Unsupervised domain adaptation using eigenanalysis in kernel space for categorisation tasks

Suranjana Samanta, Sukhendu Das
  • IET Image Processing, November 2015, the Institution of Engineering and Technology (the IET)
  • DOI: 10.1049/iet-ipr.2014.0754

Domain Adaptation by Eigen Analysis

What is it about?

Domain adaptation is a technique needed for robust machine learning algorithms. It deals with the situation when the training and the testing samples do not follow the same underlying distribution. In this work, we have used the Eigenvectors and Eigenvalues of the training ans the testing sample to find a suitable transformation. The transformed instances of the training samples have identical Eigen analysis results as that of the test samples.

Why is it important?

The proposed method is fast to compute and also robust as it can easily handle non-linear transformation of data.

Perspectives

Suranjana Samanta
Indian Institute of Technology Madras

The work proposed in this paper describes a simple yet effective way to transform instances of the training set to match the distribution of the test set. The method has been applied to a wide range of dataset such as, image, video and text, which shows the effectiveness of the proposed method.

Read Publication

http://dx.doi.org/10.1049/iet-ipr.2014.0754

The following have contributed to this page: Suranjana Samanta