Robust Experimental Design via Generalised Bayesian Inference

The recent publication titled 'Robust Experimental Design via Generalised Bayesian Inference' presents a novel framework known as Generalised Bayesian Optimal Experimental Design (GBOED). This framework builds on the principles of Bayesian inference, which is traditionally used to quantify information gain in experimental designs. However, conventional Bayesian methods can falter when the underlying statistical models are inaccurately specified. GBOED addresses this issue by employing Generalised Bayesian inference, which substitutes the likelihood in Bayesian updates with a loss function, thereby enhancing robustness. The study introduces a new acquisition function, the Gibbs expected information gain (Gibbs EIG), derived from an extended information-theoretic framework. Empirical results indicate that GBOED significantly improves resilience to outliers and incorrect assumptions about outcome noise distributions, making it a valuable tool for researchers aiming to conduct more reliabl…