Daily Archives: December 8, 2008

When training generative models, we usually optimize parameters to optimize joint likelihood.  However, this is in no way (or is it? let me know if you know better) a guarantee that you’ll do better on many real-world benchmarks such as … Continue reading →