# Causal Bayesian Optimization

**Authors:** Virginia Aglietti, Xiaoyu Lu, Andrei Paleyes, Javier Gonz\' alez  
**arXiv:** [2005.11741](https://arxiv.org/abs/2005.11741)  
**Date:** 2020-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2005.11741v1/

## Key points

- The paper "Causal Bayesian Optimization" addresses the problem of globally optimizing a variable of interest in a causal model where interventions can be performed.
- The authors propose a novel algorithm called Causal Bayesian Optimization (CBO), which combines ideas from causal inference, uncertainty quantification, and sequential decision making.
- CBO balances two trade-offs: exploration-exploitation and observation-intervention when combining real interventional data with estimated intervention effects computed via do-calculus.
- The approach generalizes Bayesian optimization by incorporating causal information to improve decision-making strategies and avoid suboptimal solutions.
- The practical benefits of CBO are demonstrated through experiments in synthetic settings and two real-world applications.
- The first application involves optimizing the performance of an online advertising campaign by selecting ad placements on different websites.
- The second application involves optimizing the energy consumption of a building by controlling its heating system.
- Overall, this paper presents a valuable contribution to the field of optimization under uncertainty by introducing a new algorithm that incorporates causal information into Bayesian optimization for improved decision-making strategies in complex systems with interconnected nodes.

## Summary

The paper "Causal Bayesian Optimization" addresses the problem of globally optimizing a variable of interest in a causal model where interventions can be performed. This problem is relevant in various fields, including biology, operational research, and communications, where the goal is to optimize an output metric of interconnected nodes. The authors propose a novel algorithm called Causal Bayesian Optimization (CBO), which combines ideas from causal inference, uncertainty quantification, and sequential decision making. The approach generalizes Bayesian optimization by incorporating causal information to improve decision-making strategies and avoid suboptimal solutions. CBO balances two trade-offs: exploration-exploitation and observation-intervention when combining real interventional data with estimated intervention effects computed via do-calculus. The practical benefits of CBO are demonstrated through experiments in synthetic settings and two real-world applications. The first application involves optimizing the performance of an online advertising campaign by selecting ad placements on different websites. The second application involves optimizing the energy consumption of a building by controlling its heating system. Overall, this paper presents a valuable contribution to the field of optimization under uncertainty by introducing a new algorithm that incorporates causal information into Bayesian optimization for improved decision-making strategies in complex systems with interconnected nodes.