Machine Learning (ML) Implementation Model

A Guide to Deployment in Industry

The implementation model serves as a guide to facilitate the implementation of Machine Learning (ML) in industry. While the model covers some technical details, the majority of its focus is on the challenges specific to actual implementations, particularly in manufacturing and operations settings. These challenges sit at the intersection of management and engineering, with skills required from both in order to put the technology into practice. Contents of the implementation model can be summarized into five stages leading up to production:

The magnitude of implementation challenges is often increased in manufacturing & operations, relative to other industries, due to the difficulty of collecting data in fast-paced, physical environments. However, for settings in which rate, volume, sensitivity, and complexity are high, ML methods can yield significant gains. Not only will this model provide a baseline understanding to those who haven’t approached these problems in practice before, it also aims to dive deeper into some of the persistent challenges of implementation. It sources from the knowledge of the MIT community of students, partners, alumni, and faculty. Recommendations are made mostly for the individual solving a problem with ML, but can also help guide an organization’s leadership to empower their teams with these tools.

Implementation Model in Action

Providing concrete guidance for ML application, the model walks through various stages of project workflow to capture nuanced considerations—from organizational planning, project scoping, data engineering, to algorithmic selection—in resolving execution challenges. Included within the model are:

Reviewed
Iterated
Through Practice

The model is continuously updated to incorporate changing requirements and new learnings. With active case studies from the MIT LGO program, ongoing face-to-face collaboration between business and technology is captured to translate theories into practice.

For additional information on the implementation model, please reach us via our Contact Form