Víctor Campos, BSC researcher at the Autonomic Systems and e-Business Platforms group, has been awarded with the best MSc thesis 2018 by the Catalan Association for Artificial Intelligence at its 21st International Conference (CCIA 2018) .
Campos is currently pursuing his PhD on the intersection between Deep Learning and High Performance Computing at BSC, supported by Obra Social "la Caixa" through La Caixa-Severo Ochoa International Doctoral Fellowship program. His awarded thesis is titled "Learning to Skip State Updates in Recurrent Neural Networks". Recurrent Neural Networks (RNN) provide state-of-the-art results in most tasks involving sequences, but they involve sequential computations that may result in slow predictions. In his thesis, Campos presents Skip RNN, a novel RNN architecture that learns which elements in the input sequence can be omitted, providing speed improvements without damaging the accuracy of the model.
This work has been developed in collaboration with researchers at Google, Columbia University, UPC and BSC. The project was accepted in the International Conference on Learning Representations (ICLR) 2018, one of the most relevant conference on deep learning.
Further information about the project here.