Robust and efficient dimension reduction

1
Robust and efficient dimension reduction

• Marco Riani Matilde Bini
• Univ. di Parma Univ. di Firenze

2
Principal Component Analysis (PCA)

• Statistical properties of eigenvalues and
  eigenvectors are difficult to study
• Most results are obtained under normal
  assumptions

3
Multivariate transformations

• Very difficult to test and validate a particular
  transformation due to masking and swamping
  problems

and

• are sensitive to the presence of outliers