Neural Optimal Design of Experiments for Inverse Problems

This paper introduces a novel learning-based framework, Neural Optimal Design of Experiments (NODE), for optimal experimental design in inverse problems. The key innovation is a single optimization loop that jointly trains a neural reconstruction model and optimizes continuous design variables (e.g., sensor locations) directly. This approach avoids the complexities of bilevel optimization and sparsity regularization, leading to improved reconstruction accuracy and reduced computational cost. The paper's significance lies in its potential to streamline experimental design in various applications, particularly those involving limited resources or complex measurement setups.