2025-09-30 –, Louis Armand 1 - Est
Domain adaptation addresses the challenge of applying ML models to data that differs from the training distribution—a common issue in real-world applications. SKADA is a new Python library that brings domain adaptation tools to the sci-kit-learn and PyTorch ecosystem. This talk covers SKADA’s design, its integration with standard ML workflows, and how it helps practitioners build models that generalize better across domains.
Real-world data is messy, and models often break when deployed on data that differs from their training distribution. This is where domain adaptation becomes essential—and yet, it's rarely easy to implement or benchmark in practice. Enter skada, an open-source Python package that brings powerful domain adaptation tools into the Scikit-Learn and PyTorch ecosystems with a consistent, modular API.
No prior knowledge of domain adaptation is required, but familiarity with scikit-learn, skorch and supervised learning will help.
Outline
- 0–5 min — What is domain adaptation, and why does it matter?
- 5–15 min — Live demo: using SKADA to fix a model that breaks under domain shift
- 15–20 min — Tour of SKADA’s API: estimators, pipelines, benchmarks
- 20–25 min — Deep dive: integrating SKADA into torch and skorch
- 25–30 min — Q&A and tips for going further
Takeaways
- Understand what domain adaptation is and why it is crucial
- Learn how to use SKADA with familiar tools (
Pipeline,GridSearchCV, etc.) - Leave with working code examples and links to tutorials and benchmarks
Audience
This session is ideal for data scientists, researchers, and engineers interested in open-source tooling, model robustness, and practical ML deployment.
I'm a third-year PhD student working on domain adaptation for biosignal applications.
Postdoctoral researcher at Inria Saclay