Variational Free Encoders in Machine Learning

The objective function that underwrites variational order encoders in deep learning is exactly the same objective function in dynamic causal modeling. It rests upon the Variational approach in which you know the functional forms of your your posters or your beliefs about the unknowns and as soon as you impose a fixed functional form on these densities you create this approximation to the marginal likelihood, known as an evidence lower bound or an elbow ELBO. The winning model is a model that has the highest evidence or the highest elbow but there's even a semantics associated with this because we're talking about log probabilities so if one model is 20 times more likely than another model that means its log evidence will be three nuts