Pamela Perez (GAMA-1) – Presenting: “An MLOps approach to address the complexities of delivering a ML/AI product.
Rapid advances in Machine Learning (ML) offer great promise, and the goal of any R2O organization is to bridge the gap between these activities and their application.
ML offers unique challenges in this endeavor. Typically research focuses on the development of models; however, these models play a small part in the production of ML solutions. Operationalizing ML products includes a multitude of additional activities from data collection and cleaning, feature engineering and hyperparameter tuning, all of the way to deployment and monitoring.
MLOps is a relatively new discipline aimed at addressing these complexities. The objective of this presentation is to introduce an MLOps approach for orchestrating the many components of ML based products to provide more robust, accurate, and rapidly available ML applications. It highlights elements of an MLOps offering that are distinct from a purely DevOps approach, such as ML model experimentation which facilitates the comparison and versioning of hundreds of models as well as regulation of the data. At the other end of the pipeline is monitoring model performance, tracking accuracy, and identifying bias and/or drift. Some of the tools the team is using from Amazon Web Services are Amazon SageMaker Studio, Amazon AgeMaker Pipelines, Amazon Jupyter Notebook to develop the methodology to address the complexities of delivering a ML/AI product.
Click the image below to enlarge
or download the PDF of the poster.