Seminar from Norbert Tsopze, Professor at Université Yaounde 1

 

Date: 13/05/2025
Time: 11h00
Place: IC2, Boardroom
Speaker: Norbert Tsopze
 
 

Machine Learning models explainability for text and audio classification

 

Abstract :

The interpretability of the Machine Learning models outcomes to the end user is one of the most important properties which favors the adoption of these models in various domains like medecine, law, … From the point of view of interpretability, Machine Learning models can be classified into two categories: interpretable models and “black-box” models. The explanability of the black box consists in the most of the cases in finding the similar behavior interpretable model, and interpret this model.

In the first part of this presentation, the problem of interpretability and explicability of models will be introduced; then some approaches to the explicability of “black box” models and the used evaluation metrics will also be explored. The second part focuses on a deep model (CNN+FCN) for explaining the classification outcomes. This part will mainly focus on two approaches we developed : the adapted version of the LRP algorithm to the convolution part of the model in order to find the the different n-grams responsible for classification and a prototype-based approach for classification. The presentation ends by the use of the SSL Wav2vec for improving Tone Recognition Performance in a Low-Resourced Language.

Biography :

Norbert Tsopze is a senior lecturer at the Department of Computer Science of the University of Yaounde I. His research interests include data mining, formal concept analysis, machine learning, text analysis, social network analysis, classification. He earned a Master’s degree in Computer Science from University of Yaounde I in 2004, followed by a PhD in Computer Science from the University of Yaounde I and the University of Artois (France) in 2010. From 2011-2012, he worked at University of La Rochelle (France) as a postdoctoral fellow.

 
ESPERANTO is a Research and Innovation Staff Exchange (RISE) program funded by the Marie Sktodowska-Curie Actions of the Horizon 2020 European Research Framework Program. This 5-year project started on January 1, 2021 and wil end on December 31, 2025.
https://cordis.europa.eu/project/id/101007666.