Loading...

Multi-label Classification by Considering Label Dependencies

Farahnak Ghazani, Fatemeh | 2015

2030 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 47796 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Soleymani, Mahdieh
  7. Abstract:
  8. In multi-label classification problems each instance can simultaneously have multiple labels. In these problems, in addition to the complexities of the input feature space we encounter the complexities of output label space. In the multi-label classification problems, there are dependencies between different labels that need to be considered. Since the dimensionality of the label space in real-world applications can be (very) high, most methods which explicitly model these dependencies are ineffective in practice and recently those methods that transform the label space into a latent space have received attention. A class of these methods which uses output space dimension reduction, first uses an encoding method to transform the ouput space to lower dimensional latent space. Afterwards, they predict the latent space from the feature space and convert them to the original output space using a suitable decoding method. These methods consider label dependencies implicitly in the latent space, which is common between all the labels. The encoding method can be explicit which learns a function to code any label vector or implicit which learns a code matrix for label space directly by optimizing a cost function. In this research, the proposed methods use implicit encoding to reduce output space dimensionality and learn latent space common between all the labels. In these methods, we try to propose appropriate optimization problems for the multi-label classification problem. The proposed methods consider label dependencies implicitly in finding the latent space. In these methods kernel tricks can also be used to predict the latent space from the input space. Moreover, the threshold can be learnt for each instance, instead of using a predefined threshold for all instances. Extensive experiments on several datasets demonstrate the effectiveness of the proposed method compared with previous methods
  9. Keywords:
  10. Multi-Label Classification ; Label Space Dimension Reduction ; Implicit Encoding ; Label Dependencies

 Digital Object List

 Bookmark

...see more