Course talk:CPSC522/Support Vector Machines

From UBC Wiki

Contents

Thread titleRepliesLast modified
Critique 3206:48, 12 February 2018
Critique 121:37, 7 February 2018
Critique 2121:34, 7 February 2018

Critique 3

Scheme[wikitext]

  • The topic is relevant for the course. 4
  • 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. 5
  • There was appropriate use of (pseudo-) code. 1
  • It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic). 4
  • It is correct. 5
  • 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). 5
  • It links to appropriate other pages in the wiki. 5
  • The references and links to external pages are well chosen. 5
  • I would recommend this page to someone who wanted to find out about the topic. 5
  • This page should be highlighted as an exemplary page for others to emulate. 4
  • If I was grading it out of 20, I would give it: 17
AINAZHAJIMORADLOU (talk)07:49, 7 February 2018

Thanks for the feedback. Do you have any specific ideas on how to improve the article?

AlistairWick (talk)21:38, 7 February 2018

Comments[wikitext]

Overall, the page is written clearly and it's easy to understand. I don't see any flaws but I have some comments that might be helpful to you.

  • I'm not a native speaker but I think it's better to say that 'you can see some results in figure x' or 'figure x shows some of the obtained results for the problem' instead of saying "in the image on the right" or using words like "below" or "above" to reference a figure or image. Thus, your figures will have labels and images can easily be found.
  • Another thing that comes to my mind is that if you're using a picture from another resource, you should add the reference to that image. The results showing margins for SVM seems familiar to me. I'm not sure about this so don't get me wrong.
  • It's not very important but I think it's better to use the link on the whole word not just part of it. In line 9 of section linear regression perspective: using "NP-hard" in the link rather than NP-"hard".
AINAZHAJIMORADLOU (talk)06:45, 12 February 2018
 
 

SVM:

Put implementations of SVM Resize pictures (instead of thumbnail do framed) they are hard to see “Builds on” section paragraph 1 … I don’t know what “(read: not curved, unbroken)”

Builds on section is lengthy. Consider moving some of the information to the content section. Such as the “perception” description for example.

Additionally your mention of “multi-class problems, regression, and to non-linear splits” etc in the abstract could then be moved to the builds on sections.

Last sentence first paragraph of builds on I would suggest rewording. “the problem, then, is of where to place this hyperplane for a given data set.”

Overview first paragraph, put "support vectors" in bold

  • The topic is relevant for the course. 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). 5
  • The formalism (definitions, mathematics) was well chosen to make the page easier to understand. 4 (make definitions more obvious. Is would be nice if they sick out more. Put them in bold and introduce them.)
  • The abstract is a concise and clear summary. 4 (see previous comments)
  • There were appropriate (original) examples that helped make the topic clear. 5
  • There was appropriate use of (pseudo-) code. 0 (add some (: )
  • It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic). 5
  • It is correct. 5
  • It was neither too short nor too long for the topic. 4 (I think I could be expanded well from what you have in your to add section)
  • It was an appropriate unit for a page (it shouldn't be split into different topics or merged with another page). 5
  • It links to appropriate other pages in the wiki. 4 (could add a few more)
  • The references and links to external pages are well chosen. 5
  • I would recommend this page to someone who wanted to find out about the topic. 5
  • This page should be highlighted as an exemplary page for others to emulate. 5

If I was grading it out of 20, I would give it: 17

VanessaPutnam (talk)19:55, 4 February 2018

I'll fix up some of the language in the article, and bold the definitions. Agree the builds on section could be altered, I think I'll move a lot of that down to the content section as another reviewer suggested.

Any thoughts on what type of code you'd like to see? The diagrams are generated in Matlab, but I'm not sure if that's suitable.

AlistairWick (talk)21:37, 7 February 2018
 

Critique 2

Comments[wikitext]

  1. Overall, a good high-level description of what the algorithm does; comparisons with other methods like least-squares regression and Perceptron are helpful and the article is generally easy to follow. It would be nicer if there were more explanations on how the SVM works, by addressing the following questions:
    1. How do you select the "support vectors"?
    2. How do you actually locate the hyperplane? Is there some kind of iterative method you have to apply?
    3. How do you guarantee convergence?
  2. There is a brief description of using the "kernel trick" to extend the application of SVM to non-linear tasks (combining multiple SVMs), but it should be elaborated further as what the "kernel trick" actually does is still unclear. i.e. how do you combine multiple SVMs?
  3. Second and Third paragraph of "Builds on" section could be moved to the "Motivation" section. Mentioning the limitations of least-squares approach and Perceptron would motivate the concept of the SVM. The first paragraph about binary classifier may be enough for the "Builds on" section.

Minor Comment[wikitext]

The diagram should be resized to fit the width of the screen; on 1366x768 resolution the last graph (soft-margin SVM) is being clipped

Scheme[wikitext]

  • The topic is relevant for the course. 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). 5
  • The formalism (definitions, mathematics) was well chosen to make the page easier to understand. 2
  • The abstract is a concise and clear summary. 4
  • There were appropriate (original) examples that helped make the topic clear. 4
  • There was appropriate use of (pseudo-) code. 1
  • It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic). 3
  • It is correct. 5
  • 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). 5
  • It links to appropriate other pages in the wiki. 3
  • The references and links to external pages are well chosen. 3
  • I would recommend this page to someone who wanted to find out about the topic. 5
  • This page should be highlighted as an exemplary page for others to emulate. 4
  • If I was grading it out of 20, I would give it: 16
KumseokJung (talk)22:54, 5 February 2018

Thanks for the tips! I've redone the images since you wrote this, and I'm working on expanding the explanations. I think you're right in that it's worth mentioning how a solution is reached (fairly generic minimization/optimization routines); I'll talk about some of the methods typically applied. I for sure need to cover the kernel trick, it's one of the things that makes modern SVMs so powerful.

AlistairWick (talk)21:34, 7 February 2018