring metrics. Generated captions for images with equations are quite accurate and, in most cases, coincide with the real ones. Such a solution can be used in mathematical programs to automatically translate images into LaTeX and further solve and analyze the resulting equations or expressions. The scope of use can be expanded by training the network on images with handwritten equations. A similar technology is used in the PhotoMath application, however, it has low accuracy and a small set of supported mathematical symbols. The described solution is devoid of both problems.
URI: https://libeldoc.bsuir.by/handle/123456789/45891
Appears in Collections:NDTCS 2021

Files in This Item:
File Description SizeFormat 
Vyaznikov_Developing.pdf476.9 kBAdobe PDFView/Open
Show full item record Google Scholar

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.