Comparative Study on the Efficiency of Deep Learning Model Training in Cloud Environments: Google Colab vs AWS

Abstract

Deep learning has become a major foundation in the development of modern artificial intelligence technologies, especially in the applications of image recognition, natural language processing, and recommendation systems. However, the training process of deep learning models requires large and efficient computing resources. This study aims to evaluate the efficiency of training deep learning models on two popular cloud platforms, namely Google Colab and Amazon Web Services (AWS). The method used is a comparative experiment with a simple Convolutional Neural Network (CNN) model trained using the CIFAR-10 dataset, and Identical training hyperparameters were applied on both platforms. The results show that Google Colab demonstrates greater cost efficiency as it provides GPUs for free, while AWS provides faster training performance and slightly higher validation accuracy. This study concludes that platform selection should be tailored to the user's needs, both in terms of budget, project scale, and system stability. These findings offer preliminary guidance for selecting cloud platforms in small- to medium-scale deep learning projects.