Machine Learning :: Reading Group

Session #5: Time Series

December 25, 2013 · 7:00 PM
This location is shown only to members

Our upcoming session will deal with time series prediction & classification.

I suggest we'll go over the following interesting recent works. We need a few volunteers that will help us look at the essence of these works. Please send me a message if you would like to assist.

1) Online Learning for Time Series Prediction, Anava et al., COLT 2013 -- http://bit.ly/IxJFO6 (A fresh look at a veteran algorithm. Connecting the world of time series prediction with the theoretical toolbox of online learning.)

2) Behavioral Dynamics on the Web: Learning, Modeling, and Prediction, Radinsky et al., ACM Transactions on Information Systems 2013 -- http://bit.ly/18uiPTx (Dealing with web-scale time series. The what, how, and why.)

3) Playlist Prediction via Metric Embedding, Chen et al., KDD 2012 -- http://bit.ly/1cSCEku (A very interesting way to think of time series prediction. It connects nicely with the discussions we had on music recommendation and app prediction.)

4) If time* will permit I will share a work done at my workplace.

* :)

** Please send me a message if you would like to host this/next session (round about 20 people).

Join or login to comment.

Hit enter to add your reply

Ron B.

Hi,

I am starting a thread on 'related links.' (Help by adding links that you find them related to our discussion and worthy.)

- Re: Comment on 'learning over several sequences', here is a link to a nice survey on 'online portfolio selection'. The first author has done his PhD on this topic (and has real interesting ideas). link: http://bit.ly/1dKhJk6

0 · December 28, 2013
- Amit M.
  
  When we discussed identification of periodic signals I've talked about the fact that FFT does not capture non-sin-like signals well. This article uses ANOVA to find such periodical signals (a method I've also used when looking for periodic signals in astrophysical observations) http://1.usa.gov/1eq7... (be careful - a pdf)
  
  2 · January 1, 2014
Ron B.

Hi, We had a fun session. Amit presented a great crush course on time series analysis which lead to interesting discussions. I've learnt a few new thing (thanks Amit!). Then, we covered Chen el al.'s work on learning to generate playlists. Thanks for Citi innovation lab tlv folks (Oren/Sasha/Bella) for hosting us.

What's next?

1) Graphlab's co-founder is organising an applied machine learning day (next month): http://bit.ly/19qjKoD -- note that Kira (co-author of paper #2) is presenting her work.

2) Upcoming sessions -
-- I would like us to discuss 'deep learning', and looking for volunteers. Please contact me if you have some knowledge / wanna present an intro on this topic.
-- Please contact me if you have an idea for an interesting topic / work that you think we should cover.

Thanks!
Ron

1 · December 26, 2013
- Mich
  
  i'll be happy to if there's no deep-learning expert (i'm not).
  
  0 · December 26, 2013
Assaf R.

Thanks for organizing! It was a very interesting session, and looking forward to the next ones.

I apologize for having to leave early. How long did you stay there?

0 · December 26, 2013
Hershel S.

Thanks to Ron for organizing and the Citi folks for hosting.

3 · December 26, 2013
Guy

Great discussion tonight.
Sorry I had to leave early...

Looking forward for the next interesting meeting.

1 · December 25, 2013
Ron B.

Hi All,

We'll meet this Wed @19:00 in Citi Innovation Lab TLV, Kiryat Atidim Tower (4th floor). You can park freely near by - at Atidim pool and sport centre. We're meeting in the nearest tower. Please drop me a msg in case you need more information.

Ron

0 · December 23, 2013
- Mich
  
  no security issues are expected at the lobby?
  
  0 · December 23, 2013
- Ron B.
  
  בדרך כלל משאירים רשימת שמות בלובי. שלח לי את שמך ואעביר אותה לאנשי סיטי.
  
  0 · December 23, 2013
Hershel S.

I look forward to seeing everybody in a couple of weeks. In the meantime, I'd like to mention that the company where I work, WorldQuant (www.worldquant.com), is hiring math people, with backgrounds in math, physics, statistics, computing, engineering, etc. The job is developing mathematical models for trading in financial markets. Here is a link to the job description:
http://www.jobs-israel.com/posi...

If you're interested, send a c.v. to me ([masked]) and I'll give it directly to the GM, so that it doesn't get buried in the piles that arrive via the website.

0 · December 12, 2013
rouli n.

I would really appreciate some Time Series tutorial from an ML perspective - feature extraction and selection, modeling and so on, before diving into cutting-edge research, if that makes sense.

1 · December 3, 2013
- View all 3 replies
- Ron B.
  
  Sec. 2.1 (and 2.2) of Paper #1 pretty much sums what we need. It discusses the AR + ARMA models. See also Sections 3.1 & 3.2 of paper #2.
  
  0 · December 3, 2013
- Ron B.
  
  I will try to cover Markov Chains for paper #3 and others.
  
  0 · December 3, 2013
- Ron B.
  
  and I am still waiting for volunteers. Please contact me if you can/wanna assist.
  
  0 · December 3, 2013