Deep Neural Networks for Regression Problems: From Data to Deployable Julia Models (1/3)

Participants will be guided through the full modelling workflow: from data preparation and exploratory analysis to neural network design, training, validation, tuning, model comparison, and final model selection.

The sessions will include demonstrations using app.samplingforml.edu.eu.org, an interactive tool for exploring sampling in machine learning, and training DNNs for regression effectively. Where appropriate, the series will also draw on PCA and heatmap-based exploration to show how data structure, sampling decisions, and variable relationships can influence downstream model performance.

Emphasis will be placed on the practical modelling choices that matter, including architecture design, activation functions, loss functions, optimisation, and evaluation of predictive accuracy. The series is intended for participants interested in correctly applying neural networks to continuous-outcome prediction problems in a transparent and practically useful way.