Difference between revisions of "Machine Learning"

From Cyborg Anthropology
Jump to: navigation, search
 
(19 intermediate revisions by 3 users not shown)
Line 1: Line 1:
<private>
+
[[Image:machine-learning-maggie-nichols.jpg|center|600px]]
 +
===Definition===
 +
Machine Learning is a process of training a computer algorithm to properly classify future inputs after having trained the algorithm with sample data. A program is first trained with known inputs, and "learns" these patterns through one or more statistical methods. Then it can classify new input based on the inputs it has seen before. Machine Learning is often used in computer vision to classify images.<ref>Machine Learning in Computer Vision. Ajay Joshi, Anoop Cherian and Ravishankar Shivalingam. Dept. of Computer Science, University of Minnesota http://www-users.cs.umn.edu/~cherian/ppt/MachineLearningTut.pdf Accessed Oct 2011.</ref>
  
The original link to this is [http://metaoptimize.com/qa/questions/186/good-freely-available-textbooks-on-machine-learning?utm_source=twitterfeed&utm_medium=twitter here]. I'd suggest visiting it for an updated list with lots of relevant comments. The following is more for my own notes.
+
===History===
 +
Ray Solomonoff published the first report on non-semantic machine learning in 1956, titled ''An Inductive Inference Machine''.<ref>Solomonoffm, R. J. An Inductive Reference Machine. Technical Research Group, New York City.  http://world.std.com/~rjs/indinf56.pdf Accessed Oct 2011.</ref> He also was a pioneer in algorithmic probability, publishing several papers in the 1960s. It wasn't until the late 1970s that machine learning started emerging as a more focused field of study in computer vision. Machine learning has also been applied in many other fields such as [[Natural Language Processing|natural language processing]], speech and handwriting recognition, and sentiment analysis.  
  
====Freely Available Books on Machine Learning====
+
===Examples===
The Elements of Statistical Machine Learning is a great text covering most core topics in supervised learning along with a bit of unsupervised learning and some other specialist areas (like high dimensional problems).
+
A program can be trained from security camera images to recognize whether a person is present in the frame. Someone would first train the program by giving it several images without a person present in the frame as the negative input, and then give it several images with a person in the frame as the positive input. The program would then be able to classify future images it is given with reasonable accuracy. This would allow a computer to take an action based on whether it sees a person or not. Tom M. Mitchell provided a widely quoted definition: "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E."<ref>Mitchell, T. Machine Learning. McGraw Hill, 1997. Pg. 2.</ref>
  
Information Theory, Inference, and Learning Algorithms covers what the title suggests. It doesn't have a great deal of depth on the machine learning topics but is good for an overview of the techniques used in Bayesian inference.
+
More recently, machine learning has been applied to create search engine algorithms. Researchers at Cornell University created a prototype, [http://www.cs.cornell.edu/~tj/striver/ STRIVER], which was able to improve its results over time based on which results the visitor clicked on.<ref>Ibid.</ref>
  
Gaussian Processes for Machine Learning is the definitive reference for Gaussian processes.
+
==References==
 +
<references />
  
Data-Intensive Text Processing with MapReduce contains patterns for implementing your algorithm in the Map-Reduce framework.
+
[[Category:Book Pages]]
 +
[[Category:Finished]]
 +
[[Category:Illustrated]]
  
Reinforcement Learning: An Introduction covers the fundamentals of bandit algorithms and reinforcement learning in fully observable worlds (MDPs). Note it says very little about generalisation and practically nothing about acting in partially observable worlds (POMDPs). Since this book was published there has been substantial work in all areas of reinforcement learning; while the book will give you the basics you'll have to do a lot of reading in the literature to catch up to current work.
+
__NOTOC__
 
+
D.Barber: [http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/pmwiki.php?n=Main.Textbook Bayesian Reasoning and Machine Learning]
+
 
+
Andrew Ng's [http://see.stanford.edu/see/materials/aimlcs229/handouts.aspx CS229 Machine Learning] course notes. Comes with video lectures.
+
 
+
Convex Optimization – Boyd and Vandenberghe
+
 
+
Reinforcement Learning - An Introduction
+
 
+
Introduction to Information Retrieval
+
 
+
Introduction to Machine Learning course notes (based on a forthcoming book by Shai Ben-David and Shai Shalev-Shwartz) - 110 pages of comprehensive notes. The notes have a slight bias towards the more theoretical side of ML - PAC, Rademacher complexity, etc., with theorems and all that.
+
 
+
Mike Jordan and Martin Wainwright's [http://www.nowpublishers.com/product.aspx?product=MAL&doi=2200000001 Graphical Models, Exponential Families, and Variational Inference]
+
 
+
[ttp://idiom.ucsd.edu/~rlevy/textbook/text.html Probabilistic Models] in the Study of Language by Roger Levy at UCSD. Still in draft form so incomplete (some chapters have yet to be written) freely available.
+
 
+
</private>
+
 
+
[[Category:Book Pages]]
+
[[Category:Unfinished]]
+

Latest revision as of 15:52, 16 December 2011

Machine-learning-maggie-nichols.jpg

Definition

Machine Learning is a process of training a computer algorithm to properly classify future inputs after having trained the algorithm with sample data. A program is first trained with known inputs, and "learns" these patterns through one or more statistical methods. Then it can classify new input based on the inputs it has seen before. Machine Learning is often used in computer vision to classify images.[1]

History

Ray Solomonoff published the first report on non-semantic machine learning in 1956, titled An Inductive Inference Machine.[2] He also was a pioneer in algorithmic probability, publishing several papers in the 1960s. It wasn't until the late 1970s that machine learning started emerging as a more focused field of study in computer vision. Machine learning has also been applied in many other fields such as natural language processing, speech and handwriting recognition, and sentiment analysis.

Examples

A program can be trained from security camera images to recognize whether a person is present in the frame. Someone would first train the program by giving it several images without a person present in the frame as the negative input, and then give it several images with a person in the frame as the positive input. The program would then be able to classify future images it is given with reasonable accuracy. This would allow a computer to take an action based on whether it sees a person or not. Tom M. Mitchell provided a widely quoted definition: "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E."[3]

More recently, machine learning has been applied to create search engine algorithms. Researchers at Cornell University created a prototype, STRIVER, which was able to improve its results over time based on which results the visitor clicked on.[4]

References

  1. Machine Learning in Computer Vision. Ajay Joshi, Anoop Cherian and Ravishankar Shivalingam. Dept. of Computer Science, University of Minnesota http://www-users.cs.umn.edu/~cherian/ppt/MachineLearningTut.pdf Accessed Oct 2011.
  2. Solomonoffm, R. J. An Inductive Reference Machine. Technical Research Group, New York City. http://world.std.com/~rjs/indinf56.pdf Accessed Oct 2011.
  3. Mitchell, T. Machine Learning. McGraw Hill, 1997. Pg. 2.
  4. Ibid.