The objective of this project is the design of a new machine learning pipeline for Farmy that considers the standards of MLOps. This pipeline should manage the model life cycle (training, testing, monitoring, retraining) and should consider humans in a loop strategy. The lack of annotated data about most diseases, making it challenging to come with a performant classifier in a straightforward manner. Having experts in the ML loop helps move faster and smoother to production, even if the system… Show more The objective of this project is the design of a new machine learning pipeline for Farmy that considers the standards of MLOps. This pipeline should manage the model life cycle (training, testing, monitoring, retraining) and should consider humans in a loop strategy. The lack of annotated data about most diseases, making it challenging to come with a performant classifier in a straightforward manner. Having experts in the ML loop helps move faster and smoother to production, even if the system is not perfect yet. A successful human-in-the-loop strategy should consider active learning to decrease the number of annotated examples, by choosing only the examples able to improve the model performance. Show less