Repository navigation
Expand file tree
/
Copy pathsemi-supervised.html
More file actions
112 lines (111 loc) · 5.24 KB
/
Copy pathsemi-supervised.html
File metadata and controls
112 lines (111 loc) · 5.24 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
<!DOCTYPE html>
<html>
<head>
<title>Semi-Supervised Learning Guide</title>
</head>
<style>
body{
font-size:30px;
}
</style>
<body bgcolor="#fdfdfb" style="height:100%;">
<table width="90%" align="center" border="0" margin="50" height="100%">
<tr>
<td>
<h1 align="center">Semi-Supervised Learning</h1>
<div align="center"><img src="images/semi-supervised.png" alt="how supervised learning works" width="300px" border="1"></div>
<p>
Semi-supervised learning is a hybrid approach to machine learning that sits
between supervised (all labeled data) and unsupervised (no labeled data) learning.
It is specifically designed for situations where you have a massive amount of data,
but only a tiny fraction of it is labeled.
</p>
<hr>
<h2>☆ How it works?</h2>
<p>The general process often follows these steps:</p>
<ol>
<li><strong>Initial Training</strong>
<blockquote>
The model is trained on the small set of labeled data, just like in supervised learning.
</blockquote>
</li>
<li><strong>Pseudo-Labeling</strong>
<blockquote>
The model is then used to predict labels for the unlabeled data. Because the model isn't perfect yet, these are called "pseudo-labels."
</blockquote>
</li>
<li><strong>Filtering</strong>
<blockquote>
The predictions with the highest confidence are added to the training set as if they were real labels.
</blockquote>
</li>
<li>
<strong>Retraining</strong>
<blockquote>
The model is retrained on the now-expanded dataset (original labeled data + high-confidence pseudo-labeled data).
This cycle repeats to refine accuracy.
</blockquote>
</li>
</ol>
<hr>
<h2>☆ Key Concepts and Assumptions</h2>
<p>For semi-supervised learning to work, algorithms rely on three core assumptions:</p>
<ul>
<li><strong>Continuity Assumption</strong>
<blockquote>
Points that are close to each other in the data space are likely to share the same label.
</blockquote>
</li>
<li><strong>Cluster Assumption</strong>
<blockquote>
Data naturally forms groups (clusters). If two points are in the same cluster, they likely belong to the same category.
</blockquote>
</li>
<li><strong>Manifold Assumption</strong>
<blockquote>
High-dimensional data (like images) actually lies on a lower-dimensional "surface" or manifold.
The model tries to learn this surface to better separate classes.
</blockquote>
</li>
</ul>
<hr>
<h2>☆ Common Techniques</h2>
<table border="1" cellpadding="10" cellspacing="5" margin="2">
<tr>
<th>Technique</th>
<th>How it works</th>
</tr>
<tr>
<th>Self-Training</th>
<td>The model labels its own unlabeled data and uses the best guesses for its next round of training.</td>
</tr>
<tr>
<th>Co-Training</th>
<td>Two different models look at the same data from different "views" (e.g., text vs. images) and teach each other.</td>
</tr>
<tr>
<th>Graph-Based SSL</th>
<td>Data points are treated as "nodes" in a web. Labels "spread" like a virus from labeled nodes to their closest unlabeled neighbors.</td>
</tr>
</table>
<hr>
<h2>☆ Real-World Use Cases</h2>
<ul>
<li>Medical Imaging</li>
<li>Speech Recognition</li>
<li>Web Classification</li>
</ul>
<hr>
<section id="youtube-video"></section>
<h2>☆ Watch this YT Video!</h2>
<div align="center"><iframe border="1" width="560" height="315"
src="https://youtube.com/embed/C3Lr6Waw66g"
frameborder="0"
allowfullscreen>
</iframe></div>
</section>
</td>
</tr>
</table>
</body>
</html>