TFIPDL: Development of new information fusion techniques in deep learning processes

Start date: December 1, 2018
End date: November 30,2019


The general objective of the project was to improve the applicability and interpretation capacity of deep neural networks by developing new information fusion mechanisms in the convolution and pooling phases.

During the project, new information fusion mechanisms have been developed applicable in the pooling and convolution stages of a deep neural network. These functions make it possible to optimize the operation of the network, adapting to the specific characteristics of each problem to be treated and allowing a certain interpretability of the results; that is, a certain understanding of why a given answer or solution is the one obtained.

In addition, these mechanisms are computationally inexpensive, which allows their effective integration in the learning process of the network. As an application of the developments, the mentioned mechanisms have been incorporated into obstacle detection devices for personal mobility vehicles.


Phd. Angela Bernardini, coordinator of the Artificial Intelligence & Data Analytics area at NAITEC




Project financed by the Government of Navarra through the 2019 call for grants to Technology Centers and Research Organizations for carrying out collaborative R&D projects.