NORMA eResearch @NCI Library

Knowledge Distillation of the ResNet50 Model for Ocular Diseases Analysis

Thakare, Saurabh Babarao (2022) Knowledge Distillation of the ResNet50 Model for Ocular Diseases Analysis. Masters thesis, Dublin, National College of Ireland.

[thumbnail of Master of Science]
PDF (Master of Science)
Download (1MB) | Preview
[thumbnail of Configuration manual]
PDF (Configuration manual)
Download (1MB) | Preview


Massive amounts of annotated training data contribute to deep convolutional neural network development. It is usually difficult and expensive to obtain data annotations or classification in practice. Deep classification models are prone to overfitting when training with scarce amounts of medical images. Modern deployment demands require interpretability in cases involving computational speed, memory, and complexity of algorithms where performance alone is not enough to satisfy practical needs. In order for deep neural networks to progress further, it is necessary to understand their complex models, how much computation speed and memory are consumed by models, and how models will be portable with different devices. In order to resolve this issue, this paper used the Knowledge distillation of the ResNet50 method using the ocular diseases dataset and compared and visualized the results. The study thoroughly analyzed the result of the model with knowledge distillation and without knowledge distillation of the ResNet50 model in various temperature conditions. This paper is able to create a small student CNN12 model using the knowledge distillation of the teacher ResNet50 model. At t=50, t=70, t=90, and t=100, the accuracy of CNN12 is achieved is 51.4%, 60.9%, 63.7%, and 76.5%, respectively and this accuracy compared with teacher ResNet50 model.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Knowledge Distillation (KD); ResNet50; CNN12; Ocular Diseases; data augmentation
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
R Medicine > RE Ophthalmology
Divisions: School of Computing > Master of Science in Data Analytics
Depositing User: Tamara Malone
Date Deposited: 13 Mar 2023 17:20
Last Modified: 13 Mar 2023 17:20

Actions (login required)

View Item View Item