Scalable inference in the N-torus using variational methods and model augmentation

Dr Alex Navarro, AstraZenica. Part of the statistics seminar series

With the exponential growth in the amount of data generated and stored, methods capable of dealing with large datasets have become increasingly important. In this talk, I address the problem of inference in large datasets of multiple angle measurements in a toroidal topology. To tackle this problem, I propose a novel approach drawing on statistical physics and probabilistic machine learning. This approach is based on the covariance function machinery developed for Gaussian Processes and performs inference by introducing a variational approximation. To illustrate the capabilities of the proposed method to cope with large datasets, I showcase experimental results in both synthetic and real-world applications and compare to other established methods.