Divulgação - Defesa Nº 266

Aluno: Vinícius Oliveira Barros

Título: “Interpretability of an Automatic Handwritten Signature Verification Model”

Orientador: Byron Leite Dantas Bezerra - (PPGEC)

Examinador Externo: Cleber Zanchettin - (UFPE)

Examinador Interno: Diego José Rátiva Millán (PPGEC)

Data-hora: 31 de Março de 2023, às 08:30h.
Local: REMOTO (https://meet.google.com/jnv-dhrg-tcz)


Resumo:

         In handwritten signature verification, convolutional neural networks are used in many different configurations and produce vastly different albeit satisfactory results when extracting signature features. One problem posed by the usage of these models is that, depending on the application, the need for insight into what features were used to output a given feature sets a barrier to the usage of those networks. In this work, we use Integrated Gradients and Saliency maps to extract regions of relevance from a sample input signature. We also analyze those regions where signature features such as initial and final pen strokes, pressure, speed, and connections, among others, to investigate if a proposed feature-extracting network points to the same criteria used by handwritten signature verification experts. Our experiments show that the initial and final pen strokes are the most commonly-occurring, high-relevance feature region outputted by the selected attribution algorithms with a relative frequency of 66.6%. These results relate to those obtained by the study of the criteria chosen by experts in a similar setting.

Go to top Menu