# Density Estimation Using Real NVP ```toc ``` - ![[dcc9909e54da929309a50a552074048b_MD5.webp|Open: Pasted image 20241119171117.png]] - follow up to [[NICE - non linear independant components estimation]] - different sets of latent variables for different resolutions $z = (z^{(1)}, …, z^{(L)})$ - 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 $p_\theta(x|z)p_\theta(z)$ - [[268644ad8a5d0ae0a9b1f6c93465616e_MD5.webp|Open: Pasted image 20241119171440.png]] ![[268644ad8a5d0ae0a9b1f6c93465616e_MD5.webp]] - transform into something more flexible[[71485bd3414e589c2d9e74c3f27fe674_MD5.webp|Open: Pasted image 20241119171554.png]] ![[71485bd3414e589c2d9e74c3f27fe674_MD5.webp]]