Lars Lorch

Papers

Total Citations

H-Index

About

Lars Lorch is a researcher whose work sits at the intersection of Bayesian machine learning and trustworthy AI, with a particular focus on developing principled methods for incorporating domain knowledge into neural network models. His most cited work, "Output-Constrained Bayesian Neural Networks" (2019, 10 citations), introduces a novel framework that addresses a fundamental limitation of standard Bayesian neural networks: while priors are typically defined in parameter space, meaningful expert knowledge often exists in function space. Lorch's solution, Output-Constrained BNNs (OC-BNNs), allows practitioners to encode prior knowledge about what a model's output should or should not be in specific regions of the input space, effectively bridging the gap between statistical rigor and practical applicability. This contribution is particularly valuable for safety-critical applications where models must respect physical or logical constraints. Beyond this core work, Lorch's research continues to advance the frontier of uncertainty quantification and robust decision-making under model misspecification, making him a notable voice in the growing field of reliable machine learning systems.