Course talk:CPSC522/Learning Markov Logic Network Structure

Hi Jordon,

First of all, let me apologize. I totally mixed up the due date of critiquing with the due date of Final Draft. Let me thank you for contributing in our Wiki pages, and what a fantastic job you have done. The page is perfect in almost every aspect. Only my few pointers will be: Why there isn't the link to the both papers at top(not sure we have to do this, but from most of other pages, I think it seems like a good idea.) I would prefer a less detailed pseudo code (shorter and easier and just an overview) with a link to this more detailed and complete one. Other than these two Fan-tas-tic ! Good job!

Cheers,

Mehrdad Ghomi

Hi Mehrdad,

No worries about the mixup.

I added in the papers at the top as you suggested.

As for the pseudocode, I know some students seem to prefer having the detailed pseudocode in there, while others prefer a simpler overview as you (and Bahare) have suggested. I plan to keep the pseudocode in there but add in the plain-English descriptions as well, and put them before the pseudocode. I figure that's a best-of-both-worlds solution.

Clear skies, Jordon

Dear Jordan,

It is perfect page with understandable introductions and detailed explanations. Also, It is well structured and leading me to understand the topic step by step. Only thing I am able to point out is that would it be better if you keep figure 3 and 4 above the reference, I think it would looks better for appearance.

Best regards,

Jiahong Chen

Hi Jiahong, thank you for the feedback!

I agree with you about the placement of Figures 3 and 4, but I haven't seen instructions on how to force content to be below an image when it's in a frame. I'll keep looking, but any pointers would be welcome.

Clear skies,
Jordon

Hi Jordon Great page! Its always awesome to read your pages because of your writing style and interesting examples! Few suggestions:

1. You can explain some of the concepts with maybe a single line (example-FOL, L-BFGS, GSMN)
2. Some example of the Clause Construction Operators would be interesting

Some questions:

1. " ground terms are those with no variables" so it is a constant or a function with no variables?
2. Is there any benefit of forming cliques?
3. In the TNode construction section you have mentioned " matrix MP that is used to find the edges between the TNode". So its a Matrix of Boolean values as given in the algo.

How is 0/1 used for finding the edges between the TNode?

Hi Samprity,

I think both of your suggestions are reasonable, so I'll see what I can do. I don't think examples of the clause construction operators were given in the paper, but I'll give some thought to making one up.

As for your questions:

1. A ground term has only constants in its arguments (eg. IsInCPSC522(samprity) or SamePerson(jordon, HusbandOf(jennifer)). I'll see if I can make it more clear on the page.

2. The cliques restrict the search space (clauses can only exist if their nodes form a clique). I think that's already in there, but I may not have written it in a clear way, so I'll take another look at it.

3. It's supplied to the Markov network structure algorithm as training data. I haven't looked into it (since they don't directly discuss it in the paper), but I assumed you could treat the 0s and 1s as labels for a binary classifier. I'll take another look at how I've written it and see if I can make it more clear in the page.

Clear skies,
Jordon

Thanks for the clarifications! I do see the benefit of cliques now that you have mentioned this line.

Hi Jordan,

That's a wonderful page. I like how you explained topics.

The only thing is that it was better to explain the codes separately. For example for beam search you could explain that separately. It is difficult to go through codes to understand that parts.

Cheers,

Bahare

Hi Bahare,

I agree about not having just the pseudocode in those sections; I was running short on time and wanted to make sure there was at least some content there. I plan to add plain-English descriptions of the algorithms for the final version.

Clear skies,
Jordon