Machine Learning into Practice: Deep Dive into MLOps
This foundational course offers a comprehensive journey through the various stages of deploying and maintaining machine learning models to applications using the MLOps paradigm.
MLOps is a paradigm that aims to deploy and maintain machine learning models in production reliably and efficiently. The word is a compound of "machine learning" and the continuous delivery practice (CI/CD) of DevOps in the software field.
Praktische info:
Leertraject
Inschrijven?
- Voorwaarden: programming skills (preferably Python), foundational understanding of machine learning concepts
- Prijs: €400-€1200
Bring your own laptop, with:
- enough power (minimum 8GB RAM)
- administrative rights to install the necessary programs
Classes:
- 16 December 2024 (9h -17.30h)
- 17 December 2024 (9h - 17.30h)
- 18 December 2024 (9h - 16h) + closing reception
georganiseerd door:
Through hands-on workshops participants will gain insights into the core steps of MLOps: Data preparation and versioning, model deployment, monitoring, scaling, and continuous training. They will understand the significance of having a clear understanding of what to expect in real-world scenarios when deploying a machine-learning model.
Additionally, this workshop covers the specific challenges of deploying LLMs and RAG solutions. We conclude the workshop with techniques for downscaling models to edge devices for real-time processing.
Throughout the course, participants will acquire practical skills and knowledge essential for navigating ML deployment smoothly, empowering them to face various real-world challenges.
Programme
Day 1 - 16 December 2024
Workshop: Docker & Kubernetes (4 x 1,5h)
This workshop will give you practical hands-on experience with using Docker and Kubernetes for MLOps. It will also give you the theoretical foundations to make you comfortable with relying on Docker and Kubernetes for MLOps.
- Docker basics.
- Using Docker for ML training.
- How to migrate from Jupyter Notebooks to production-ready docker containers.
- GPU acceleration in Docker containers.
- How to deploy production ML models on Kubernetes.
- Upgrading production ML models in Kubernetes.
Teachers: Merlijn Sebrechts & Sander Borny (IDLab, Ghent University, imec)
Data logging for industrial edge devices (1h)
Having a good data logging setup is crucial for developing industrial machine vision algorithms that run on the edge.
In this sessions we present how we set up our current projects and show several use-cases to demonstrate the effectiveness.
Teacher: Gilles Ballegeer (VINTECC)
Day 2 - 17 December 2024
Applying DevOps to Machine Learning (4h)
In this hands-on workshop we offer an introduction to MLOps, focusing
on how DevOps practices can be applied to machine learning. Together,
we will work through a mini project where we will deploy, serve, scale,
and monitor a machine learning model in production. The session covers
key topics such as model deployment, versioning, scaling, and continuous
training. It also addresses important aspects of monitoring and dealing
with concept drift. Depending on time and interest, optional topics
like A/B testing and LLMOps may also be included. Throughout the
workshop, participants will be introduced to recommended tools and best
practices relevant to MLOps.
Teacher: Cedric De Boom (Dataminded)
AI assistants: from POC to production (3h)
AI assistants are being widely adopted in businesses as a means for
efficient context-aware interaction. While setting up a custom AI
assistant is not rocket science, different challenges exist to get it up
and running in a production environment. In this workshop we will
discuss these challenges and set up our own RAG-based solution in the
cloud. We will zoom in on tools and frameworks that can be used for
traceability and automated quality monitoring.
Teacher: Dimitri De Rocker (Datashift)
Day 3 - 18 December 2024
ML model optimizations for efficient edge AI deployment (1h)
State-of-the-art machine learning models for image recognition and
natural language processing require a huge amount of computational
resources that are not available on resource constrained edge devices.
In this lesson, we will discuss the different options that are available
to optimize these models, reducing their computational cost and memory
footprint, making it possible to use them on mobile and embedded
devices.
Teacher: Sam Leroux (IDLab, Ghent University, imec)
Fine-Tuning Small LLMs for PII Detection and Secure Edge Deployment (3h)
You wil learn how to fine-tune small language models for detecting
Personally Identifiable Information (PII), with a strong focus on data
privacy and secure edge deployment. You'll learn to generate synthetic
data for training without exposing real sensitive information and
perform local fine-tuning entirely on-device to eliminate cloud-related
risks. We'll cover best practices for securely deploying your model on
edge devices, including implementing encryption methods. The session
also explores testing strategies to validate your model's PII detection
accuracy while upholding security standards. By the end, you'll be
equipped to build and deploy a privacy-focused AI application that
operates offline on an edge device, safeguarding sensitive information
throughout its lifecycle.
Teacher: Joseph Miano (Superlinear)
Introduction to Federated Learning with Python & Flower (2h)
In this session you will be introduced to Federated Learning in a
practical way, a technique that allows you to train machine learning
models without centrally storing the data. We will cover the basic
principles, look at available frameworks and discuss the challenges of
this technology. In the second, hands-on part you will get started with
the Flower framework and Python to apply this knowledge in practice.
Teacher: Thomas Van den Bossche (Odisee)
Lesgevers / sprekers
PhD holder and certified cyborg. I enjoy weird food, hiking, and games with a good story. I want to make the world a better place.
I’m a senior researcher at imec and teaching fellow at Ghent University in the IDLab research group in Belgium.
Sander Borny received the M.Sc. degree in information engineering technology from Ghent University, in 2016. He is currently a Research Assistant with IDLab, Ghent University - imec. His focus is providing cloud infrastructure, DevOps and DataOps to various projects in the fields of smart cities, smart grids, and the Internet of Things.
Machine Vision Engineer at VINTECC
Responsible for full project completion, from initial steps in data gathering to production deployment. Projects mostly consist of a deep learning stack followed with edge deployment.
Senior Data Engineer @Lighthouse
- Data specialist
- PhD in Artificial Intelligence
- Passionate and certified teacher
- Allround pianist
AI Director at Datashift
Datashift specializes in artificial intelligence, business intelligence, data science and data governance. It helps companies across industries to find and unlock the power of their data and to transform it into business value. It enables a shift from intuition to understanding. From understanding to action. From action to impact.
Datashift makes sense of data.
Sam Leroux
Dr. ing. Sam Leroux behaalde in juli 2014 summa cum laude een Master of Science diploma in Information Engineering Technology aan de Universiteit Gent, België. In september van dat jaar ging hij aan de slag bij het departement Informatietechnologie aan de Universiteit Gent. Na het behalen van zijn doctoraat in 2019 werkt hij als postdoctoraal onderzoeker in het Distributed Machine Learning team bij IDLab Gent. Zijn belangrijkste onderzoeksinteresses zijn energie-efficiënt deep learning, edge computing en Tiny ML. Sam Leroux geeft ook verschillende cursussen in zowel de bachelor- als de masteropleiding Information Engineering Technology.
Expertise: Machine learning, Deep learning, Edge computing, resource efficient machine learning, TinyML
I am a data scientist and machine learning engineer with 5+ years of experience building models for computer vision, natural language processing, and tabular datasets. With a B.S. in neuroscience and an M.S. in computer science, I am especially excited about the development of neural networks and the increasing complexity of problems they can solve.
Over the past several years, I have had the opportunity to work on analytics for large-scale medication adherence outreach programs, multi-task neural networks for brain microscopy image segmentation, transformer-based NLP models to detect COVID-19 outbreaks from news articles, model explainability in credit risk assessment, machine learning for fraud detection, and more.
Thomas doet aan Odisee Hogeschool toegepast onderzoek naar Data & AI. Zijn centrale vraag: hoe zetten organisaties AI op een verantwoorde en efficiënte manier in? Als medeoprichter van Webstrive combineert hij dat onderzoek met de dagelijkse praktijk, en vertaalt hij complexe AI-concepten naar concrete toepassingen voor de werkvloer. Hij heeft daarbij oog voor strategie, privacy en de noden van de eindgebruiker, in lijn met GDPR en de EU AI Act.
Daarnaast geeft Thomas workshops die de volledige cyclus van een AI-project doorlopen: van strategische sessies over prototyping tot validatie. In de prototyping-workshops bouw je in enkele uren een werkend prototype met vibe coding-tools zoals Lovable, v0 en Bolt. In andere sessies ontdek je hoe AI-agents je dagelijkse werk versnellen: Claude Code voor wie ontwikkelt, Claude Cowork voor wie documenten en bestanden beheert.
Thomas verzorgde al opleidingen en workshops voor onder andere UGent, VOKA, Unizo, Embuild en Etion. Elke sessie wordt afgestemd op sector en kennisniveau, gaande van een eerste kennismaking met prompt engineering tot meerdaagse trajecten waarin teams hun eigen AI-case uitwerken.
