Inference of Structure Using LBP

An open question in cell biology is the molecular structure inside cell membranes. Due to recent advances in cryogenic electron tomography, it is possible to see these membranes in their native structure with very low quality. In this work we explore the problem of inferring structure from a noisy observation using graphical models and loopy belief propagation (LBP). The solution is rotation and translation but not scale invariant. It is also tolerant to noise and outliers. We also explore three avenues of improving convergence for difficult datasets: generalized belief propagation, sequential LBP, and the use of larger factors.