Deep Convolutional Neural Networks as Generic Feature Extractors

Lars Hertel, Erhardt Barth, Thomas Kaster, Thomas Martinetz

Abstract

Recognizing objects in natural images is an intricate problem involving multiple conflicting objectives. Deep convolutional neural networks, trained on large datasets, achieve convincing results and are currently the state-of-the-art approach for this task. However, the long time needed to train such deep networks is a major drawback. We tackled this problem by reusing a previously trained network. For this purpose, we first trained a deep convolutional network on the ILSVRC-12 dataset. We then maintained the learned convolution kernels and only retrained the classification part on different datasets. Using this approach, we achieved an accuracy of 67.68% on CIFAR-100, compared to the previous state-of-the-art result of 65.43%. Furthermore, our findings indicate that convolutional networks are able to learn generic feature extractors that can be used for different tasks.

Original languageEnglish
Title of host publication2015 International Joint Conference on Neural Networks (IJCNN)
Number of pages4
PublisherIEEE
Publication date28.09.2015
ISBN (Print)978-1-4799-1961-1
ISBN (Electronic)978-1-4799-1960-4
DOIs
Publication statusPublished - 28.09.2015
EventInternational Joint Conference on Neural Networks 2015 - Killarney, Ireland
Duration: 12.07.201517.07.2015
https://ieeexplore.ieee.org/document/7280683/

Fingerprint

Dive into the research topics of 'Deep Convolutional Neural Networks as Generic Feature Extractors'. Together they form a unique fingerprint.

Cite this