Looper: An End-to-End ML Platform for Product Decisions

Scalable End-to-End ML Platforms: from AutoML to Self-serve

“…Another benefit of datadriven software platforms is illustrated by a common strategy [31] is to limit retention of data and trained ML models to, say, 35 days, which improves data privacy but requires infrastructure for feature lineage tracking and automation for ML model retraining. ML platforms [20,30,39] often support and automate workflows that train ML models on data to perform prediction, estimation, ranking, selection, and other ML tasks. Such applications necessitate regular data collection and retraining of ML models [45], which provides strong motivation for platforms in practice.…”

“…End-to-end ML platforms support workflows with a broader scope, including data collection and preparation as well as tracking and optimization of product-impact metrics to drive business value. While relatively new, several industry platforms [20,30] drive products that are used by billions of users and operate on powerful data sources. Such platforms automate data collection and model retraining, and also support online causal evaluation for products they enable.…”

“…End-to-end platforms are categorized as specialized and generalpurpose [30]. Specialized platforms focus on certain application categories and use cases (e.g., image [22], text [8], or speech analysis for harmful content [32]), which allows them to tailor their interfaces to application concepts and product specifics.…”