HEP@VUB Colloquium: Karen Terveer on Information Field Theory

Reconstructing cosmic-ray radio data with Information Field Theory

Information Field Theory (IFT) is a Bayesian framework for high-dimensional signal inference which integrates physics-informed priors, such as spatial correlations and physical constraints, directly into the reconstruction of continuous signals. Through variational inference, IFT approximates high-dimensional, potentially non-Gaussian posteriors to provide rigorous uncertainty quantification alongside point estimates. In this talk, I will present the application of IFT to the reconstruction of radio signals from extensive air showers. While traditional simulation-matching methods currently set the benchmark for reconstruction precision, they are often computationally intensive and typically provide only point estimates. Conversely, data-driven machine-learning approaches offer remarkable inference speed but may require extensive training datasets and can be challenging to interpret physically. IFT offers a complementary approach: it remains inherently physics-informed through its priors and requires no pre-training, yet it uses many of the efficient high-dimensional optimization techniques found in modern ML. The presented method for air shower reconstruction utilizes signal timing and fluence simultaneously, achieving a precision of 25.4g/cm^2 in X_max and 12% in radiation energy. Additionally, it offers a three-order-of-magnitude speedup over standard techniques and benchmarks the first simultaneous reconstruction of timing and fluence data for LOFAR.