ISSN 2673-494X
Two-page Reports on Science

Cuban Sci. 2021, 2(2): 1–2

Published on 13 Sep 2021 PDF (167K)
Bidirectional Pooling for Deep Neural Networks
Marilyn Bello, Gonzalo Nápoles, Koen Vanhoof, Rafael Bello
  • Department of Computer Science, Central University of Las Villas, Cuba
  • Faculty of Business Economics, Universiteit Hasselt, Belgium
  • Department of Cognitive Science & Artificial Intelligence, Tilburg University, Netherlands
“This work introduces a new neural network architecture that uses bidirectional associations-based pooling to extract high-level features and labels from multi-label data. Unlike the pooling approaches reported in the literature, our proposal does not require input data to have any topological properties as typically occurs with images and videos. The numerical results show that our bidirectional pooling helps reduce the number of problem features and labels while preserving the discriminatory power of the network.”

References

  1. Goodfellow, I., Bengio, Y., Courville, A., Bengio, Y., Deep learning, MIT press Cambridge (2016)
  2. Charte, F., del Jesus, M.J., Rivera, A.J., Multilabel classification: problem analysis, metrics and techniques, Springer (2016)
  3. Gibaja, E., Ventura, S., Multi-label learning: a review of the state of the art and ongoing research, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 4 (2014) 411–444
  4. Yu, D., Wang, H., Chen, P., Wei, Z., Mixed pooling for convolutional neural networks, Lecture Notes in Computer Science, 8818 (2014) 364–375
  5. Bello, M., Nápoles, G., Sánchez, R., Bello, R., Vanhoof, K., Deep neural network to extract high-level features and labels in multi-label classification problems, Neurocomputing, 413 (2020) 259–270
  6. Hinton, G.E., Salakhutdinov, R.R., Reducing the dimensionality of data with neural networks, Science, 313 (2006) 504–507