Modern Agile for Machine Learning

I had the pleasure of presenting ‘Modern Agile for Machine Learning’ at Toronto Machine Learning Micro-Summit Series to 100+ industry Machine Learning practitioners. The talk was based off my personal experiences in applying Extreme Programming practices for enterprise ML projects at Dessa.

Abstract

The world of enterprise Machine Learning is vastly different than that of small-scale use cases and kaggle competitions. Inexperienced ML teams that are used to the latter will encounter endless problems when they attempt to implement production-grade ML models in enterprise.

Many ML teams heavily focus on model performance and fail to follow standard coding practices, ultimately sacrificing code quality in favour of quick results. While this workflow is manageable for small projects, it is not sustainable for productionalizing large-scale, collaborative ML pipelines in enterprise, as ML teams accumulate massive amounts of technical debt and high cost-of-change, which can paralyze progress and delay production efforts.

In this talk, I will introduce how machine learning teams can apply Modern Agile and Extreme Programming engineering principles, such as test-driven development and refactoring, to Machine Learning Development, in order to deliver high-quality, flexible ML solutions with low cost-of-change that will save them massive amounts of time during development and production.