Track: High School STEM Poster Competition

Abstract

Recent advancements in deep learning have yielded remarkable results in various fields such as image classification, object detection, and natural language processing. This study focuses on image classification and implements and compares two different Convolutional Neural Network (CNN) models based on the AlexNet and ResNet architectures using the CIFAR-10 dataset. The motivation for this research stems from the remarkable achievements of deep learning and the significance of image classification. The CIFAR-10 dataset, containing various categories of objects and animals, is utilized as a benchmark dataset to evaluate the generalization capability of the models. The first model, based on AlexNet, exhibits a gradual increase in training accuracy with increasing epochs. However, the test accuracy plateaus after approximately 20 epochs, indicating the challenge of achieving high accuracy with a relatively simple architecture. In contrast, the second model, based on ResNet, effectively addresses the gradient vanishing problem, enabling the training of deeper neural networks. Experimental results show stable increases in both training and test accuracy, with the test accuracy maintaining a high level. The ResNet-based model achieves an accuracy of 90.69% in the final test, demonstrating superior performance compared to AlexNet. This study particularly validates the effectiveness of deep neural network architectures, including ResNet. Future research will explore methods to further enhance performance using various datasets and models, addressing the observed potential limitations.

Keywords

Privacy Security, Privacy Paradox, Large Language Model, Natural Language Processing and Data Protection,