Speakers

Mohamed-Achref Maiza
Mohamed-Achref Maiza
Senior Data Scientist, Renault DigitalOptimization in multi-label classification and system orchestration with TFX and Kubeflow Pipelines

Business track

Language: English.

Mohamed-Achref has a background in computer science and engineering. He has been working on deep learning applications for car diagnostic and visual quality inspection.

The talk would deep dive into technical aspects such as loss optimization strategies, mathematical properties, optimizing and satisficing metrics, hyper-parameter tuning, components for ML pipeline automation, application for car diagnostic and image classification with neural networks.

Mohamed-Achref Maiza
Senior Data Scientist, Renault DigitalOptimization in multi-label classification and system orchestration with TFX and Kubeflow Pipelines
Luis
Dr. Luis Moreira-Matias
Head of Data Science, KreditechTop Deadly Sins of Applied Machine Learning in Finance

Business track

Language: English.

Luis has a Ph.D. on Machine Learning and a history track of publishing/serving (50+ pubs.) on the major Data Science venues worldwide such as KDD, AAAI, IEEE TKDE or ECML/PKDD. His path in Industry encompasses a long successful track on deploying ML methods in commercial products with proven added value throughout the Finance, Mobility, Retail and Energy industries, from EMEA to APAC. Recognized as atypical for an aptitude in presenting complex AI concepts in a translated manner for general audiences, being regularly invited for Keynotes worldwide (ranging from Brisbane, Australia to Las Palmas, Spain).

In this talk, Luis will be uncovering the common pitfalls on bringing applied DS projects to life in Finance and some of the state-of-the-art solutions that his team brought up to address them. Curious for more? See you there.

Dr. Luis Moreira-Matias
Head of Data Science, KreditechTop Deadly Sins of Applied Machine Learning in Finance
Oles Petriv
Oles Petriv
CTO, Reface.AIML research: how to keep going

Keynote

Language: Ukrainian.

For the last seven years, he has been actively researching and developing computer vision and natural language processing systems. He is the author of a machine learning course on the Prometheus platform and an in-depth training course at the ARVI Lab.

He has extensive experience in video processing using deep learning methods for detecting objects and actions, predicting image depth maps, semantic segmentation and generating subtitles for images and video studios in Hollywood.

He has developed one of the first automation systems to control the placement of groceries at the store shelves using neural networks. He led the development of many projects for automated analysis of news in various languages, recognition of entities, analysis of conceptual drift and representation of language structures using machine learning systems.

Oles Petriv
CTO, Reface.AIML research: how to keep going
Borys_Pratsiuk
Borys Pratsiuk
CTO, ScalarrML models in production. Installation and configuration

Technical track

Language: Ukrainian.

Borys graduated from the Chair of Physical and Biomedical Electronics of the KPI with honors in 2007 on the specialty “Physical and Biomedical Electronics”. He defended his dissertation at the Faculty of Electronics in the KPI in 2012.

Borys works CTO at Scalarr.

He will tell why ML model training is not the main task for model deployment to production.
Also go through the Kubeflow setup and pipelines maintain.
And he will show a real case of how it changes the philosophy of ML team and how they write more standard code for production deployment.

Borys Pratsiuk
CTO, ScalarrML models in production. Installation and configuration
Michael Korkin
CTO, EverguardHow to Identify An Object If You Must

Business track

Language: Russian.

Many industrial workspaces remain extremely dangerous for workers. Steel plants, construction sites, oil rigs, mines, and other industrial environments have high rate of accidents, including fatal.

Our company is on a mission to radically improve worker safety with computer vision, deep learning, and sensor fusion. Unlike typical applications of computer vision, we are confronted with harsh environments, low visibility, lack of basic data communications, and difficulty in obtaining training datasets. All of that calls for highly innovative solutions and thinking outside the bounding box.

Michael Korkin
CTO, EverguardHow to Identify An Object If You Must
Vladislav-Zavadskyy
Vladislav Zavadskyy
Head of Research, NeoRenderAnimable Neural 3D Models

Technical track

Language: Ukrainian.

Vladislav studied at the KPI at the Faculty of Informatics and Computer Engineering, where he defended a diploma about Neural Architecture Search based on Reinforcement Learning.
He’s currently working on creating efficient neural representations of scalar fields. On previous projects, he worked on Computer Vision problems (mostly generative models), NLP and time series analysis.

Today meshes are used ubiquitously to represent 3D shapes. They’ve been around for a long time and several optimizations have been developed to rasterize them efficiently. However, they still have a few drawbacks which limit their possible applications.

NeoRender is a company, which strives to create a new way to represent 3D shapes, which will be free of such drawbacks. On this talk, you’ll get a glimpse of what NeoRender doing on the example of a rigged (animable) 3D model of humans call SMPL. You will learn how to convert textured 3D meshes to neural network representations, which then can be used to approximate physical interactions efficiently and also rendered in a variety of poses.

Vladislav Zavadskyy
Head of Research, NeoRenderAnimable Neural 3D Models
Raid Arfua
Senior Machine Learning Engineer, SoftConstructObject tracking system using multiple NNs with an application in Sports

Technical track

Language: Ukrainian.

ML Engineer with a wide range of expertise in data science engineering who participated in Recommendation Systems, Computer Vision, Big Data projects. Holds two master degrees in Theory of Probability and Mathematical Statistics, and Economics. Has a highly positive attitude towards applying ML and Data Science in the most suitable way. He has a lot of interests in Sciences, Philosophy, Sports, and combines it in everyday work.

Tracking objects in Sports requires specific complex solutions: how to combine multiple NNs in one modular system to solve different tasks like object detection, classification, clusterization, etc. How to incorporate domain knowledge to increase the accuracy of tracking. Which classical video processing methods allow improving tracking. How to test separated modules and systems in general. Examples of video from football games will be presented.

Raid Arfua
Senior Machine Learning Engineer, SoftConstructObject tracking system using multiple NNs with an application in Sports
Muntaniol
Maryna Muntaniol
Senior Consultant, People Advisory Services, EY UkraineData rules - the impact of Data Science on business

Business track

Language: Ukrainian.

Maryna is the leader in Digital HR and focuses on organizational performance improvement and HR management. Maryna joined EY in 2015. She specializes in analysis of processes efficiency, HR automation, strategy development, IT systems benchmarking and selection, organizational structures and processes redesign.
Maryna consults companies on employer branding, conducts studies in the field of integrated talent management. She is a speaker at public events, trainer at trainings and seminars. Maryna holds a National Technical University of Ukraine Igor Sikorsky Kyiv Polytechnic Institute diploma the with a major in Economic Cybernetics.

Maryna Muntaniol
Senior Consultant, People Advisory Services, EY UkraineData rules - the impact of Data Science on business
Holovko
Yuliia Holovko
Manager, Practice People Advisory Services, EY UkraineData rules - the impact of Data Science on business

Business track

Language: Ukrainian.

Yuliia joined EY in 2014. She specializes in labor market studies and HR management.
Yuliia conducts compensation and benefits surveys in Ukraine and CIS, develops grading pay structures. She has experience in integrated HR management , HR audits, short and long-term employee incentive programs development.
Yuliia graduated from Taras Shevchenko National University of Kyiv with a degree in Enterprise Economics.

Yuliia Holovko
Manager, Practice People Advisory Services, EY UkraineData rules - the impact of Data Science on business
Sokolov
Michael Sokolov
CTO, Dex Technologies17 elegant ways of shooting yourself in the foot during question-answering system development

Technical track

Language: Ukrainian.

Michael is a data strategist, NLP expert and systems architect .
Successfully implemented dozens of large projects, including projects for the US government and achieved state-of-the-art results in QA systems.

He runs his own data science R&D company and coordinates several companies in the development and delivery of AI software and solutions for IT, Healthcare, Finances, Cybersecurity, Retail and other fields.

His talk will cover the most common pitfalls during general and domain-specific question-answering systems development. The talk also describes 17 helpful patterns for how to avoid these pitfalls. Besides talk, you will learn how to manage question-answering systems design, functionality, automation, scalability and cost reduction. The presentation will also help you comprehend the most common operational aspects of question-answering systems, based on existing experience. For a wide audience this knowledge can be extrapolated to NLP and software engineering in general.

Michael Sokolov
CTO, Dex Technologies17 elegant ways of shooting yourself in the foot during question-answering system development
Mykola Maksymenko
R&D Director, SoftServeBridging Quantum Computing and Machine Learning in Tensor Flow

Technical track

Language: Ukrainian.

Mykola Maksymenko, R&D Director at SoftServe, drives technological development in applied science and AI, human-computing interactions, and sensing. Mykola holds a Ph.D. in Theoretical Condensed Matter Physics, with over ten years of research experience, previously working at the Max Planck Institute for the Physics of Complex Systems and the Weizmann Institute of Science.

Quantum Computing promises a lot of potential to many computationally intense areas, but so far most of the novel algorithms are tested on classical simulators of quantum processors. Here Tensor Networks are one of the backend workhorses for efficient compression of quantum state and manipulations with quantum logic gates.

On the other hand, Tensor Networks can be seen as a new trainable Machine Learning object, in some cases being more expressive than Deep Neural Networks. It also allows to compress existing DNN architectures and speed up inference times.

I will outline what Tensor Networks are from physics perspective to Machine Learning and DNN compression and provide examples using TensorNetworks library on top of a Tensor Flow.

Mykola Maksymenko
R&D Director, SoftServeBridging Quantum Computing and Machine Learning in Tensor Flow
Justin Shenk
CTO, VisioLabAutomating Food Recognition for Canteens with Machine Vision

Business track

Language: English.

Justin has MSc in Cognitive Science with focus on AI (University of Osnabrueck, Germany) and in PhD Candidate (Radboud University).

Automated food checkout with machine vision will be the future of canteens. Our solution of a food recognition deep learning framework is designed to scale for hundreds of canteens and restaurants. Learnings gathered from our team in Kyiv and operations in German canteens will be shared.

Justin Shenk
CTO, VisioLabAutomating Food Recognition for Canteens with Machine Vision
Alexander Shevchenko
Lead Data Analyst, LetyshopsHow we perform AB-tests: methodology and organisation of work

Business track

Language: Ukrainian.

Bachelor of Science in Applied Physics. Graduated Taras Shevchenko National University of Kyiv, Radiophysics Department. Worked as Data Analyst in several leading Game Development companies in France, Germany and Ukraine (Ubisoft, Goodgames, Wargaming). Main qualification: marketing and product data analytics. Last year spent developing and supporting product AB-tests - which are the main topic of a presentation.

Company experience gained after split tests implementation will be presented - methodological and purely practical issues both with general assessment of this data-driven approach according to particularly Letyshops company experience.

Alexander Shevchenko
Lead Data Analyst, LetyshopsHow we perform AB-tests: methodology and organisation of work
Kostiantyn Radchenko
Product Analyst / Project Manager, LetyshopsHow we perform AB-tests: methodology and organisation of work

Business track

Language: Ukrainian.

Kostiantyn has a Master's Degree in Laser and Optoelectronic Engineering, is working on a PhD thesis, field of research is biomedical engineering, with a focus on developing a decision support systems. Data analytics is his passion. At LetyShops he works in the Partner Development department, conducts in-depth analysis and forecasting of the service users' behavior, their interests.

Presentation contains information about company organisational structure that supports Data Driven Development Method in dynamic environment with several simutaniously developed hypothesisys and several independed multidisciplinar teams working on one product.

Kostiantyn Radchenko
Product Analyst / Project Manager, LetyshopsHow we perform AB-tests: methodology and organisation of work
Vladyslav Horbatiuk
Vladyslav Gorbatiuk
ML Software Engineer, Snap IncChallenges in hands tracking on mobile devices

Technical track

Language: Russian.

Studied at NTUU ``Igor Sikorsky KPI``, faculty of Informatics and computer science, technical cybernetics department.

We would like to discuss challenges in hands tracking development on mobile devices.
What kind of challenges we faced
- Model criteria (сomputational budget, model size, accuracy, robustness, temporal consistency)
- Data (synthetic data, real data, combining different data)
- Product trade-off (use-case specific solution, different use-cases with examples, prior information)
We will talk about problems in detail and propose possible solutions.

Vladyslav Gorbatiuk
ML Software Engineer, Snap IncChallenges in hands tracking on mobile devices
Serhii Tiurin
Data Scientist, Data Science UAWorkshop: Creating web-app for DS stuff with Streamlit

Workshop

Language: Russian.

Serhii is a data scientist with deep business understanding. Worked on many ML/DL projects like: price optimization, sales forecasting, churn prediction.

Flask is not the only one!
So, you created your model in the Jupyter Notebook, what to do next? How to create an interface for it, so other people can use it as a web service?
Streamlit is an open-source library enables you to quickly turn pure Python scripts into bespoke apps without any ``app-building knowledge”. No need to write a backend, define routes and handle HTTP requests.
I`ll tell you how to create a web service for your data in hours, not days, using Streamlit.

Serhii Tiurin
Data Scientist, Data Science UAWorkshop: Creating web-app for DS stuff with Streamlit
Nikishaev
Andrey Nikishaev
System Architect, MoneyveoWorkshop: Object Detection with Single Shot Networks

Workshop

Language: Russian.

He graduated from Shevchenko university with a bachelor’s degree, specialty physical-mathematic.
He has 20+ years in IT. The last project was about the creation of a recommendation system (front, back, ETL, DevOps).

Before that, he did analytics of blockchain projects for investment funds and private investors in cooperation with BlockScience.

He was also engaged in developing recommender systems, development of information dissemination algorithms in social networks, automatic photo retouching, traffic analytics, style transfer grids and more.

Also, he recently launched a practical online course on Object Detection: https://learnml.today

Andrey will talk about the following at the workshop:
- The base ideas behind the object detection
- The difference between different network architectures
- Techniques used in training
- Train model

Andrey Nikishaev
System Architect, MoneyveoWorkshop: Object Detection with Single Shot Networks
Alexander Honchar
Co-Founder and CTO, Neurons LabWorkshop: Why your state-of-the-art ML can't predict markets

Workshop

Language: Russian.

Predicting financial markets with the help of mathematics and algorithms is a tempting idea that is making brightest academia minds and mediocre amateurs spending hours on searching the “Holy Graal”.

This workshop aims to give you a hands-on sneak peek into best practices in adapting financial data to “make the markets”. We will start with a “normal” machine learning baseline and step-by-step will improve data preparation, normalization, labeling, and validation in order to correspond to the harsh realities of the unpredictable financial world. You will be able to see, that the key lies not in the complexity of the algorithm, but in the profound understanding of the underlying data.

After the workshop, you will not just get practical skills with financial time series, but also will deepen your understanding of the machine learning pipelines, which can be helpful not only in finance but in healthcare, military, and other applications very sensitive to the failures.

Alexander Honchar
Co-Founder and CTO, Neurons LabWorkshop: Why your state-of-the-art ML can't predict markets

Hosts

Daniel Che
Comandante, Che – Guerrilla Marketing

Daniel Che
Comandante, Che – Guerrilla Marketing
Jane Klepa
Executive director, 1991 Open Data Incubator

Jane Klepa
Executive director, 1991 Open Data Incubator