A summary of How to implement SVMs

John Platt; DOI

For the summary of the paper, go to the Summary section of this article.

Table of Contents

• A summary of How to implement SVMs
  • Table of Contents
  • First Pass
    • Category
    • Context
    • Contributions
  • Second Pass
    • Background Work
    • Motivation
    • Figures, Diagrams, Illustrations, and Graphs
    • Clarity
    • Relevant Work
    • Author Assumptions
      • Correctness
    • Future Directions
  • Summary
  • Summarization Technique
  • Citations

First Pass

Discussion about the title, abstract, introduction, section and sub-section headings, and conclusion

The essay How to implement SVMs by John Platt (as part of the larger Support vector machine collection of essays in the July/ August edition of the 1998 IEEE Intelligent Systems magazine) [1] discusses how to implement a Support Vector Machine (SVM). This essay goes into great detail on implementation strategies for handling larger data sets, as well as methods for training SVMs. Topics include understanding the Quadratic Problem (what SVMs aim to solve), sequential minimal optimization (reaching a global minimal value), and where to find SVM implementations.

Category

What type of paper is this work?

This essay seems to be a tutorial/ workshop paper about SVMs.

Context

What other types of papers is the work related to?

I would expect papers that are about implementing SVMs from scratch would be related to this essay.

Contributions

What are the author’s main contributions?

Their main contributions are an understanding of how SVMs work as well as how to implement them efficiently.

Second Pass

Background Work

What has been done prior to this paper?

Work has already been done on experimenting optimal SVM algorithms and minimization functions.

Motivation

Why should we care about this paper?

We should care about this paper as it provides an understanding of what a SVM is and how they function.

Figures, Diagrams, Illustrations, and Graphs

Are the axes properly labeled? Are results shown with error bars, so that conclusions are statistically significant?

The figure and charts have proper labels and captions that explain what they are representing.

Clarity

Is the paper well written?

For the most part, yes. However, the essay expects the reader to be knowledgeable about SVMs prior to reading the essay. This is shown mostly through the usage of mathematical notation specific to the problem domain, and linking to other work to explain it. While this is a short essay for a magazine, a brief sentence or two about the notation would have been appreciated.

Relevant Work

Mark relevant work for review

The following relevant work can be found in the Citations section of this article.

• A Tutorial on Support Vector Machines for Pattern Recognition [2]

Author Assumptions

What assumptions does the author(s) make? Are they justified assumptions?

The author assumes that the reader, should they implement their own SVM algorithm, will be using a commercial numerical analysis package.

Correctness

Do the assumptions seem valid?

Without understanding the nature of the numerical analysis packages of 1998, I would assume that this assumption is correct. I base this on that the author mentions that free numerical analysis packages (not if they were open sourced or not) run slower than commercial packages and may have errors due to precision mistakes.

Future Directions

My own proposed future directions for the work

I’m not interested in creating my own SVM algorithm. However, having a better understanding of how SVMs work as well as the different minimization functions that they implement, would be nice to know.

Summary

A summary of the paper

The essay How to implement SVMs by John Platt (as part of the larger Support vector machine collection of essays in the July/ August edition of the 1998 IEEE Intelligent Systems magazine) [1] discusses how to implement a Support Vector Machine (SVM). The author goes into detail about what an SVM is trying to accomplish (minimize a quadratic problem on a high dimensional matrix), what techniques exist to solve this problem, as well as available programs to allow for researchers to utilize SVMs in their work.

Overall, the essay does a good job of explaining the problem space as well as implementation details, however, the essay is very much a product of its time. There is less of a need to develop new SVM algorithms as there are many that are provided off of the shelf in free and open source numerical analysis packages [3] [4]. Additionally, the suggestion that readers should purchase a numerical analysis package to create their own SVM is dated in my opinion, as again, there are many free options available [5].

Summarization Technique

This paper was summarized using a modified technique proposed by S. Keshav in his work How to Read a Paper [0].

Citations