tag:blogger.com,1999:blog-8730014808678923743.post2350584671912230782..comments2019-02-09T13:28:00.420+03:00Comments on almost stochastic: The EM AlgorithmDenizhttp://www.blogger.com/profile/06043980704154563908noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-8730014808678923743.post-45672395741038030002013-05-26T12:20:33.959+03:002013-05-26T12:20:33.959+03:00Thanks, I added this as a note in the post.Thanks, I added this as a note in the post.Denizhttps://www.blogger.com/profile/06043980704154563908noreply@blogger.comtag:blogger.com,1999:blog-8730014808678923743.post-44734741805121784732013-05-26T11:56:11.507+03:002013-05-26T11:56:11.507+03:00Not only it makes sense to choose $p(\mathbf{y}|\m...Not only it makes sense to choose$p(\mathbf{y}|\mathbf{x},\theta)$as${q(\mathbf{y})}$, that is the best you can do. If you find the${q(\mathbf{y})}$that maximizes$\log \E_{q(\mathbf{y})} \left[ \frac{p(\mathbf{x}, \mathbf{y} | \theta)}{ q(\mathbf{y})} \right]$by taking derivative and setting it to zero, you end up with$p(\mathbf{y}|\mathbf{x},\theta)\$. bitkidokuhttp://www.blogger.com/profile/09141709808294868945noreply@blogger.com