Lossy Compression

A few techniques I have been studying:

  • Linear subspace projection (DCA, PCA).
  • Non-linear subspace projection (KDCA, KPCA, KRUCA).
  • Lossy networks (Autoencoder, narrow networks).
  • Compressive multi-kernel method.
  • DCA/KDCA desensitization.
  • Constrained PCA/KPCA, constrained DCA/KDCA.

Reconstructed images using different subspace projection methods on Yaleface
Reconstructed images using different subspace projection methods (random projection vs PCA vs DCA) on the Yaleface dataset. See my MLSP 2016 paper for more detail.