Inference & Sampling
When a posterior cannot be written down, it must be approximated and sampled. This base covers variational inference and the ELBO, energy-based models, and Langevin dynamics / SGLD — a way to sample from an unnormalized density using only its gradient, and the inner loop of CryoWGEN.
Hundreds of walkers start uniform and, under Langevin dynamics, drift along the gradient of the log-density with added noise until they settle on the two modes. A larger step is faster but, taken too far, overshoots detail. Only the gradient is needed — no normalizing constant.
Articles in this base 3 articles
Variational inference & the ELBO
Approximating an intractable posterior by optimizing a tractable family, bounded below by the evidence lower bound.
Energy-based models
Probabilistic models that assign an unnormalized energy to every configuration, defining a density through the Boltzmann form.
Langevin dynamics & SGLD
Sampling from an unnormalized density using only its score, by following a noisy gradient ascent.