Machine Learning: Fine-Tuning and Cross-Validation

Training a model is only the beginning. In practice, producing reliable and generalizable machine learning systems requires a deep understanding of why models succeed — and why they fail. This course moves participants beyond the basics of model fitting, equipping them with the theoretical foundations and practical tools needed to optimize, regularize, and rigorously evaluate machine learning models. From implementing gradient descent to constructing end-to-end Scikit-learn pipelines, participants develop the critical mindset and technical rigor that advanced machine learning practice demands.

While Machine Learning: Fine-tuning and Cross-validation can be taken as a standalone course, it is also part of EPFL’s learning pathway in applied machine learning. It is the second of four Microcredentials leading to the Certificate of Open Studies (COS) in Applied Machine Learning https://www.formation-continue-unil-epfl.ch/formation/applied-machine-learning/, providing a flexible pathway toward an EPFL credential in applied machine learning.

Upon successful completion of this course, participants will be able to:

Implement the gradient descent optimisation algorithm from scratch using Python
Standardise features to ensure efficient convergence of gradient-based optimisers
Analyse model complexity by decomposing errors into bias and variance components
Apply ridge and lasso regularisation techniques to manage overfitting and improve generalisation
Execute robust model evaluation using nested cross-validation and grid search strategies
Implement end-to-end machine learning pipelines using Scikit-learn pipelines and custom transformers

Nongraded

Data analysts and quantitative analysts looking to bridge the gap to machine learning; research scientists who need advanced, custom predictive frameworks; and software engineers transitioning into the machine learning and AI space to build real-world applications.