Improving generalization is one of the main challenges for training deep neural networks on classification tasks. In particular, a number of techniques have been proposed, aiming to boost the performance on unseen data: from standard data augmentation techniques to the ℓ2 regularization, dropout, batch normalization, entropy-driven SGD and many more. In this work we propose an elegant, simple and principled approach: post-synaptic potential regularization (PSP). We tested this regularization on a number of different state-of-the-art scenarios. Empirical results show that PSP achieves a classification error comparable to more sophisticated learning strategies in the MNIST scenario, while improves the generalization compared to ℓ2 regularization in deep architectures trained on CIFAR-10.

Post-synaptic Potential Regularization Has Potential

Tartaglione, Enzo;Perlo, Daniele;Grangetto, Marco
2019-01-01

Abstract

Improving generalization is one of the main challenges for training deep neural networks on classification tasks. In particular, a number of techniques have been proposed, aiming to boost the performance on unseen data: from standard data augmentation techniques to the ℓ2 regularization, dropout, batch normalization, entropy-driven SGD and many more. In this work we propose an elegant, simple and principled approach: post-synaptic potential regularization (PSP). We tested this regularization on a number of different state-of-the-art scenarios. Empirical results show that PSP achieves a classification error comparable to more sophisticated learning strategies in the MNIST scenario, while improves the generalization compared to ℓ2 regularization in deep architectures trained on CIFAR-10.
2019
International Conference on Artificial Neural Networks
Munich, Germany
17-19/09/2019
Artificial Neural Networks and Machine Learning – ICANN 2019: Deep Learning. ICANN 2019
Springer
11728
187
200
978-3-030-30483-6
978-3-030-30484-3
Tartaglione, Enzo; Perlo, Daniele; Grangetto, Marco
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2318/1712221
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 4
  • ???jsp.display-item.citation.isi??? 3
social impact