I wrote an email out to the members of the Paris Machine Learning meetup group and just translated it for the blog ( the French version is here: Une rétrospective rapide des présentations du Paris Machine Learning Meetup)
With 2014 ending, here are some notes on what happened last year and a half at the Paris Machine Learning meetup.From an organisational point of view, our sponsors allowed us to host the meetup in several places. These sponsors, DojoCrea, Criteo, Google, TheFamily, NUMA, and Hopwork and Aneo, showed their understanding of how important algorithms are becoming in our world. We are very grateful to them that they have enabled the exchange of ideas within the community. For brevity, we have had 19 meetings since June 2013 and have become, with 1650+ members, the largest meetup in the world on the subject of Machine Learning outside the United States (top 6 including the US). One of the innovative aspects of our meetings has been the use of remote video presentations. The idea is that our community needs to hear new ideas faster than through traditional filters. These presentations were made from Mountain View, Boston, New York, Houston, Davis, Montreal, Tubingen and Israel. Everyone probably also remembers the somewhat complex meetup last June that took place with three other European cities (Zurich, Berlin, London). In general, Franck, Frederick and I are always amazed at the wide variety of interventions and the quality of the Q&A sessions.We are also begining to fill up our meetup roster faster than before with the next three meetups presenters nearly filled. In order to help in this complicated programming process, we just posted a form that will allow you, if you are interested to intervene in the meetup, to let us know about your interest, it's here: http://goo.gl/forms/mRXiaoio8p
From a technical standpoint, we have seen the emergence of CNNs (Convolutional Neural Networks) through several presentations, but we have also seen that through several Kaggle challenges such techniques as Random Forests seemed to be a good tool for fast and effective results. We can now see Kaggle and similar services as a way to perform some sorts of hyperparameter search and a matching service between datasets and relevant algorithms. The problem with these services is that they do not allow us to judge the ability of an algorithm to perform exploration rather than exploitation. We had at least two such presentations on exploration this past year. This is very new in the field of Machine Learning that has most been exploiting correlations, but the dimensionality of ever increasing datasets (Big (ger) Data) suggests that these techniques are bound to grow in importance. In the same vein, having an algorithm that is explanatory seems to be also a great challenge. We could talk more about these fascinating topics, but we invite you to look at the archives of the meetup that contain all the links to presentations and videos of the meetups. A listing by theme can be found below.
From a more general point of view, we are also witnessing a growing disconnect between the real capabilities of algorithms and their romanticized version in the press "tech" coverage or in the mouths of people who should know better. This is one of the reasons why Franck and I have decided to create the 'Hors series' meetups where we ask presenters to provide some context within a particular perspective or trade.
Finally, do not forget to post your open jobs on the LinkedIn group which now has more than 580 professionals in the Paris area. To register for the meetup, you can do it here of course.
Meanwhile, Franck Bardol, Frederic Dembak and myself wish you a happy 2015.
Igor Carron.Survey
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.
No comments:
Post a Comment