SDS 433: Data Science Trends for 2021 — with Ben Taylor

Model Maintenance

Model maintenance is a critical aspect of data science, often overlooked in the rush to deploy models. Ben Taylor emphasizes the importance of continuous model deployments and prediction-level insights to ensure models remain effective over time. He highlights the challenges of feature drift, where changes in data can lead to model inaccuracies, and the need for systems to detect and address these issues proactively 1. Taylor notes that many companies only react to feature drift after encountering problems, stressing the importance of proactive measures 2.

Continuous model deployments and prediction-level insights are crucial for maintaining model effectiveness.

--- Ben Taylor

Proactive feature drift detection and continuous learning are essential for long-term model success, requiring data scientists to anticipate and address potential issues before they arise.

   

AI Production

AI productionalization involves transforming experimental AI models into robust, production-ready systems. Ben Taylor discusses the challenges of this process, including the need for continuous learning and the ability to rapidly deploy new models in response to changes like COVID-19 3. He highlights the importance of having a structured process for retraining models, as many AI projects lack this, leading to inefficiencies and lost data 3.

AI productionalization requires a structured process for retraining models to avoid inefficiencies.

--- Ben Taylor

Additionally, Taylor touches on data privacy challenges, noting that even anonymized data can sometimes be reverse-engineered to reveal personal information, underscoring the need for federated learning approaches 4.