Regularization for Inverse Problems and Machine Learning: A Tutorial

• Roberto Gutierrez Beraldo UFABC
• Leonardo Alves Ferreira Samsung R&D Institute Brazil
• Ricardo Suyama UFABC

Resumo

In this tutorial, we explore different paradigms for solving image processing tasks such as denoising and deblurring. These tasks can be interpreted as ill-posed inverse problems, grounded in physical-mathematical models that require regularization to produce meaningful solutions. Alternatively, they can be framed as supervised regression tasks within a machine learning context, where regularization techniques are used to improve generalization to unseen data. We compare these two perspectives, highlighting their similarities and dif-ferences-particularly in how regularization is understood and applied in each framework.

Palavras-chave: Graphics, Deblurring, Inverse problems, Image processing, Noise reduction, Tutorials, Machine learning