Amaanullah, Fairuuz Nurdiaz (2024) Discreate Variational Graph Autoencoder-# untuk Klasifikasi Citra Digit. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of ACCESS CLOSED BY AUTHOR]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text (ACCESS CLOSED BY AUTHOR)
5002201042_Undergraduate_THesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (4MB) | Request a copy |
|
Text
5002201042-Item Removal Request.docx Download (233kB) |
Abstract
Graph neural networks (GNN) mulai banyak dikembangkan dikarenakan kemampuannya dalam mengekstraksi fitur yang lebih representatif dibandingkan neural networks (NNs) pada umumnya. Sebagai gambaran pada data citra, GNNs mampu memberikan informasi konektivitas piksel (delapan konektivitas) lebih lengkap dibandingkan NNs lainnya (empat konektivitas) yang didasarkan atas matriks ketetanggaan. Lebih lanjut, untuk meningkatkan kemampuan GNNs dalam merekonstruksi citra, digunakan model variational autoencoder (VAE) didasarkan atas distribusi kontinu. Namun, VAE dikenal sering menyebabkan posterior collapse. Oleh karena itu, penelitian Tugas Akhir ini mengusulkan sebuah model yang mengintegrasikan GNNs dengan discrete VAE-# pada kasus klasifikasi citra digit yang dinamakan Discrete Variational Graph Autoencoder (DVGAE-#). Penelitian ini bertujuan untuk memberikan variabel laten diskrit yang representatif pada klasifikasi citra digit. Pada penelitian ini, model yang diusulkan diuji pada dataset sekunder pengenalan citra digit. Hasil eksperimen menunjukkan bahwa model yang diusulkan mampu mengklasifikasikan citra digit lebih akurat dibandingkan model pembandingnya dengan cross error = 0.0713 dan akurasi F1-score adalah 0.9699.
=================================================================================================================================
Graph neural networks (GNNs) have been attracted for their ability to capture more representative features against common neural networks (NNs). For instance, on image data, GNNs can provide more informative pixel connectivity (eight connectivities) based on an adjacency matrix compared to other NNs (four connectivities). Moreover, to upgrade the GNN’s ability to image reconstruction, we combined GNNs with variational autoencoder (VAE) from continuous distributions. However, VAE is a well-known model for the posterior collapse issue. Therefore, this study proposes a model that integrates GNNs and discrete VAE-# in digit classification cases, namely Discrete Variational Graph Autoencoder (DVGAE-#). We tested the proposed model on the public dataset for digit classification. The experiment results showed that the proposed model can classify digit image more accurate compared to the baseline models with cross error = 0.0713 and F1-score adalah 0.9699.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Variational Autoencoder, Representasi Latent Diskrit, Graph Neural Networks, Discrete Latent Representation |
| Subjects: | Q Science > QA Mathematics Q Science > QA Mathematics > QA166 Graph theory Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) |
| Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis |
| Depositing User: | Fairuuz Nurdiaz Amaanullah |
| Date Deposited: | 12 Feb 2024 07:54 |
| Last Modified: | 08 Nov 2024 06:57 |
| URI: | http://repository.its.ac.id/id/eprint/106371 |
Actions (login required)
 |
View Item |