Skip navigation


  • DSpace logo
  • Home
  • Collections
  • Researcher Profile
  • Explore by
    • Researcher Profile
  • VGU library
  • Help
  • User Guide
  • Sign on to:
    • My DSpace
    • Receive email
      updates
    • Edit Account details

VGU RESEARCH REPOSITORY


Please use this identifier to cite or link to this item: https://epub.vgu.edu.vn/handle/dlibvgu/1778
DC FieldValueLanguage
dc.contributor.authorTruong Minh Hieuen_US
dc.date.accessioned2024-02-16T04:37:50Z-
dc.date.available2024-02-16T04:37:50Z-
dc.date.issued2022-
dc.identifier.urihttps://epub.vgu.edu.vn/handle/dlibvgu/1778-
dc.description.abstractRecent COVID-19 pandemic outbreaks lead to the global shortage of RT-PCR test kits. Subsequently, the hospitals and healthcare centers are utilizing lung CT scans to alternate the chemical method. Having an enormous number of cases, deep learning enhanced approaches have been exploited to reduce the repetitive classification task. Several high-quality pieces of research on the convolutional neural network (CNN) backbones were conducted and resulted in significant improvements. This thesis aims to further enhance the state-of-art performance with the adoption of modern recurrent neurons to CNN heads. In the following experiments, the oldfashioned fully-connected layers are replaced by a spare and compact wiring strategy called neural circuit policies with liquid time-constant units encapsulated inside. The architecture was confirmed to have outstanding robustness, stability, dynamics, and interpretability when handling sequential information. However, it does not support non-sequential data. For compatibility, input images have to be reshaped to simulate sequential characteristics. In particular, we investigated sequence modeling on the y-axis, z-axis, and patches. The experiments are organized into three comparison stages to save time and computational resources. Models are selectively upgraded from lightweight to middleweight and then maximum potential concerning hardware capacity. Ultimately, the sequential modeling along the z-axis (Z-NCP) provided the best result. However, it is still a bit behind the traditional fullyconnected networks due to the limitation of hardware power. With the current explosive development of technology, Z-NCP can be revisited soon with more enlarged configurations.en_US
dc.language.isoenen_US
dc.publisherVietnamese-German University-
dc.rightsAttribution-NonCommercial 4.0 International*
dc.rights.urihttps://creativecommons.org/licenses/by-nc/4.0/*
dc.subjectRecurrent neuronsen_US
dc.subjectConvolutional neural networksen_US
dc.titleAn adoption of recurrent neurons to convolutional neural networksen_US
dc.typeThesisen_US
item.fulltextWith Fulltext-
item.languageiso639-1other-
item.grantfulltextrestricted-
Appears in Collections:Computer Science (CS)
Files in This Item:
File Description SizeFormat Existing users please Login
An adoption of recurrent neurons to convolutional neural networks.pdf2.01 MBAdobe PDF
Show simple item record

Page view(s)

145
checked on Aug 24, 2025

Download(s)

46
checked on Aug 24, 2025

Google ScholarTM

Check


This item is licensed under a Creative Commons License Creative Commons

© Copyright 2020 by Vietnamese - German University Library.
Add: Ring road 4, Quarter 4, Thoi Hoa Ward, Ben Cat City, Binh Duong Province
Tel.:(0274) 222 0990. Ext.: 70206