The awarded team of the Information Technologies Institute (ITI) of CERTH: Nikolaos Gkalelis, Post-doctoral researcher. Mariano Ntrougkas, Research assistant. Vasileios Mezaris, Research Director
The paper authored by the researchers of the Information Technologies Institute of CERTH, Mariano Ntrougkas, Nikolaos Gkalelis and Vasileios Mezaris, titled “TAME: Attention Mechanism Based Feature Fusion for Generating Explanation Maps of Convolutional Neural Networks”, received the Best Paper Award of the 24th IEEE International Symposium on Multimedia (ISM 2022).
The awarded paper contributes to the domain of explainable artificial intelligence, proposing a new method for explaining the decisions of image classifiers that are based on convolutional neural networks. The proposed TAME method combines a target model’s feature maps from multiple layers using an attention mechanism, transforming them into an explanation map. After training the attention mechanism, explanation maps can be efficiently computed, requiring a single forward pass of the neural network.
This work was supported by the EU Horizon 2020 programme under grant agreement H2020-101021866 CRiTERIA. CERTH participates in CRiTERIA, which seeks to develop new, comprehensive and human-rights sensitive risk analysis methodologies for border protection, by developing explainable artificial intelligence methods.