FIT-M REACTOR 2021

Master class plan:

• look at the application of serverless computing in practice
• build serverless capacity
• consider the advantages and disadvantages
• create an application in live coding mode
• deploy to the cloud and get a working prototype

Requirements: no special requirements.

Master class plan: 
1.what do you need to apply Data Science? 
2. What is Data Governance: assessment of the level of maturity 
3. Pitfalls in Building a Data Management Culture 
4. Practical cases of Data Science in FMCG

Master class plan:

• How We Lived Before Unified Approaches
• What were the problems
• Highlighting approaches to project management
• Establishing Basic Requirements for Entry Projects
• Integration into the process of the basic requirements for the input of projects
• Building a consistent CI / CD landscape
• Control over the CI \ CD landscape
• Creation of a description of the coverage options CI \ CD for the input of projects
• Possible future
• Summarizing

Requirements: no special requirements

Master class plan:

• Overview of changes in the IT industry associated with the emergence of clouds
• Evolution of approaches to infrastructure configuration management
• Cloud native approach
• Possibilities of public clouds for implementing a cloud native approach
• Strengths and weaknesses of the cloud native approach

Requirements: no special requirements

Master class plan:

• Comparative characteristics of external cloud solutions (aws, gcp, azure, Ya.Oblako)
• Comparison of tools for external and internal cloud solutions
• Determining the difference between tools and approaches to working with cloud solutions

Requirements: no special requirements

Master class plan:

• What is robotization
• What does AI have to do with it
• The brilliance and poverty of Computer Vision
• Demonstration of using Computer Vision to solve practical problems
• Demonstration of the use of neural networks for the classification of texts in robotic processing of calls
• Demonstration of robotic work with an online resource

Requirements: no special requirements

Master class plan:

• Find out why you need it
• Collect data from government contracts
• What data helps to recognize the context of the necessary information and collect it
• Design NER to extract the information you need
• Put everything in a structured form

Requirements: laptops with Anaconda Navigator installed, Internet access, proficiency in Python

Master class plan:

• log management, why is it needed
• elk-stack, scheme of work, scaling
• work with logs as with data, statistics

Requirements: confident use of OS Linux.

Master class plan:

• Introductory Presentation: Essential AWS Services for Running Machine Learning
• Data preparation
• Model training
• Expanding the Model to Production
• Using the model
• Completion and Next Steps: Custom Models, AutoML, Explainability, MLOps

Requirements: basic knowledge of machine learning algorithms, laptops with Internet access and a browser (preferably Google Chrome)

Master class plan:

• How we started solving MLOps problems in PropellerAds when there were no ready-made tools yet
• What tasks did we solve and in what ways
• What tools can and should be used if we started doing it now

Requirements: no special requirements.

Master class plan

• Iterativeness - about a flexible approach, prompt response to business requests
• Repeatability - about the ability to "return where it was" in case of malfunctions, ensuring a safe "walk in the swamp" during the constant evolution of solutions
• Isolation - as a tool to "stabilize" the information space in the absence of stability of tasks and hardware

Requirements: no special requirements

The master class will consist of two parts.

The first is a presentation. What is JupyterHub, what is Kubernetes, why is it worth running JupyterHub in Kubernetes, what features to consider.

The second part is a hands-on demo on a production Kubernetes cluster.

A public repository will also be available with all the commands, instructions and code for self-practice, reproducing all the steps shown in the practical demo of the master class.

Requirements: basic familiarity with linux or basic programming skills, or familiarity with the topic of data science / data engineering. Any basic skills in IT, or just an interest in this topic, will do.

This manual is intended for the basic user. Almost anyone who works with a computer will be able to reproduce all the steps and gain valuable experience with Kubernetes, clouds and JupyterHub.

Master class plan:

As part of the master class, we will talk about the capabilities of neural network models for natural language processing and the use of cloud services to extract entities and relationships from the corpus of scientific articles on COVID-19. This approach will allow us not only to carry out semantic queries to the corpus, but also to extract and visualize the change in treatment tactics over time, and to discover key patterns in treatment tactics. The master class will be held in a demonstration mode, but all participants will have access to materials for subsequent independent practice.

Requirements: Basic Python Programming Skills

Master class plan:

• AI trends
• AI Applications in Cybersecurity
• Cyberthreats and risks associated with AI
• Cyber ​​threats and risks in relation to attacks on AI
• How AI learning models and security fit together
• Protecting AI from cyberattacks

Requirements: no special requirements.

Master class plan:

• What are Cybertwins
• How the concept of cyber twins can be used to defend against cyber attacks
• Modeling attack vectors and defense systems using cyber twins
• Tracking attackers with cyber twins
• Cyber ​​risk management with cyber twins

Requirements: no special requirements.

Master class plan:

I will tell you what opportunities are provided by modern cloud platforms that can be used in research. Let's review popular cloud providers, compare their capabilities and the most interesting products in the context of science. Let's look at a few specific examples. Let's look at what conditions (grants, discounts) are provided to scientists.

Requirements: no special requirements.

Master class plan:

• Consider delivering secrets via Bank-vaults
• Deploy an application with static and dynamic secrets
• Consider how to differentiate access and bypass restrictions
• Let's look at how to deliver secrets via Vault Agent

We will need:

• Workshop repository
• K8S cluster (minikube or any other cluster available to you, where you have admin access)
• kubectl
• helm
• jq
• curl
• GNU sed (by default on Linux, on mac os you need to install it separately, you can use brew)
• GNU make (optional)
• Docker (optional)

What will we do:

• Start an instance of a volta in a cube
• Configure it
• Deploy Banzaicloud Vault Secrets Webhook
• We will deploy the application several times, including several applications with different access in the same namespace, and consider how to transfer static secrets (KeyValue), as well as dynamic (DB Secrets Engine)
• Let's look at how you can transfer secrets to an application using the Vault Agent

introductory presentation https://docs.google.com/presentation/d/1o8Xn1594GnvgxjgCGf3NWmnIk_w-dCx7JADnrk5amaw/edit?usp=sharing"

Master class plan:

• Cloud-Native architecture
• Protected Objects
• Threats
• Attack surface
• DevSecOps architecture
• Secure development process as the basis for attack surface reduction
• Run-Time Protection

Requirements: Understanding the architecture of cloud-native applications.

Master class plan:

1) Introductory part:

• The current state of AI in insurance;
• For what tasks are AI models in demand now?
• The most popular AI models, a description of how they work

2) Practical part

• using Random Forest to select the most significant parameters affecting the target variable (using the example of data from open sources, most likely Rosstat)

Requirements: Laptops / PCs with Anaconda Python 3.8 package preinstalled, knowledge of the basics of statistics and Python programming will be enough for the listeners

Master class plan:

• Introducing the Unreal Engine
• Features of VR development
• Developing a simple game
• Demonstration of the result to those who wish

Requirements: asic programming skills in any language.

Master class plan:

• We will take an example of solving the problem of synthesis of a self driving car control model using reinforcement learning and model evolution
• This sketch will be told, the results will be analyzed
• There will be a small overview of similar approaches from the academic environment
• source codes will be given to participants for creativity and revision
• Everyone will be able to start and try to change the project so that its final model overtakes others in the validation run

Requirements: no special requirements.

Seminar annotation

The full-scale digitalization of housing and communal services today has certain limitations. The main part of the efforts of resource supplying organizations to modernize communal infrastructure is primarily aimed at operating activities, updating and building networks and equipment. “Digit” is a difficult story, but there are simply no other alternatives for housing and communal services. Sustainable development of the industry is possible only when we achieve its complete controllability and transparency, including through digitalization

Master class plan:

• Formulation of the problem
• Introducing the Jetson Nano
• Introducing Google Colab
• Introducing the Nvidia TAO Toolkit framework
• Working with dataset
• Model training in Google Colab
• Transfer and launch of the model on the Jetson Nano
• Self-learning algorithm on the Jetson Nano
• Summary statistics
• Results of the project, its limitations and development

Requirements: laptop, knowledge of basic Python, Google account, knowledge of the basics of Computer Vision, knowledge of the basics of ML, if there is a Jetson nano, of any version, we can run the resulting code on the sbc brought

Master class annotation:

It just so happened that in recent decades, the term AI has been associated mostly with technology. The embodiment of this approach is the “power race” of modern deep neural network algorithms such as GPT-3, WuDao etc. At the same time, scientific advances in various fields of knowledge from linguistics to medicine are used by these technologies in part as starting points for building models. Can modern machine learning models, in particular deep neural networks, qualitatively enrich scientific knowledge while being themselves a poorly theorized scientific field?

Requirements: no special requirements

Master class plan:

Let's understand the essence of some services as service. This includes an immersion in theory with a focus on practical examples, which the author has had the privilege of doing.

Session is divided into three parts:

• We will analyze all the basic infrastructure elements that are extremely necessary for as service - from APA GateWay, through infrastructure, code, to Databases
• Consider a number of solutions from the point of view of OpenSource and what it gives and what takes away
• For those who remain on the practical side, I am ready to propose together to solve a relatively practical problem: without touching the coding (i.e., logically), implement several business functions that the very popular proxy NGINX cannot solve, but which can be solved by using not only this solutions and of course take advantage of OpenSource properties

Requirements:

• Become familiar with what are called infrastructure components (example: Databases, Operating System, Proxy)
• Learn about the term OpenSource
• Become familiar with the term Cloud, IaaS, SaaS and the concept of as service in general

Master class plan:

• Overview of the kubernetes platform and the complexities of its operation
• Kubernetes lightweight solutions for local operation
• Installing a local platform
• Configuring local nodes and launching applications

Requirements:

• Computers with linux or macs, 4+ RAM, a virtual box installed, the Internet (this is in case the participants will also be doing work)
• Experience with linux and console is desirable

Master class plan:

• Introductory part, problems
• overview of multiple branching models \ strengths and weaknesses
• TBD review: how it works, who suits, what kind of exhaust
• feature flags
• flag types
• features of different environments
• AB flags
• conclusions

Requirements: the master class will be of interest to children working in teams, regardless of grade. The main value is to tell the details of the approach, and how to reduce the problems of interaction within the teams.

Master class annotation:

You have a small team that you are doing The Project. Who will change the world.

As part of this project, you need to create an internal service in Django: show statistics, manage internal users, and something else.

At first it looks like a cool idea, but after six months it turns into a problem: something constantly does not work, the code is incomprehensible, if you change it here, it breaks there. A year later, this service turns into a concrete disaster: it sometimes works, no one knows how, no one understands how to fix it. The salvation of the world is in the balance.

In the master class, we will talk about how to avoid this without a separate development team to support the service and look at some aspects of this in practice

Requirements: be familiar with Django, preferably in practice

Master class plan:

• First, we’re going to discuss the general concepts of NLP problems and how to solve them. Let's talk about the features of using NLU in real products
• We will implement a simple semantic search using the BERT + Nearest neighbor search link and discuss for which products you can create a baseline based on it
• Let's wrap our pipeline into a service that anyone can use. In the end, I will tell you how you can improve the results obtained and optimize the inference, I will share useful links

Requirements: knowledge of python, basic understanding of NLP and language models based on transformers

Master class plan:

• How to use bootstrap correctly
• How to parallelize computations
• What does the Poisson distribution have to do with it?

Requirements: no special requirements

Process Mining and Big Data. How to quickly find user churn points 

18 Dec, room R-206 16:20-17:20

Alexander Tolmachev

Master class plan

• graph theory in the analysis of user behavior logs
• modern libraries for analyzing user behavior processes based on logs, pros and cons in the benchmark
• practical cases of analyzing user behavior and the effect for business
• Answers on questions \

Requirements:

• github registration
• add Alexander Tolmachev's account to subscribers
• knowledge of probability theory at least some, the ability to analyze data, the ability to write in python
• install libraries: pm4py, collections, pandas, numpy, IPython, dateutil, sklearn, plotly, ipywidgets