Session Outline
In the lifecycle of developing machine learning models, continuous monitoring and retraining are key to ensuring models perform well in real-world settings. In this talk at the NDSML Summit 2023, speakers discuss how different decisions affect model performance and resource use when continuously training Language Models in production. Speakers also introduce a framework for designing a successful retraining strategy, which helps address challenges like data changes (data drift) and adapting to new data as it becomes available.
Key Takeaways
- A framework of decision points to develop a model retraining process.
- Strategies for including newly acquired data for model retraining and evaluation.
- Empirical evidence on model performance using different retraining strategies.
