Pengembangan Neural Flow Diffusion Model Dengan Kolmogorov-Arnold Networks Untuk Pembangkitan Citra

Wicaksono, Mochammad Aurich Ilham (2025) Pengembangan Neural Flow Diffusion Model Dengan Kolmogorov-Arnold Networks Untuk Pembangkitan Citra. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211024-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (16MB) | Request a copy

Abstract

Model diffusion dikenal sebagai model generative artificial intelligence yang mampu membangkitkan data baru dengan fidelitas tinggi, yaitu sangat menyerupai data asli. Salah satu prinsip utama dalam proses difusi adalah membangkitkan sampel data berdasarkan distribusi Gaussian. Jika input data memiliki dimensi yang tinggi, maka proses difusi berpotensi menghadapi curse of dimensionality. Oleh karena itu, penelitian Tugas Akhir ini menyusun sebuah model difussion yang diintegrasikan dengan model Kolmogorov-Arnold Networks (KAN). Secara singkat, penelitian Tugas Akhir ini mengestimasi fungsi proses difusi menggunakan KAN. Model yang diusulkan dinamakan dengan Kolmogorov-Arnold Networks Neural Flow Diffusion (KANNDiff). Pada Penelitian Tugas Akhir ini, KANNDiff diuji pada data citra MNIST dan data citra CIFAR-10. Dari hasil eksperimen diperoleh bahwa KANNDiff mampu merekonstruksi citra yang diberikan lebih baik dibandingkan dengan model baseline NFDM dengan ditunjukkan nilai metrik Bits per Dimension (BPD) dan Fr´echet Inception Distance (FID) yang lebih rendah.

======================================================================================================================================

Diffusion models are known as artificial generative intelligence models that are capable of generating high-quality data. One of the key features of the diffusion process is that the data samples are based on a Gaussian distribution. If the input data has a high dimensionality, then it is likely that the estimation calculation experiences the curse of dimensionality issue. Therefore, this Final Project Research constructs a diffusion model that is integrated with the Kolmogorov-Arnold Networks (KAN) model. In brief, this Final Project Research estimates the diffusion process function using KAN. The proposed model is called Kolmogorov-Arnold Networks Neural Flow Diffusion (KANNDiff). In this Final Project Research, KANNDiff will be tested on MNIST image data and CIFAR-10 image data. Based on the experimental results, KANNDiff successfully reconstructs the given image better than the baseline model (NFDM) by showing the lower Bits per Dimension (BPD) and Fr´echet Inception Distance (FID) metric values.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Neural Flow Diffusion Model, Kolmogorov-Arnold Networks, Pembangkitan Citra =========================================================== Neural Flow Diffusion Model, Kolmogorov-Arnold Networks, Image Generation
Subjects:	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:	Mochammad Aurich Ilham Wicaksono
Date Deposited:	01 Aug 2025 09:30
Last Modified:	01 Aug 2025 09:30
URI:	http://repository.its.ac.id/id/eprint/125735

Actions (login required)

View Item