causaLens logocausaLens logo alternate

The Optimization Approach to Score-based Causal Discovery

Recent breakthroughs in science and technology mean that AI systems can now autonomously discover causes and effects using “causal discovery algorithms”. This tech report introduces “score-based” algorithms, which treat causal discovery as an optimization problem.

A key type of causal discovery algorithm, “score-based algorithms”, treat causal discovery as an optimization problem by directly searching over the space of possible graphs to find those that best explain the observable data. We set out the mathematics of score-based causal discovery and outline recent developments in the topic, including the use of deep learning systems to facilitate the discovery process. This report is written primarily for a technical audience.