Welcome to AI & BigData Lab June 4th AI & Big Data Conference



    On June 4 in Odessa , our FlyElephant team , together with GeeksLab, will hold the third annual technical conference on artificial intelligence and big data - AI & BigData Lab .

    At the conference, developers will discuss implementation and application of various algorithms, tools, and new technologies for working with big data and artificial intelligence. Implemented projects will be presented, and the functionality and principles of their work will be described.

    The conference program AI & BigData Lab is already partially formed. Among the accepted reports can be noted:

    • How we taught to think digital. (Diana Limanskaya, VertaMedia Analyst Coordinator)
      When you go to the site to watch your favorite movie or series, you wait a few seconds before the advertisement is shown. In the few seconds before an ad is showing, hundreds of complex processes are taking place. I want to tell how we created a self-learning system for a digital marketing platform based on mathematical techniques, what stages we went through in its development and what problems we encountered.
    • Translation from “bad” English to “good”. (Anatoly Vostryakov, researcher at Grammarly)
      In the beginning, I want to give a short historical digression into ways to automatically correct errors in tests. In the main part of the report, I would like to highlight the latest methods of error correction using what is now called neural machine translation. That is, we translate the English text into English, but with the corrected errors in the output. Unfortunately, I am limited in specific examples from Grammerley's practice, so the report will be in the form of a review of the algorithms that are already there or will appear by the time of the report.
    • Efficiently compute k averages for a distributed big data stream. (Artyom Barger, Research Engineer at IBM, Israel)
      In this talk, I will provide a deterministic algorithm that allows you to efficiently calculate k average values ​​(k-means) in a continuous data stream in real time. The sublinear algorithm uses only logn * k ^ O (1) memory, it also easily adapts to distributed computing systems, allowing computation to be directly proportional to available computing power with reduced time. Empirical results for popular datasets will be presented at the end.
    • #DataForGood - How to change the world for the better with data analysis. (Maxim Tereshchenko, Product Owner at Zoomdata) 
      The use of Big Data to optimize and improve the effectiveness of decision-making in business is already being discussed. Almost every large corporation has a Big Data platform in its arsenal. But as part of the report, I would like to move away from business and consider the topic of using AI and Big Data for social projects. Hundreds and thousands of analysts, Data Scientists, Big Data engineers unite and implement projects that change the lives of ordinary people around the world and, especially, help residents of underdeveloped countries. Here we are talking about a completely different level of motivation and teamwork. As part of the report, I would like to discuss what drives these people, what real projects with which technologies were implemented, and how they changed people's lives. 
    • Methodology of Data Science Projects. (Sergey Shelpuk, Head of Data Science at VITech)
      Data analysis projects are a challenge not only for engineers, but also for managers. The report will be devoted to the features of such projects compared to conventional development, team roles and building customer interactions in the face of R&D uncertainty.
    • Learning deep, very deep and recursive networks. (Artyom Chernodub, NS at IPMMS NASU)
      The report provides an overview of new approaches to teaching deep and recurrent neural networks. The orthogonal initialization of weights for convolutional and recurrent neural networks and its influence on the problem of vanishing gradient effect, normalization of mini-packets (batch normalization), and difference learning (residual learning) are discussed.
    • MOLAP: New frontiers for the possible. (Konstantin Gerasimenko, CEO at Easy MOLAP, Germany)    
      A story about what MOLAP is. Comparison with traditional approaches. Advantages and disadvantages. 
    • Spike and bionic neural networks: problems and prospects. (Dmitry Novitsky, senior researcher, associate professor at the Institute of Cybernetics of NASU)
      In the world of machine learning, feed-forward neural networks have dominated for many years, which have almost nothing to do with the neurons and networks of our brain. In this report, we will get acquainted with bionic (biologically plausible) neural networks. In most of them, neurons emit and receive impulses (adhesions). What are the problems and difficulties of training such networks? In what traditionally unsolvable (or poorly solved) problems can they be effective, how effective are the human and animal brains in them? How can such networks be implemented in hardware, and what is neuromorphic computing? - These are the questions to which this presentation is devoted.

    Registration and all the details on the conference website . For readers of our blog, there is a 15% discount promotional code: FlyElephantHabrahabr.

    Also popular now: