Deep-Learning-Advantages-of-pre-trained-models
Different data types call for different components in a deep learning pipeline. In this research, the aim is to explore the advantages of a pre-trained model compared to a model trained from scratch. Training both models on the same task for comparison, various approaches are reported for data pre- processing, parameter selection, classification and evaluation. For the image dataset, a maximum accuracy of score of 0.91 is achieved for pre-trained model and a score of 0.75 for the scratch trained model. For the natural language processing dataset, a maximum accuracy score of 0.997 is achieved for pre-trained model and a score of 0.982 for the scratch trained model. Reasoning is given behind the different choices in approaches and the resulting performances as a manner of comparing the model types.