Dimensionality Reduction and Visualization


  • To design a unified or generalized framework for DR methods (UFDR). 
  • To determine a clear (and definite) relationship between the performance and design parameters (weights determining the compromise between recall and precision, and regularization or smoothness parameters for outlier detection) according to the UFDR settings.
  • From the UFDR, to design new or more general divergence-based methods where the design parameters are independent and their roles are clearly identified.
  • To design faster and stable implementations for DR methods in such a way that the sensitivity to starting parameters is avoided.
  • To develop DR approaches that can be successfully incorporated into an interface in which user may interact on the parameters in an efficient and interactive way.