Efficiently capture training steps and model updates for machine learning
ProvQML: Efficient Provenance Capture for Machine Learning as Recursive Aggregate SQL [Extended Version]
Databases
Summary
Machine learning models learn by adjusting through many training steps, but understanding exactly how each training example affects the model is hard. The authors created ProvQML, a way to track detailed records of how individual data points influence each training step and how the model changes over time. This helps explain how machine learning models learn and improve, making their decisions more transparent. They tested their method on popular datasets and showed it works efficiently.
What this means in practice
- •For machine learning engineers: Track detailed training data contributions and model changes during iterative training in large-scale ML systems.
- •For data engineering teams: Integrate provenance tracking into SQL-based data analytics pipelines that run iterative ML algorithms.
Authors
Nkechi J. Akinwale, Seokki Lee
Abstract
Expressing ML algorithms as recursive aggregate SQL queries over distributed systems enables large-scale declarative data analytics, yet no existing system supports scalable provenance capture, necessary for ML explainability. We present ProvQML, a provenance-aware recursive aggregate SQL engine that efficiently captures fine-grained provenance while evaluating ML algorithms as recursive aggregate queries. We capture two types of provenance: (i) local provenance that records the contribution of each training instance to gradient computation at a specific model state, and (ii) global provenance that encodes the sequence of model parameter updates across the iterations. Extensive experiments on widely used ML datasets demonstrate that our approach efficiently captures both provenance for various ML models over widely used ML datasets.