Workflow design patterns for the machine learning pipelines:

Machine learning pipelines are complex workflows involving multiple sequential steps, such as data ingestion, preprocessing, model training, and deployment. A fundamental and widely adopted design pattern for these workflows is to represent them as a Directed Acyclic Graph (DAG). In a DAG, each node represents a task or a component of the pipeline, and the directed edges represent the dependencies and flow of execution between these tasks. The "acyclic" property ensures that the pipeline has a defined start and end and does not enter an infinite loop, making it manageable, reproducible, and scalable.

A. Aim to explain how the machine learning model works: This describes the field of Explainable AI (XAI) or model interpretability, not the structural design of the pipeline itself.
C. Seek to simplify the management of machine learning features: This is the primary goal of a feature store, which is a specific tool or component within an MLOps ecosystem, not the overall pipeline pattern.
D. Separate inputs from features: This describes the process of feature engineering, which is a specific step (or node) within a pipeline, not the overarching workflow design pattern.

1. Google Cloud. (n.d.). MLOps: Continuous delivery and automation pipelines in machine learning. "The pipeline steps are defined as a graph of components. This pipeline graph is a directed acyclic graph (DAG)..." (Section: "ML pipeline"). This is official vendor documentation describing a core MLOps concept.

2. Baylor, D., et al. (2017). TFX: A TensorFlow-Based Production-Scale Machine Learning Platform. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. The paper describes the TFX platform where "A pipeline is a directed-acyclic graph (DAG) of components." (Section 2.1, p. 1388). DOI: https://doi.org/10.1145/3097983.3098021

3. Kubeflow Documentation. (n.d.). Introduction to Kubeflow Pipelines. "Each pipeline is a specification that describes a workflow, which is a directed acyclic graph (DAG) of tasks." This is official documentation for a leading open-source MLOps platform.