Buy Video AI & BigData Online Day

2021 Autumn

3
Tracks

15
Hours of content

18
Videos

Conference program

Data Science Solutions

09:30 – 10:00

Registration

10:00 – 10:45

Як команда a-Gnostics металургам допомагала: прогнозування споживання природного газу (UA)

Yaroslav Nedashkovsky


1. Машинне навчання в металургії, актуальні завдання.
2. Ринок газу на добу вперед: чому важливо робити точні прогнози.
3. Збір даних для індустріальних завдань машинного навчання, відмінність від типових задач, підводні камені.
4. Побудова моделі в стрімко мінливих умовах виробництва.
5. Декомпозиція задачі: перехід від загального прогнозу споживання до суми прогнозів ключових споживачів.
6. Запуск моделі в промислову експлуатацію.

11:30 - 12:15

Challenges in real-world CV product (RU)

Haik Mherian


Зазвичай всі доповіді про нові типи нейронок, про деталі навчання їх і який крутий результат вийшов.

Проте мало хто розказує про те, звідки для реального проєкту взяти дані для навчання моделі, як багато їх потрібно в еру глибинного навчання. І про проблему синтетичних даних, як з нею боротись і т.д.

Ми з колегою зробимо доповідь по реальному проєкту, що недавно запустили в продакшин, поділимося своїм досвідом роботи в реальному житті.

Під час проєкту як раз ми зіштовхнулись з проблемою даних, і що на синтетичних даних нічого нормально не працювало і розкажемо, як важливі реальні дані і т.д.

Ось посилання про наш проєкт:
https://itc.ua/news/fozzy-group-pochala-klientske-testuvannya-shtuchnogo-intelektu-kissa-ai-yakij-zdaten-obslugovuvati-gostej-na-kasi-bez-dopomogi-lyudini/
https://www.youtube.com/watch?v=60FmYfO5_8c
https://rau.ua/ru/news/budushhee-segodnja-silpo/

12:15 - 13:00

Learning robot skills from video (UA)

Kateryna Zorina


Robotic systems are an essential part of the modern world. Robots assemble products on manufacturing lines, transport goods in warehouses, or clean floors in our living rooms. In recent years, many research efforts are dedicated to learning robot skills faster, more efficiently, and with a smaller amount of human supervision. In this talk, we will discuss state-of-the-art approaches for skill learning in robotics, and also I will share the current results of my ongoing research. We will explore how expert demonstrations are used for learning dynamic robotic skills by imitation. And how reinforcement learning is used for robot learning. Then I will present our approach, which aims to replace expert demonstration (recorded robot movement) with information extracted from videos of humans performing the same skill.

13:00 - 13:45

AI in Biotech: The Clash of Math Models and Reality (UA)

Oleksandr Gurbych


Modern AI/ML technologies were well advertised, but on average, only one out of five projects succeeds.

The situation in the life sciences is even worse because of the limited public data and the extreme complexity of biological objects and their interactions.

Let's consider the case where AI worked out
1) Intro
2) Vocabulary
3) History and Foreword
4) The Receptor and The Ligands
5) Methods
a) RLGNN
b) Molecular Dynamics
c) Wet Lab Experiment
6) Results
7) Conclusions

13:45 - 14:30

Using Cloud and Text Analytics to Gain Insights from COVID-19 Papers Corpus (RU)

Dmitry Soshnikov


In this session, we show how to leverage CORD dataset, containing more than 400000 scientific papers on COVID and related topics, and recent advances in natural language processing and other AI techniques to generate new insights in support of the ongoing fight against this infectious disease.

The idea explored in our talk is to apply modern NLP methods, such and named entity recognition (NER) and relation extraction to article's abstracts (and, possibly, full text), to extract some meaningful insights from the text, and to enable semantically rich search over the paper corpus. We first investigate how to train NER model using Medical NER dataset from Kaggle, and specialized version of BERT (PubMedBERT) as a feature extractor, to allow automatic extraction of such entities as medical condition names, medicine names and pathogens. Entity extraction alone can provide us with some interesting findings, such as how approaches to COVID treatment evolved with time, in terms of mentioned medicines. We demonstrate how to use Azure Machine Learning for training the model.

To take this investigation one step further, we also investigate the usage of pre-trained medical models, available as Text Analytics for Health service on the Microsoft Azure cloud. In addition to many entity types, it can also extract relations (such as the dosage of medicine provisioned), entity negation, and entity mapping to some well-known medical ontologies. We investigate the best way to use Azure ML at scale to score large paper collection, and to store the results.

14:30 - 15:15

3D modeling using Differentiable Programming (UA)

Andy Bosyi


How enhance CNN with complex activation function written using Differentiable Programming that helps the model to recognize patterns in the 3D world. Practical lecture with real data examples and explanations deep in the linear algebra and vectorization.

15:15 - 16:00

Working with BERT (UA)

Igor Lakoza

16:00 - 16:15

Conference Closing


AI Business

09:30 – 10:00

Registration

10:00 – 10:45

Чому створення data strategy для компаній
– це першочергове завдання? (UA)

Oleksandr Krakovetskyi


Переважна більшість компаній хоче впроваджувати ті чи інші елементи штучного інтелекту. Попри це, багато ініціатив закінчуються достроково через брак даних, відсутність централізованого сховища даних тощо. Тому для компаній є критично важливим розробка стратегії даних (data strategy), що дозволить значно пришвидшити реалізацію проєктів у сфері штучного інтелекту?

Невеличкий спойлер: чи є випадки, коли вам не потрібна стратегія даних? Звісно. Якщо у вас немає даних :-)

10:45 – 12:15

Panel Discussion: "Challenges to AI Adoption"

Рarticipants: Oleksandr Krakovetskyi, Yaroslav Nedashkovsky, Oleksandr Gurbych, Mykola Mykytenko

12:15 - 13:00

Дослідження екосистеми Data Science в Україні (UA)

Ross Chayka


  • Кількість спеціалістів в Україні, їх розподіл за професійним рівнем.

  • Кваліфікаційні рівні Junior, Middle, Senior зараз.

  • Вимоги до кожного кваліфікаційного рівня.

  • Університети та навчальні заклади з найбільшим вкладом.

  • Конференції, івенти та комьюніті майданчики.

  • Компанії в яких працюють Data Science спеціалісти, типовий розподіл.

  • Кар'єрні шляхи в Data Science.

  • Дослідження ринку зарплат зараз.

  • Суміжні спеціальності.

13:00 - 14:30

Panel Discussion "Ринок талантів DS в Україні"

Рarticipants: Oleksandr Krakovetskyi, Andy Bosyi, Dmitry Soshnikov, Michael Konstantinov

14:30 - 16:00

Panel Discussion "Nowdays and future of AI"

Рarticipants: Oleksandr Gurbych Dmitry Soshnikov, Andy Bosyi, Michael Konstantinov, Haik Mherian

16:00- 16:15

Conference Closing


Machine Learning

09:30 – 10:00

Registration

10:00 – 10:45

Zero-shot learning capabilities of CLIP model from OpenAI (UA)

Yurii Pashchenko


Let's talk about the CLIP model from OpenAI and why it can be considered the main network of 2021, using an example of its use in zero-shotlearning for classification, detection, segmentation and style transfer tasks.

10:45 – 11:30

Attention: Visual Transformers (UA)

Denys Drabchuk


RNN is dead long live the Transformers?

No! ConvNets are dead long live Visual Transformers!

  • Why CNN are good for Computer Vision?

  • The Dawn of Convolution from LeNet to EfficientNet.

  • Giant Steps: AlexNet, VGG, Inception, ResNet, MobileNet, NAS.

  • CNN for classification, detection, segmentation and other tasks (e.g. image captioning, etc).

  • BiT: Big Transfer.

  • Why CNN is bad for Computer Vision?

  • What is receptive field?

  • Attention is all you need!

  • From BiT to ViT.

  • An Image is Worth 16x16 Words.

  • Future of ViT.

  • From BYOL to CLIP.

  • Swin Transformer.

  • MLP, Convolution, Attention, Mixer.. What next?

11:30 - 12:15

How Neural Networks Think and What They See (RU)

Michael Konstantinov


1. Modern neural networks logic: from MLP to ViT.
2. Convolution vs Attention vs Mixer.
3. Level of abstractions in neural networks.
4. Visualization techniques.

5. Beyond neural visualization: Interpretation is All You Need.

12:15 - 13:00

Assiciative Memories, Hopfield and more:
from Attention to Biological neurons (RU)

Dimitri Nowicki


First in this talk, we will review the modern Hopfield networks with very high capacity and gradient-based update rule. We will show that such a nework can be equivalent to the Attention mechanism used in Transformers. Then we will introduce Kernel associative memory that reaches similar results with help of the "kernel trick". In final part of this presentation we will introduce recent results in reproducing detailed models of biological neurons with help of deep neurall networks (DNN) as well as capability of biologically plausible neurons to solve the tasks where large DNN is needed. The talk will be acompanied with the live demo of described networks and algorithms.

13:00 - 13:45

MLOps: your way from nonsense to valuable effect (approaches, cases, tools) (UA)

Mykola Mykytenko


Let's talk about how to build a process that will help the Data Sсience team quickly, successfully and painlessly place models in a productive environment, how to implement processes that will make the team work easier and faster, look at the tools, approaches and cases that allow you to achieve this.

14:30 - 15:15

Kubeflow for end2end machine learning lifecycle (RU)

Kyryl Truskovskyi


We are going to cover the machine learning model lifecycle, see what tools are used for each step of this journery. After that, we are going to explore some real-world examples and will deep dive into one of them with Kubeflow as a base.

15:15 - 15:30

Conference Closing


Organizer
Lemberg Tech Business School

Lemberg Tech Business School: organization with a 10-year history of successful conferences: Lviv Mobile Development Day, GameDev Conference, Lviv PM Day та Lviv Freelance Forum.