*Result*: Structured pruning utilizing DepGraph with L2 norm to reduce the number of CNN parameters.

*CNN architectures have been shown to perform well in several tasks, including image classification, natural language processing, and speech recognition. The architecture's performance is directly proportional to the model's size, which presents a significant challenge in its practical application. This research proposed structured pruning on a Convolutional Neural Network (CNN) using a Dependency Graph (DepGraph) with an L2 norm pruning criterion. This approach can reduce the number of parameters so that the CNN architecture is lighter. Additionally, it was intended to enhance the model's processing speed, specifically in terms of the rate at which images can be processed. To prove that DepGraph with L2 norm can effectively competently reduce the huge amount of parameters in CNN architecture, this study uses three commonly used CNN architectures: EfficientNetV2-S, MobileNetV3-S, and ResNet-18. Each pruned architecture is trained to build a classification model. The models' performance is evaluated by accuracy, precision, recall, and F1-score. The experimental results revealed that structured pruning using DepGraph with an L2 norm pruning criterion effectively reduces the number of parameters in these models. The DepGraph with an L2 norm pruning criterion decreases the number of parameters EfficientNetV2-S, MobileNetV3-S, and ResNet-18 architectures by 74.36%, 73.40%, and 74.96%, respectively. Nevertheless, the performance of each model is maintained with no significant decrease in accuracy. [ABSTRACT FROM AUTHOR]*