Jakob Zeitler on Bayesian Optimisation for Materials and Brownies | FAI CDT

In this interview, Jakob Zeitler, a recent PhD graduate from UCL, discusses his research journey in causal inference and Bayesian optimization. He explains the fundamental questions of causal inference, such as determining whether A causes B and quantifying the effect, and highlights the importance of assumptions that are hard to validate empirically, leading to sensitivity analysis. He then describes his shift to Bayesian optimization, motivated by the desire to work in controlled environments where assumptions are more easily satisfied. Using the analogy of baking brownies, he illustrates how Bayesian optimization efficiently explores high-dimensional spaces to find optimal recipes, with applications in materials science and drug discovery. He shares a concrete example from industry where Bayesian optimization recommended experiments that contradicted chemists’ expectations but yielded surprising results. Zeitler also discusses his internship at Spotify, where he worked on synthetic control methods, leading to a paper and two patents, including a novel language-learning feature. He mentions winning the G-Research thesis competition and his current postdoctoral position at Oxford in the SmartBiome project. Throughout, he emphasizes the importance of choosing relevant and interesting research topics, collaborating with great people, and maintaining a balance between scientific rigor and practical impact.