Member-only story
Score-Based Causal Discovery with LLMs
Practical Hands-on Techniques for Integrating LLMs Into Causal Discovery Pipelines
This article is also available as a podcast! If you’re on the go or just want to absorb the content in audio format, you can listen to the full episode below 👇 The podcast is also available on Spotify and Apple Podcasts.
Causal discovery tackles them by recovering a graph where nodes are variables and edges encode causal influence. The previous article introduced constraint-based methods, which infer this graph through conditional independence tests. They are statistically elegant, but sensitive in practice: errors in early tests propagate, and performance degrades quickly with noisy or high-dimensional data.
Score-based methods take a different approach. Instead of testing independences directly, they assign a score to candidate graphs and search for the structure that best explains the data. This makes them more flexible, more robust to noise, and easier to combine with prior knowledge.
