AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Variational Inference: A Standard Method for Training
I like variational inference in that it's modular, but also because it's optimization based and I was familiar with that. So my work was on choosing a different divergence that than what was being used at the time to minimize an upper bound of the log marginal likelihood of the data. And yeah, having an algorithm that minimizes that is the first thing, but then a mortisation came in that deviating me a little bit from looking further into approximate inference algorithms.