Ecologia e sostenibilità ambientale

1 CFR, 5 ore - I semestre

Docente invitato

Salvatore Virdis | Asian Institute of Technology, Thailand

Orario

7-15 gennaio 2027, TBC

Aula

TBC

Salvatore Virdis | Asian Institute of Technology, Thailand

Alghero
7-15 gennaio 2027 - TBC

This 5-hour lecture introduces the fundamental concepts of artificial neural networks through a practical regression problem. It is designed for MSc and PhD students working with environmental and geospatial data who may have limited previous experience with neural networks or machine learning.
Starting from familiar concepts of linear and multiple regression, the lecture gradually introduces the main components of a neural network, including inputs, weights, bias, activation functions, hidden layers, outputs, loss functions, and model training. A daily environmental time series will be used as a practical example, with five climatic variables serving as predictors of Lake Surface Water Temperature (LSWT).

Microsoft Excel will be used as the main learning environment, not as a specialized neural-network platform, but as a transparent way to expose the calculations that normally remain hidden within machine-learning software. Students will be able to follow, step by step, how input variables are combined with weights, how neurons generate outputs, how predictions are produced, and how errors are calculated. Model training will then be illustrated by optimizing the network weights to minimize prediction error.

The emphasis of the lecture is therefore on understanding the logic behind neural networks rather than on programming or software-specific implementation. By the end of the session, students should be able to explain how a basic feed-forward neural network operates, how it extends conventional regression, and how it can be applied to environmental prediction problems.