×

Data Science UA Conference
  • ×
  • Data Science UA Conference

    5th International
    Conference in Kyiv
    2018
    |
    NOVEMBER 24
    |
    KYIV

    DATA SCIENCE UA

    Data Science UA Conference brings together leaders in machine learning,
    analytics, BI, data science, AI for a day-long exploration of how data trans-
    forms the World today and what is going to be tomorrow.

    500
    Participants
    18
    Speakers
    3
    Stages
    10
    Hours of networking

    Stages


    Tech

    Business

    Workshops

    Panel discussion

    Speakers


    Volodymyr Tkachuk

    Head of Edge Research, Ring Ukraine

    Theme: How to deploy Machine Learning algorithm on embedded

    Alexandr Honchar

    AI Solution architect, MAWI

    Theme: Multitask learning: learn more to learn better

    Taras Hnot

    Senior Data Analyst, SoftServe

    Theme: Сustomers profiling based on psychometric characteristics

    Vasyl Mylko

    CEO and Founder, Ingeenee

    Theme: AI solves NP-complete

    Viktor Sakharchuk

    Independent CV/ML R&D professional

    Theme: Confidence Measures For Stereo Vision: an Engineer’s View

    Denis Dovgopoliy

    Founder at GrowthUP Group

    Theme: Features in attracting investors to the early stage AI-startups

    Andy Bosyi

    Founder and CEO of MindCraft.ai

    Theme: Active learning metrics or when to stop label the data

    Workshops


    Alexandr Honchar

    AI Solution architect, MAWI

    Workshop: Applied multitask learning with Keras

    Panel discussion


    Borys Pratsiuk

    Head of R&D, Ciklum

    Panel Discussion

    Denis Dovgopoliy

    Founder at GrowthUP Group

    Panel Discussion

    Hosts


    Jane Klepa

    Executive director,
    1991 Open Data Incubator

    Daniel Chernega

    Comandante in
    “Che – Guerrilla Marketing”

    Tickets


    Buy 5 tickets Get 10%
    off
    Buy 10 tickets Get 15%
    off
    25% discount for student

    1450
    June 20 – August 31

    1950
    September 1- September 30

    2150
    October 3 – October 31

    2750
    November 1 – November 23

    3000
    November 24

    Become one of our Partners

    The Data Science UA Conference attracts both, the best and brightest attendees,
    and many of the top companies using data science.

    Silver partners


    Bronze partners


    Media Accreditation

    APPLY NOW >

    Become a volunteer

    APPLY NOW >

    Contacts


    Location

    Oasis concert hall Kyiv

    Kiev, Lipkovsky street, 1A
    “Ultramarine”, 3rd floore


    Contact information

    +38 099 055 23 92
    info@data-science.com.ua
    #datascienceua


    ×

    Volodymyr Tkachuk

    Head of Edge Research, Ring Ukraine

    Theme: How to deploy Machine Learning algorithm on embedded

    About speaker:
    In 2008, he graduated from NTUU “KPI” in “Applied Mathematics”.
    Worked as a programmer, researcher, researcher, technical manager in such companies as Microsoft (Copenhagen), the Department of Pattern Recognition in Kibtsentr (Kyiv), Materialise NV (Kyiv, Shanghai) and Ring Ukraine (Kyiv).
    I was engaged in projects of 3D reconstruction of a scene on two-dimensional shots, segmentation of a three-dimensional medical image, optimization of 3D printing algorithms.
    The current work in Ring Ukraine is connected with the development and implementation of algorithms to be performed on “smart” devices such as courier calls and Ring’s cameras.
    Briefly about the report:
    Learning complex algorithms, such as large neural networks, is still a resource-intensive procedure that requires complex hardware. At the same time, the emergence of “light” architectures (for example, MobileNet), and the simultaneous increase in the power of embedded systems, creates the ability to perform machine-trained algorithms on “smart” devices.
    There are various ways to optimize and run their algorithms. The easiest way is to use community-supported and developed frameworks such as Mini Caffe and TensorFlow lite. Some manufacturers of built-in systems are equipping their products with chips optimized for neural networks and offering libraries to work with them. Finally, you can always program itself to implement any network in the language C / C ++ with optimization for a particular processor.

    ×

    Denis Dovgopoliy

    Founder at GrowthUP Group

    Theme: Features in attracting investors to the early stage AI-startups

    About speaker:
    He graduated from the Kiev Polytechnic Institute (1996) and International Institute of Management, MBA Program (2000). He is a co-founder of the first Ukrainian business accelerator for IT startups of the GrowthUP.
    One of the leaders of the Ukrainian venture and entrepreneurial community.
    Founder and organizer of several significant industry events: IDCEE, iForum, Silicon Valley Open Doors, Startup Crash Test, etc.
    Before founding BVU Group, he held C-level positions in telecom companies. He founded several local companies.

    ×

    Denis Dovgopoliy

    Founder at GrowthUP Group

    Theme: Features in attracting investors to the early stage AI-startups

    About speaker:
    He graduated from the Kiev Polytechnic Institute (1996) and International Institute of Management, MBA Program (2000). He is a co-founder of the first Ukrainian business accelerator for IT startups of the GrowthUP.
    One of the leaders of the Ukrainian venture and entrepreneurial community.
    Founder and organizer of several significant industry events: IDCEE, iForum, Silicon Valley Open Doors, Startup Crash Test, etc.
    Before founding BVU Group, he held C-level positions in telecom companies. He founded several local companies.

    ×

    Taras Hnot

    Senior Data Analyst, SoftServe

    Theme: Сustomers profiling based on psychometric characteristics

    About speaker:
    Taras is a data analyst with a strong experience in performing large-scale data mining, statistical modeling and patterns extraction. He has been involved into various projects in digital retail industry, including developing analytical tools for a customer profiling and segmentation, building complex personalization and product recommendation solutions and customer behavioural modeling using psychometry and neuroeconomics techniques. Taras is a co-author and an author of multiple scientific and media publications, including “Anomaly Detection – Unsupervised Approach”, “Recommender Systems Comparison: The Best Performing Algorithm” and “Bitcoin Network Analytics”.
    Briefly about the report:
    We will talk about retailers’ customers profiling, but not based on gender, age, occupation, family. Profiling will be based on customers’ psychometric characteristics or OCEAN scores. These scores would be used to show the type of ad that customer will like (transformational or informational/comparative or non-comparative / one or two-sided ads etc.). Psychometric characteristics are extracted from textual information, written by the customer, and shopping patterns using state-of-the-art techniques in machine learning like XGBoost, Random Forest, LSTM models.

    ×

    Alexandr Honchar

    AI Solution architect, MAWI

    Theme: Multitask learning: learn more to learn better

    About speaker:
    Alexandr is a machine learning expert with experience in computer vision and time series analysis, who worked with Ukrainian (MAWI, ARVI), Russian (Mlvch), American (Inma AI), and Italian (HPA Srl) companies. He is mainly interested in bringing research ideas to production and making value from latest theoretical developments in AI. Currently he is working in MAWI (USA-based company with R&D department in Ukraine) on biomedical signal analysis, in particular ECG, and applying machine learning for classical applications like medical diagnostics and developing novel cases as well. Meanwhile, Alexandr teaches deep learning seminars in University of Verona and writes a popular blog on Medium.
    Briefly about the report:
    Machine learning as a discipline went through a long way from statistical pattern recognition in clean and tiny data to learning meaningful representations from complex high dimensional data on scale with deep learning algorithms. Indeed, it reminds in a way on how human brains process the information… but not close enough. One of the amazing abilities of human brain is performing several different conclusions from one piece of data – seen a cat, we don’t just identify a class “cat” with 97% of confidence, we also see the color, how far it is from us, try to detect the mood and cuteness of the animal. The same happens when we read text or hear the speech – we understand a lot from read of heard sentences and the same should do deep learning algorithms. In this talk we’ll study the concept of multitask learning – doing several tasks on the same data, understand why it helps machine learning to generalize and understand problem better and review cases in computer vision, NLP, reinforcement learning and other fields so you can apply this idea in your business.

    ×

    Alexandr Honchar

    AI Solution architect, MAWI

    Briefly about the Workshop:
    In this workshop, we will study multitask learning in practice. Multitask learning is a concept of designing machine learning algorithms in such a way, so they learn to solve several problems using the same set of parameters. Not surprisingly it helps to generalize and learn new and more useful features from the data than with using a single task and a single loss function. We will briefly review the theory behind multitask learning but will concentrate on implementing these models (mainly neural networks) by ourselves using Keras and apply it to different datasets (computer vision, signal processing) to prove their effectiveness.

    ×

    Borys Pratsiuk

    Head of R&D, Ciklum Theme

    About speaker:
    Borys graduated with honors from the Chair of Physical and Biomedical Electronics of the KPI in 2007 on the specialty “Physical and Biomedical Electronics”, and in 2012 he defended his dissertation at the Faculty of Electronics in the KPI. Boris Pracyuk works in the Ciklum R&D department. And also has his own startup – Fino (financial assistant).

    ×

    Jane Klepa

    Executive director,
    1991 Open Data Incubator

    About Jane:
    Executive Director at 1991 Open Data Incubator, Co-founder & CMO at SPREAD.
    Startups mentor and adviser. Teenagers mentor at ukrainian business schools (Computer academy “Step” and “Creators”). Actively involved in the tech ecosystem development in Kyiv and Ukraine.
    Expert of a corporate social responsibility. Her first exposure to Ukraine’s IT world was when she worked as a PR-manager at tech events. Then moved to work as an organizer of largest tech events (Ukraine, Russia, Poland).

    ×

    Daniel Chernega

    Comandante in
    “Che – Guerrilla Marketing”

    About Daniel:
    Founder of “CHE-guerrilla marketing”.
    A passionate, futuristic, technocratic dreamer, and evangelist of change.
    “I believe that data science will change the perception of how we live and conceive. Humanity is on the verge of colossal changes, which is why the best we can do is to be surrounded by people who are just dreaming that same way. Those who dream not only to observe changes but also to be their main driving force.
    So, see at Data Science UA Conference 2018.

    ×

    Viktor Sakharchuk

    Independent CV/ML R&D professional

    Theme: Confidence Measures For Stereo Vision: an Engineer’s View

    About speaker:
    Volyn National University – master of science – theoretical physics. Lutsk National Technical University- Ph.D. student. 10+ years of commercial experience as the software/hardware engineer.
    Main skills
    Rapid prototyping and implementation of complex SW/HW solutions for 2D and 3D pattern recognition and object detection from data of different origin, multisensor fusion. Designing environment perception systems for UAV/UGV for their orientation in space. Image processing.
    Applying computer vision and machine learning methods for solving real-life problems.
    Briefly about the report:
    The report will be about the practical aspects of dealing with data obtained from the binocular stereo system. Usually, such data are corrupted by noise and outliers and are rarely used in the raw form. It may lead to incorrect interpretations of the distance measurements and improper reasoning about the scene being observed (phantom obstacles, missing objects etc). Also, it is crucial to obtain the confidence measure to get a reliable sensor fusion system. Therefore estimating a reliable confidence of the data is an important step in the data processing pipeline of every autonomous system
    We will touch the confidence estimation approaches for the data from common stereo disparity estimation algorithms being used in robotic perception systems that work in real-time

    ×

    Vasyl Mylko

    CEO and Founder, Ingeenee

    Theme: AI solves NP-complete

    About speaker:
    Co-created R&D at SoftServe in 2008, and led it as R&D Director for 10 years.
    Recent prominent researches have been done in Emotional Intelligence, Machine Empathy, Biometry, Embedded AI, IoT, AR. Our BioLock authentication by ECG was a finalist in the 2017 SXSW Interactive Innovation Awards
    Co-designed several cutting-edge technologies with/for Samsung since 2016.
    Earlier in R&D: built UX Design Office and Security Consulting
    Briefly about the report:
    Despite chip/memory/storage/networks giant leap since 1970s, a dozen of known optimization problems are still unsolved in short time using cheap resources with acceptable quality. Naming a several of them: Traveling Salesman Problem, Constraint Satisfaction Problem, Protein Folding, Earth Simulation. They are of combinatorial complexity. On the one hand, those problems are attacked with HPC by supercomputers. On the other hand it could be possible to apply new kind of AI, to solve them with cheaper resources. There could be a link between intelligence and computational universe – AI can build complexity.
    At Ingeenee, we are building an ingenious engine, that can find a maximum sightseeing in any geographical area and build the route through the cities with POIs, within time and budget constrainsts simultaneously. I will share stories how we solve this NP-complete problem (TSP+CSP) with ML and Evolutionary Computing.

    ×

    Andy Bosyi

    Founder and CEO of MindCraft.ai

    Theme: Active learning metrics or when to stop label the data

    About speaker:
    Andy has 25 years experience in software development. He built his first Artificial Neural Network in 1986, worked on various IT projects related to big data and data analytics, from 2016 switched on data science projects and set up a company MindCraft fully dedicated to creating data-driven solutions to bring new ideas from data insights and reduce staff costs
    Briefly about the report:
    Most sources are lively talking about machine learning systems, often not taking into account the fact that 80% of the time and energy of analytic goes to data mining, analysis, and labeling.
    Scope big data system it’s a huge expense for low-skilled (and not only) work. Here comes the help of semi-supervised learning methods, which allow one to teach models for partially labeled data
    In the report, we will consider the case when data labeling and learning process go iteratively with maximum efficiency. The main question here is when it is enough to spend resources on data labeling while maintaining the qualitative metrics of the model.