Short notes, stray thoughts, and the occasional technical write-up.
[2] From Maximum Likelihood Estimation (MLE) to Evidence Lower Bound (ELBO): Why Exact Posteriors Are Intractable
Why training a latent-variable generative model needs a posterior that has no closed form — and how the ELBO sidesteps it by rewriting the objective in terms that are actually tractable. Second post in the Generative Models series.
[1] KL Divergence Explained
Entropy, KL divergence, and why forward vs reverse KL determines whether a model collapses onto one mode or spreads across all of them. First post in a series building up to the ELBO.
