Break the Black Box; Human Learning from Machine Learning – Emil Pedersen, Spotify

Session Outline

The old question: Correlation or causation? In a world of exploding amounts of data, how can you keep not only training machines, but extract key insights and make better decisions as a human? Machine learning does not provide all the answers, but it enables a new set of tools like a Swiss army knife. In this session, we will go through some of them, and how they can provide insights and value.

Key Takeaways:

• Visualize hypotheses with graphs (DAGs)
• Interpret models to find tipping points and relationships (partial dependency plots with SHAP values)
• Mimic randomized controlled trials / AB-tests to quantify impact and value (Causal analysis)