Efficient MLOps Pipeline for Transfer Learning and Reuse of Pre-Trained ML Models
Vimal Kumar, Debjani Ghosh, Shivom Srivastava · 2023
This research paper aims to develop an MLOps pipeline for transfer learning to improve the efficiency and accuracy of machine learning models. The commencement of addressing a related undertaking in deep learning, frequently involves applying the oft-used practice of utilizing pre-established models as footing. However, setting up an efficient transfer learning pipeline can be challenging due to issues like data preprocessing, model selection, and deployment. The proposed pipeline addresses these issues by incorporating best practices in MLOps, including version control, continuous integration and deployment, and automated testing. In an effort to assess the proficiency of the pipeline, numerous experiments were conducted on diverse datasets and then juxtaposed against other established methodologies. The findings unveil that our conduit vastly enhances the efficacy and precision of the transfer learning. This inquiry holds momentous implications for the AI society, as it furnishes a detailed blueprint to construct efficient pipelines in this realm. Furthermore, it is feasible to expand our methodology effortlessly towards diverse fields of machine learning. This enables scholars and specialists alike to optimize the evolution of models successfully. On the whole, our investigation underscores the significance of incorporating MLOps methodologies into machine learning model creation and the conceivable advantages it can furnish regarding capability to grow, reproducibility and overlying performance.