Mean Shift Intro 
Mean Shift Intro
by Harrison Kinsley
Video Lecture 39 of 42
Not yet rated
Views: 871
Date Added: August 11, 2016

Lecture Description

Mean Shift is very similar to the K-Means algorithm, except for one very important factor: you do not need to specify the number of groups prior to training. The Mean Shift algorithm finds clusters on its own. For this reason, it is even more of an "unsupervised" machine learning algorithm than K-Means.

The way Mean Shift works is to go through each featureset (a datapoint on a graph), and proceed to do a hill climb operation. Hill Climbing is just as it sounds: The idea is to continually increase, or go up, until you cannot anymore. We don't have for sure just one local maximal value. We might have only one, or we might have ten. Our "hill" in this case will be the number of featuresets/datapoints within a given radius. The radius is also called a bandwidth, and the entire window is your Kernel. The more data within the window, the better. Once we can no longer take another step without decreasing the number of featuresets/datapoints within the radius, we take the mean of all data in that region and we have located a cluster center. We do this starting from each data point. Many data points will lead to the same cluster center, which should be expected, but it is also possible that other data points will take you to a completely separate cluster center.

Course Index

Course Description

The objective of this course is to give you a holistic understanding of machine learning, covering theory, application, and inner workings of supervised, unsupervised, and deep learning algorithms.

In this series, we'll be covering linear regression, K Nearest Neighbors, Support Vector Machines (SVM), flat clustering, hierarchical clustering, and neural networks.


There are no comments. Be the first to post one.
  Post comment as a guest user.
Click to login or register:
Your name:
Your email:
(will not appear)
Your comment:
(max. 1000 characters)
Are you human? (Sorry)