Density Estimation Using Real NVP

• 
• follow up to NICE - non linear independant components estimation
• different sets of latent variables for different resolutions $z=(z^{(1)},\dots ,z^{(L)})$
  • latents corresponding to finer details can be left out during inference

Also for VAE

• Eg: Parameterize the approximate posterior in a VAE using a flow for better inference $p_{\theta }(x|z)p_{\theta }(z)$
  • Open: Pasted image 20241119171440.png
• transform into something more flexibleOpen: Pasted image 20241119171554.png