Course talk:CPSC522/Identity Uncertainty in a restaurant data-set

Hi Bahare,

This page looks good! Here are three suggestions: 1. For section 'Structure learning for the X = Y hypothesis' part, it is not clear to me, would you mind to explain more about how the Bayesian network works? 2. What does 'sloppyX and sloppyY' means, I cannot understand 'record X and record Y being sloppy respectively.' 3. Would you mind to write something about the 'Recursive Conditioning.'? The simplest example will be helpful.

Sincerely,

Junyuan Zheng

Hi Junyuan,

Thanks for your feedback!

1. I did not get your question. What do you mean by "how it works?"

2. I will edit it to make it more understandable.

3. I have a section about recursive conditioning and we had it in the slides before. Are you looking for an example or something special?

Cheers,

Hi Bahare,

Sorry for this late reply, after your presentation, I understand this now.

Hi Bahare,

Interesting work! Here is my feedback:

General

• A quick proofread would be good, but the overall readability is fine.

Abstract

• "solves" is a strong term. Does your work actually solve the problem, or does it perform better than some baseline?

Hypothesis

• The hypothesis, as stated, seems a bit general to me. Are you testing whether considering the dependencies helps in EVERY situation?

Modeling the problem

• You state that the prior ratio is given; what would be the source?

Structure learning for X=Y

• Could you elaborate on how we can infer P(descX and descY|X=Y) using that network? For example, is nameX multinomial with the correct spelling as well as every possible misspelling? And if so, is there a conditional probability table with entries for every possible spelling given there being an error or not?

Discussion

• I don't believe cross-validation requires learning parameters by hand; did I misunderstand your statement?

Hi Jordan,

Thanks for your feedback. I learned so many things from you.

I will consider them in finalizing my page.

Hi Jordan,

Again thanks for your helpful critique.

General

A quick proofread would be good, but the overall readability is fine.

I proofread the page. I hope that helps.

Abstract

> "solves" is a strong term. Does your work actually solve the problem, or does it perform better than some baseline?

I did not know this fact the “solve” is a “strong term”. I changed it to “helps to solve”. I don’t know if it is enough or not.

Hypothesis

The hypothesis, as stated, seems a bit general to me. Are you testing whether considering the dependencies helps in EVERY situation? I changed this part too “helps to solve in a restaurant data-set”

Modeling the problem

You state that the prior ratio is given; what would be the source? I add a part to state that it is calculated by the train-set or using prior knowledge.

Structure learning for X=Y

Could you elaborate on how we can infer P(descX and descY|X=Y) using that network? For example, is nameX multinomial with the correct spelling as well as every possible misspelling? And if so, is there a conditional probability table with entries for every possible spelling given there being an error or not? I tried to add this part. I thought that causes more confusing. So let’s think of that as a black box.

Discussion

I don't believe cross-validation requires learning parameters by hand; did I misunderstand your statement?

I changed this part. In this case, I’m calculating parameters by hand. That’s why I was not able to do that multiple times.

Hi Bahare,

Nice page! I think it is a good page that provides sufficient figures pseu-do codes, and you also designed a quite good experiment to exam your hypothesis. Simple suggestion is that it might be better to have more external links in your abstract. And it might be better to have a more detailed evaluation.

Best regards,

Jiahong Chen

Hi Jiahong,

Thanks for your suggestions.

I will try to add more links before finalizing.

Cheers,

Bahare

Hi Bahare,

Nice Page. Here are some suggestions:

1. Add some explanation for pseudo maybe better to understand.
2. Do some proof reading.

Bests

Yu Yan

Hi Yu,

Thanks for your feedback.

We already had recursive conditioning in the course content. So I thought we don't need that in the page.

Cheers,