Skip to main content
Login | Suomeksi | På svenska | In English

Unsupervised learning for image classification

Show full item record

Title: Unsupervised learning for image classification
Author(s): Lu, Yao
Contributor: University of Helsinki, Faculty of Science, Department of Computer Science
Discipline: Computer science
Language: English
Acceptance year: 2016
Abstract:
This thesis is an investigation of unsupervised learning for image classification. The state-of-the-art image classification method is Convolutional Neural Network (CNN), which is a purely supervised learning method. We argue that despite of the triumph of supervised learning, unsupervised learning is still important and compatible with supervised learning. For example, in the situation where some classes have no training data at all, so called zero-shot learning task, unsupervised learning can leverage supervised learning to classify the images of unseen classes. We proposed a new zero-shot learning method based on CNN and several unsupervised learning algorithms. Our method achieves the state-of-the-art results on the largest public available labelled image dataset, ImageNet fall2011.


Files in this item

Files Size Format View
yao_thesis.pdf 3.653Mb PDF

This item appears in the following Collection(s)

Show full item record