Professors
- Beat Wolf (beat.wolf@hes-so.ch)
- Bertil Chapuis (bertil.chapuis@hes-so.ch)
Academic Calendar
Automn Semester 2024
Abstract
This module presents powerful techniques to manage the lifecycle of machine learning models, covering in particular baseline models, infrastructure (clusters, cloud, edge AI and resource management) and tooling (frameworks), model training and debugging, model evaluation and tuning, data management (sources, storage, versioning, privacy), systems testing (CI/CD) and explainability, deployment (batch, service, edge), monitoring (data drift) and continual learning. Emphasis is placed on practical tools, real use-case scenarios, and the relevant hardware and software platforms.
Additional topics such as business requirements and objectives, project management for ML, team structure, user experience as well as responsible use of ML systems, including sustainable AI, are also considered.
Prerequisites
- Basic knowledge of machine learning, deep learning, data management and data engineering.
- Good command of an imperative programming language, basic knowledge of Python.
- Basic knowledge of probability, statistics, linear algebra (vectors, matrices).
Learning Objectives
- Recognising the complete lifecycle of machine learning projects, from data requirements to development, deployment, and monitoring.
- Demonstrating skills in maintaining ML code and data, version and integrate it, and define appropriate environments, with emphasis on practical applications such as data cleaning and preprocessing.
- Deploying ML models at scale, monitoring their performance and adapting models to changing requirements, with a focus on assessing and adjusting to data drift and shifts in data distribution.
- Analysing relevant tools and real use-case scenarios, such as real-time services management; critically analysing the implications and applications in practical scenarios.
- Selecting software and hardware platforms based on the requirements of different scenarios, demonstrating a thorough understanding of the needs and constraints of each.
- Extracting and integrating insights from guest lectures by industry professionals (subject to availability), demonstrating the ability to interpret expert knowledge from scientific literature and online resources, and applying it effectively to complement their hands-on experience.
Contents of Module
- Brief recap of machine learning and deep learning.
- Introduction to the lifecycle of a Machine Learning project.
- Understanding data needs and requirements for ML projects (e.g. versioning, storage, processing, labeling, augmentation, simulation).
- ML Development: defining the environment, maintaining the ML code, integrating ML code (versioning, evaluation, baselines).
- ML Deployment: running models at scale (e.g. batch vs online, model compression, cloud / edge deployment), ensuring system availability, monitoring performance, adapting to changes (data distribution shifts, failures, metrics, logging, continual learning).
- Exploration of tools and real-world scenarios.
- Overview of relevant hardware and software platforms.
- Selection of advanced topics such as:
- Trustworthy AI (incl. regulatory aspects).
- Guest lecture(s) from industry professionals (subject to availability).
- Project management and business perspective (e.g. job roles, teams).