Joint Constrained Clustering and Feature Learning based on Deep Neural Networks

Author: 
Date created: 
2017-08-23
Identifier: 
etd10355
Keywords: 
Constrained Clustering
Feature learning with Convolutional Neural Networks
User Feedback
Active Learning
Abstract: 

We propose a novel method to iteratively improve the performance of constrained clustering and feature learning based on Convolutional Neural Networks (CNNs). There is no effective strategy for neither the constraint selection nor the distance metric learning in traditional constrained clustering methods. In our work, we design an effective constraint selection strategy and combine a CNN-based feature learning approach with the constrained clustering algorithm. The proposed model consists of two iterative steps: First, we replace the random constraint selection strategy with a carefully designed one; based on the clustering result and constraints obtained, we fine tune the CNN and extract new features for distance re-calculation. Our model is evaluated on a realistic video dataset, and the experimental results demonstrate that our method can improve the constrained clustering performance and feature divisibility simultaneously even with fewer constraints.

Document type: 
Thesis
Rights: 
This thesis may be printed or downloaded for non-commercial research and scholarly purposes. Copyright remains with the author.
File(s): 
Senior supervisor: 
Greg Mori
Department: 
Applied Sciences: School of Computing Science
Thesis type: 
(Thesis) M.Sc.
Statistics: