Course talk:CPSC522/Graphical Models

Hi Yu Yan,

Nice work, a very detailed one! Here are my scores and more details below that. please let me know if you need any further clarification.
Scale of 1 to 5, where 1 = strongly disagree and 5 = strongly agree:

[5] The topic is relevant for the course.
[3] The writing is clear and the English is good.
[5] The page is written at an appropriate level for CPSC 522 students (where the students have diverse backgrounds).
[4] The formalism (definitions, mathematics) was well chosen to make the page easier to understand.
[3] The abstract is a concise and clear summary.
[3.5] There were appropriate (original) examples that helped make the topic clear.
[4] There was appropriate use of (pseudo-) code.
[5] It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic).
[4.5] It is correct.
[3] It was neither too short nor too long for the topic.
[3] It was an appropriate unit for a page (it shouldn't be split into different topics or merged with another page).
[4] It links to appropriate other pages in the wiki.
[5] The references and links to external pages are well chosen.
[4] I would recommend this page to someone who wanted to find out about the topic.
[3.5] This page should be highlighted as an exemplary page for others to emulate.

Comments:

First of all, I agree with the most comments that Jordon already posted. Some of the pointed issues have been rectified at the time that I am writing this, however, I would like to point out and emphasize few matters:

Format:
It would be great if you apply a consistency of formatting different sections through your page, as it would make reading much easier

• Centralizing mathematical formulas
• Where is figure 4? it seems that you skipped that or you are editing now.
• Emphasizing new terms by making them Bold (already resolved)
• Linking to other pages (Already resolved)
• "Builds on" links (Already resolved)

Language:

It would be a good Idea if you read your page one more time and try to fix some grammatical issues. I would recommend you to install Ginger plugin for Google Chrome (A free proof reading plugin)

Abstract:

I found your abstract a little concise, you can definitely explain the flow of the materials that are presented in a page by briefly mentioning section headings and connecting them together.

Undirected Graphical Models:

I can’t find the g_j in the equation? do you mean f_j ? in the following part: “,g_j are factors"

Inference in graphical models:

I believe that this section should be taken out and is represented as an independent page in Wiki, as the complete explanation would be out of this page’s scope. A brief explanation with a pointer to a dedicated page would let the author to more specifically investigate Inferences in graphical models.

Factor graph and propagation algorithm:

in the Pseudo code section it would be good if you briefly explain the the variables at the beginning as what Murphy and Poole did in their books.

Learning in Graphical Models:

I guess you already rectified the matters pointed out by Jordon as I believe it is now in a perfect size.

Good job,

Yaashaar Hadadian

Hi Yaashaar Hadadian,

Thanks for your feedback. It is very useful for me to make a final draft. I will modify my draft again according to your suggestions. Thanks again.

Bests, YuYan

A solid draft! Here are my scores, with general and section-specific comments below that. Let me know if any clarification or discussion is needed.

Scale of 1 to 5, where 1 = strongly disagree and 5 = strongly agree:

---

• (5) The topic is relevant for the course.
• (3.5) The writing is clear and the English is good.
• (5) The page is written at an appropriate level for CPSC 522 students (where the students have diverse backgrounds).
• (4) The formalism (definitions, mathematics) was well chosen to make the page easier to understand.
• (4) The abstract is a concise and clear summary.
• (4) There were appropriate (original) examples that helped make the topic clear.
• (3) There was appropriate use of (pseudo-) code.
• (4) It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic).
• (4.5) It is correct.
• (4) It was neither too short nor too long for the topic.
• (4) It was an appropriate unit for a page (it shouldn't be split into different topics or merged with another page).
• (2) It links to appropriate other pages in the wiki.
• (2.5) The references and links to external pages are well chosen.
• (4) I would recommend this page to someone who wanted to find out about the topic.
• (3.5) This page should be highlighted as an exemplary page for others to emulate.

Comments:

---

• A proofreading pass for grammar issues is recommended.
• It might be useful to highlight significant terms using bold or italics.
• The examples and figures were useful, but not always original. I note that Figure 5 is credited to Murphy, which is good. If you made the other figures from scratch, then well done on professional-looking figure generation; but it might be useful to make the figures more consistent with each other (don't worry, it's NOT a high priority).

Abstract:

• The abstract is a bit too concise. I'd extend it to 2-3 sentences and include things like the names of the two main representations.
• needs "More general than" section that links to Bayesian network, Markov network, and HMM pages

Representation of Graphical Models:

• might be useful to have some links to the Bayesian network and Markov network pages here as well, but not strictly necessary since you'd have them in the "More general than" section

Directed Graphical Models:

• There are pages for both Bayesian networks and HMMs. It seems appropriate that you have a short summary of significant aspects for each, and so I don't think there's significant content duplication here; but there should definitely be links to the pages here for user convenience.

Undirected Graphical Models

• There's a bit more detail here than for the DGM section, and so there might be some concerns about content overlap.
• I don't see a definition for the g_i in the equation for P(X); since they aren't conditional probabilities, it would be worth mentioning what they are. Not much detail is required, since that's covered in the Markov networks page; simply mentioning that the g_i are factors over cliques and linking to the relevant part of the Markov networks page should be sufficient.
• Markov blankets are applied in reference to Bayesian networks as well as Markov networks, and so that part might be better placed in your section about conditional independence. Also, I believe Markov blankets are already minimal.
• The Markov blankets as listed in Figure 4 are a bit confusing. The Markov blanket of A is C', the Markov blanket of B and C is {C',C"}, and the Markov blanket of C" is {B, C', C}. Murphy mentions Markov blankets in the Chapter 19 pdf in your bibliography for a bit of clarification.

Inference in graphical models:

• Parts of this section seem to be in reference specifically to Bayesian networks; and so some of this content might be better off on the Bayesian networks page For example, the material related to Figure 5 might be better suited there.
• A definition of the Markov blanket might be useful in the conditional independence section, as mentioned previously.
• The formulas related to P(R|W) uses a mix of T's and 1's as truth values. I'd recommend sticking with T's for consistency with the figure.

Factor graph and propagation algorithm:

• Since it can be applied to both DGM and UGM, this is a useful section to have in this page.
• "A factor graph is a bipartite graph representing the factorization of a function." Any function? What kind of function?
• When you define x_s_j, you have i in is_j; do you mean i in S_j?
• What is the junction tree algorithm? What about message passing algorithms? External links would be useful if you don't want to write them up here.
• I'm finding the pseudocode difficult to read, though the comments help a bit. It would help to be more clear about what c, k, C, Mc, Dc, and xc represent.

Learning in Graphical Models:

• Part of your first paragraph contains information that might be better placed in your introduction to the nature of graphical models.
• Need more detailed descriptions of the components and notation for the definitions of D and L.
• External links would be useful here for things like MLEs and the EM algorithm.

Structure Learning:

• This section seems less detailed and more rushed than the rest of the document; terms are given mathematical symbols that are not really used; and a number of terms (d-separation, Likelihood Score, etc.) would benefit greatly from external links.

References look fine.

Hi JordonJohnson,

Thanks for your valuable feedback. Lack of hight light and grammar error issues are very important points that I will modified them immediately. And from reading your wiki page about Markov Network, I found the overlap content and I will make my Undirected Graphical Models section be more general and try to find something new that more related to graphical models. I will also add contents to Inference in graphical models and explanations for the pseudo code as you suggest. Actually, structure learning part is a little difficult for me and I still learn it. So it seems less detailed. I will add contents in this part later. Thanks again for your feedback. It really do lots of help for me. I will let you know when I am done with that.

YuYan

Hi YuYan,

This is a great draft, and here are some suggestions:

1. I think part of the information in the Undirected Graphical Models section is covered by the http://wiki.ubc.ca/Course:CPSC522/Markov_Networks#Probabilities_and_Factors , I think its the same, so maybe you can add an external link.

2. Summing out variables part you can add a link to http://wiki.ubc.ca/Course:CPSC522/Variable_Elimination , I think this will be easier for people to understand.

3. Factor graph and propagation algorithm part, I think you need to rewrite the pseudo code, it's a little hard to read. Here is an example I think it will be helpful: https://en.wikipedia.org/wiki/Pseudocode

4. As for reference, maybe add some notation to tell people where you use the material.

Sincerely,

Junyuan Zheng

Hi Junyuan Zheng,

Thanks for your feedback. for 1. I will check the Markov_Network wiki page and try to make this part be more different with it. Also external link is a good idea.

2. I will add the external link.

3. I will add more explanation to my pseudo code. I will let you know when I am done with that. Thanks again.

YuYan

Hi Yu Yan,

Great job. I appreciate your work on this topic although there are still many contents I do not understand. I have a suggesion for you. In this page you use many mathematical equations and symbols which make it hard for people without any background knowledge to understand the ideas and concepts. I hope you can use more plain statements and fewer mathematical equations to make this page less technical and academic.

Sincerely,

Ke Dai

Hi, Ke Dai, thanks for your feedback. I will try to add more plain statements to make these mathematical equations easier to understand. I will let you know when I am done with that.

YuYan