Loading...

Multi-Modal Distance Metric Learning

Roostaiyan, Mahdi | 2015

840 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46956 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Soleymani, Mahdieh
  7. Abstract:
  8. In many real-world applications, data contain multiple input channels (e.g., web pages include text, images and etc). In these cases, supervisory information may also be available in the form of distance constraints such as similar and dissimilar pairs from user feedbacks. Distance metric learning in these environments can be used for different goals such as retrieval and recommendation. In this research, we used from dual-wing harmoniums to combining text and image modals to a unified latent space when similar-dissimilar pairs are available. Euclidean distance of data represented in this latent space used as a distance metric. In this thesis, we extend the dual-wing harmoniums for distance metric learning by: 1. using a sparse regularizer term in the latent space. 2. Using a group lasso (L2-L1 norm) regularizer to modal weights in dual-wing harmoniums. Grouping effect of group lasso norm can be considered as a modal selection technique for feature generation. Every new feature can be generated from one or more modal. 3. Using the above proposed models as layers of deep networks for distance metric learning. Experimental results show the effectiveness of the proposed methods in retrieval and classification applications. Results on NUSWIDE dataset show up to 7% improvement of K-NN classification accuracy and MAP retrieval measure in deep network mode
  9. Keywords:
  10. Sparse Representation ; Distance Metric Learning ; Multi-Modal Data ; Dual-Wing Harmonium ; Deep Networks ; Group Lasso

 Digital Object List

 Bookmark

No TOC