# Online machine learning tutorial

Does anyone know some good tutorials on online machine learning technics? I.e. how it can be used in real-time environments, what are key differences compared to normal machine learning methods etc.

UPD: Thank you everyone for answers, by "online" I mean methods which can be trained in a real-time mode, based on a new inputs one by one.

• Could you clarify key aspects of "online" that you are interested in? Do you have a specific form for the data, or any options to pre-train your algorithm before the online part? – Neil Slater Jun 17 '14 at 8:41
• do you mean to analyze datastreams? – Johnny000 Jun 17 '14 at 11:42

1. There is a very nice library of online machine learning algorithms from a group at NTU, called LIBOL. This would be a very good place to start experimenting with the algorithms.
The accompanying user guide, and associated JMLR publication are very nice introductions to the basic algorithms in this field.
2. Avrim Blum has an older and more technical survey paper on online learning algorithms.
• The first two links are dead. – Suzana Jun 23 '15 at 14:30

There's a lot of online tutorial. Especially in youtube, but if you will want accurate website you can see from here http://ttic.uchicago.edu/~shai/icml08tutorial/ or http://cs229.stanford.edu/materials.html. you can visit them now.

• Link-only answers are discouraged. Can you expand on the content in this link that you think is relevant? – Sean Owen Dec 17 '14 at 13:49

You can see if you can mix Spark streaming (https://spark.apache.org/docs/1.1.0/streaming-programming-guide.html) and Spark ML Library (https://spark.apache.org/docs/1.1.0/mllib-guide.html).

Spark Streaming permits to process live data streams and Spark ML Library is a Machine Learning Library for Spark. So maybe you can do something good!

But this is a very interesting subject, I am working on it. It can be good to create a Google Community on it (https://plus.google.com/communities)?

There are plenty on youtube and here's a famous one by Andrew Ng from coursea: https://www.coursera.org/course/ml

• I am not 100% convinced by the topics on that course. For example, are SVMs actually of practical use these days? You never see a winning Kaggle entry that used SVMs as the main part. – Lembik Jun 17 '14 at 7:45
• I think OP's question is specifically about online techniques - i.e. where system is expected to learn at least partially "on the job". Not online tutorials – Neil Slater Jun 17 '14 at 8:40
• I agree with @NeilSlater since the OP mentioned "compared to normal machine learning methods". – Rohit Jun 17 '14 at 9:45
• lol, "online" is ambiguous – alvas Jun 17 '14 at 10:19