jmelesky
/
pdorg_site


			
							123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216
							<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html> <head>
<title>Support Vector Machines</title>

<style>

.slide {
  border: 2px solid #000066;
  background-color: #CCCCFF;
  position: absolute;
  width: 95%;
  height: 95%;
}


</style>

<script>
var slides = new Array();
var currentSlide = 0;

function initializeSlideshow() {
  var divs = document.getElementsByTagName("div");
  for (i=0; i < divs.length; i++) {
    var div = divs[i];
    if (div.className && (div.className == "slide")) {
      div.style.display = "none";
      div.addEventListener("click", showSlide, false);
      slides.push(div);
    }
  }

  showSlide();
}


function showSlide() {
  for (var i in slides) {
    slides[i].style.display = "none";
  }
  slides[currentSlide].style.display = "inline";
  currentSlide++;
  if (currentSlide >= slides.length) {
    currentSlide = 0;
  }
}
</script>
</head>

<body onload="initializeSlideshow();">


<div class='slide'>
<h1>Support Vector Machines</h1>
<p>(lightning talk)</p>
<p>(LPW '07) (john melesky)</p>
</div>

<div class='slide'>
<h1>Presupposing:</h1>
<p><ul>
<li>You have a bunch of something.</li>
<li>You can transform relevant attributes of those things into numbers.</li>
<li>You can connect those numbers into vectors (think coordinates in an attribute space).</li>
<li>You want to categorise them base on those numbers.</li>
</ul></p>
</div>

<div class='slide'>
<h1>The problem: find a line that separates these two categories of thing</h1>
<img style="float: right;" src="basedata.png" />
</div>

<div class='slide'>
<h1>For humans, this is easy.</h1>
<img style="float: right;" src="cleansep.png" />
<p>For mathematicians, it's actually not too hard.</p>
</div>

<div class='slide'>
<h1>For humans, this is easy.</h1>
<img style="float: right;" src="cleansep.png" />
<p>For <del>mathematicians</del> computers, it's actually not too hard.</p>
</div>

<div class='slide'>
<h1>There are two problems, though.</h1>
</div>

<div class='slide'>
<h1>Problem, the first:</h1>
<img style="float: right;" src="badline1.png" />
</div>

<div class='slide'>
<h1>Problem, the first:</h1>
<img style="float: right;" src="badline2.png" />
</div>

<div class='slide'>
<h1>Problem, the second:</h1>
<img style="float: right;" src="badset1.png" />
</div>

<div class='slide'>
<h1>Problem, the second:</h1>
<img style="float: right;" src="badset2.png" />
</div>

<div class='slide'>
<h1>Problem, the second:</h1>
<img style="float: right;" src="badset3.png" />
</div>

<div class='slide'>
<h1>Conveniently, Support Vector Machines address both of the problems i've identified.</h1>
<p></p>
</div>

<div class='slide'>
<h1>Solution, the first:</h1>
<p></p>
</div>

<div class='slide'>
<h1>Solution, the first:</h1>
<img style="float: right;" src="bordervectors.png" />
<p><ul>
<li>Create "border" vectors, parallel to eachother, touching the outermost edge of each category dataset.</li>
</ul></p>
</div>

<div class='slide'>
<h1>Solution, the first:</h1>
<img style="float: right;" src="bordervectors.png" />
<p><ul>
<li>Create "border" vectors, parallel to eachother, touching the outermost edge of each category dataset.</li>
<li>As you add new items, ensure these "borders" stay parallel.</li>
</ul></p>
</div>

<div class='slide'>
<h1>Solution, the first:</h1>
<img style="float: right;" src="supportvectors.png" />
<p><ul>
<li>Create "border" vectors, parallel to eachother, touching the outermost edge of each category dataset.</li>
<li>As you add new items, ensure these "borders" stay parallel.</li>
<li>Create your categorizing vector equidistant from your two "borders".</li>
</ul></p>
</div>

<div class='slide'>
<h1>Solution, the first:</h1>
<img style="float: right;" src="supportvectors.png" />
<p><ul>
<li>Create "border" vectors, parallel to eachother, touching the outermost edge of each category dataset.</li>
<li>As you add new items, ensure these "borders" stay parallel.</li>
<li>Create your categorizing vector equidistant from your two "borders".</li>
<li>These "borders" are called "support vectors".</li>
</ul></p>
</div>

<div class='slide'>
<h1>A joke:</h1>
<p>Q: How many mathematicians does it take to change a lightbulb?</p>
<p></p>
</div>

<div class='slide'>
<h1>A joke:</h1>
<p>Q: How many mathematicians does it take to change a lightbulb?</p>
<p>A: One, who hands it to 127 Londoners, thus reducing it to an earlier joke.</p>
<p></p>
</div>

<div class='slide'>
<h1>A question:</h1>
<p>Q: How do mathematicians categorize non-linearly-separable data?</p>
<p></p>
</div>

<div class='slide'>
<h1>A question:</h1>
<p>Q: How do mathematicians categorize non-linearly-separable data?</p>
<p>A: Munge the data until it's linearly separable, thus reducing it to an earlier slide.</p>
<p></p>
</div>

<div class='slide'>
<h1>A question:</h1>
<p>Q: How do mathematicians categorize non-linearly-separable data?</p>
<p>A: Munge the data until it's linearly separable, thus reducing it to an earlier slide.</p>
<p>Seriously. The munging is done using what are known as "kernel methods".</p>
</div>

<div class='slide'>
<h1>Kernel Methods</h1>
<p><ul>
<li>Functions that munge data</li>
<li>Very faintly magical (because i have no idea how they were derived)</li>
<li>Require some skill to choose the right one for the problem</li>
<ul></p>
</div>

<div class='slide'>
<h1>Kernel Methods + Support Vectors = Support Vector Machines</h1>
</div>

<div class='slide'>
<h1>In Perl:</h1>
<p>Algorithm::SVM - bindings to libsvm</p>
<p>(Also wrapped by AI::Categorizer)</p>
</div>

</body></html>