in equation (1) the //Approximate Inference Methods// notes,
http://adv-ml-2017.wikidot.com/forum/t-2259456/in-equation-1-the-approximate-inference-methods-notes
Posts in the discussion thread "in equation (1) the //Approximate Inference Methods// notes,"Tue, 25 Jan 2022 16:47:21 +0000http://adv-ml-2017.wikidot.com/forum/t-2259456#post-2809999Re: in equation (1) the //Approximate Inference Methods// notes,
http://adv-ml-2017.wikidot.com/forum/t-2259456/in-equation-1-the-approximate-inference-methods-notes#post-2809999
Wed, 03 May 2017 20:16:50 +0000Amir Globerson2872628
It's not needed and indeed it's not more general than the pairwise formulation, since phi_i(x_i) can always be viewed as part of some phi_{ij}. I chose to include it since it's standard in presenting approximate inference algortihms, and it's nice to see the role that phi_i(x_i) (or equivalently theta_i) play in algorithms like mean field.
]]>
http://adv-ml-2017.wikidot.com/forum/t-2259456#post-2808615in equation (1) the //Approximate Inference Methods// notes,
http://adv-ml-2017.wikidot.com/forum/t-2259456/in-equation-1-the-approximate-inference-methods-notes#post-2808615
Tue, 02 May 2017 18:15:37 +0000name
In equation (1) the Approximate Inference Methods notes, Why is the pi over all phi_i(x_i) is needed? in lecture 2, we show a definition of pairwise MRF with only phi_i,j(x_i, x_j) (slide 25 of lecture 2) thanks
]]>