Friends, until the most long-awaited event of the year in the field of data science remained 1 day! April 28 will be the fifth Moscow Data Fest. Under the cut, our story about the reports and activities of the Mail.Ru Group at the conference.
We prepared 6 reports in the main program of the festival and 4 performances at our stand.
')
Dmitry Parpulov from the Mail.Ru Machine Learning Team will tell you about the algorithm for prompting responses to incoming emails (Smart Reply), launched by mail in December 2017. You will find out what is under the hood of the service, what approaches you tried and what you finally chose. The proposed answers are better and more varied, how to evaluate the quality of the resulting model, and many more interesting things.
Dmitry Bugaychenko will consider the main architecture of the Spark ML machine learning library, and also will tell about the peculiarities of its use for solving real problems of processing large amounts of data. Special attention will be paid to a number of restrictions complicating the use of the library. Dmitry will also tell what extensions for standard elements had to be developed in order to circumvent these limitations and unleash the potential of massive distributed machine learning. The work of the standard library and its extensions, we will demonstrate the example of the problem of ranking news feed Odnoklassniki.
Vsevolod Vikulin will consider the recommendation task on the example of the Chinese goods store
Pandao . He will show typical mistakes, suggest atypical solutions. Vsevolod will talk about building a system of recommendations that any data science specialist can do on his knee using sklern.
Alexander Tobol in his report will consider:
- pipeline for building vectors of users and user search on the loaded photo;
- Neural network training: building a dataset from real user data> Neural network training> building a dataset - cook until ready;
- building a normalized user vector;
- clustering of persons on the user photo and determining the vector of the account holder;
- iron and optimization, running in the cloud, fault tolerance, launching neural networks on the CPU and GPU.
In the fall, VKontakte made a redesign of mobile applications, within which they updated the Tape of recommendations. In the process of work, many difficulties arose, both with the implementation of machine learning, connecting millions of users with millions of authors, and with the speed of work, and even with design. Andrey Yakushev will share our experience of creating such products from scratch by a small team, and also tell you how we monitor what is happening on the platform and how we find the following growth points.
Now our company is developing more than 150 projects, most of which use artificial intelligence. Mikhail Firulik will review the AI ​​technologies we use: “AI in Mail.Ru Group products: overview of cases and technologies”
Tech Talks from Mail.Ru Group experts
- 12.50-13.00 - “Distributed Trainee”, Sergey Cheparukhin, programmer-researcher;
- 14.50—15.00 - “Cases of using computer vision”, Vladimir Konev, product manager;
- 15.05–15.15 - “Face Recognition: from scratch to hatch”, Alan Basishvili, programmer-researcher;
- 17.35—17.45 - “Sequence to sequence interactive models”, Oleg Shlyazhko, leading research programmer.
At the booth we treat fresh fruit and fresh fruits all day, talk about life in Mail.Ru Group and communicate. We show the work of our stand Face Recognition.
You can also get a T-shirt from Mail.Ru and other gifts:
- Find our photographers in Mail.Ru T-shirts.
- Take a photo with them alone or with friends.
- Come to our booth 1.5 hours after the photo shoot.
- Log in to the Face Recognition counter and get your photos from #datafest by mail.
- Take your gift without leaving the counter!
In a previous post about Data Fest, we talked about conference sections.
See the schedule , come to the reports and to our booth!