Starting from a small machine learning (ML) experiment or a proof of concept (PoC)—down to the production system—an ML solution lifecycle and infrastructure cover much broader space than just machine learning model code—often consisting of multiple stages and workflows, different building blocks, and numerous components.

Building, training, deploying, and maintaining ML models in production requires well-established processes and workflows. Simultaneously, machine learning introduces fundamentally new challenges for traditional SDLC and CI/CD lifecycles due to its “data-driven” nature that defines the behavior of the system:

• Training production-grade ML models is an iterative process that involves extensive experimentation and prototyping
• ML-specific operations that depend on the data (i.e., data versioning, feature extraction, model training, evaluation, tuning, and serving)
• A complex landscape of various ML tools, libraries, frameworks, platforms, and hardware accelerators
• Production ML deployments require constant monitoring, quality control, and the ability to debug and interpret critical issues
• Success depends on the cooperation of multiple teams and stakeholders with poorly separated responsibilities and different workflow speeds

Amazon SageMaker eliminates most of these challenges by providing a fully managed ML infrastructure, tooling, and AutoML capabilities that empower state-of-the-art ML solution delivery in a time-efficient manner, and with minimal effort. AWS SageMaker is an end-to-end solution that assists during all stages of the ML model lifecycle:

• Acquire (and label) training data using Amazon SageMaker Ground Truth
• Prototype, build, and deploy ML models in interactive Jupyter environment provided by Amazon SageMaker Notebooks
• Leverage Amazon SageMaker Autopilot to automatically build high-performance models while maintaining full control and visibility
• Manage and track ML experiments using Amazon SageMaker Experiment
• Analyze ML models to detect issues and problems via Amazon SageMaker Debugger
• Monitor production ML deployments using Amazon SageMaker Model Monitor

In this session, SoftServe will share design recommendations and best practices in building large-scale machine learning systems on Amazon SageMaker using practical examples and real-life use cases. You will learn how to address modern business and technical challenges as well as how to bridge the gap between data, science, IT, business stakeholders, and end-users.

Iurii Milovanov – FEATURED SUBJECT MATTER EXPERT

Mr. Milovanov is a data science practice leader with more than 10 years of experience in building enterprise-level AI/ML, big data, and advanced analytics solutions.

He is a computer science expert with a strong emphasis on cutting-edge technologies. His research interests include various aspects of modern, progressive IT, and state-of-the-art AI, such as distributed and parallel computing, large-scale ML, natural language processing (NLP), computer vision, and speech recognition.

Yuri actively contributes to various research and scientific communities, including: the KarooGP project—a genetic programming suite used at LIGO Lab for detecting gravitational-waves, SIMOC—an interactive model of a scalable, human community located on a remote planet, and the DRLearner project—the first open source implementation of Google’s Deep Reinforcement Learning (DQN) algorithm for playing ATARI games.