ELBO (Evidence Lower Bound)

The Evidence Lower Bound (ELBO) is a technique used in probabilistic models to approximate complex calculations involved in understanding data. Essentially, it provides a manageable way to estimate how well a model explains observed data by creating a lower-bound estimate. Instead of directly computing the difficult probability, ELBO focuses on optimizing a simpler function that gets closer to the true value. By maximizing this bound, models can improve their fit to data, enabling better learning and inference even when exact calculations are computationally infeasible.