generative-note

generative model paper notes

GP-VAE: Deep Probabilistic Time Series Imputation

• missing data
  • deal with latent space instead filling the missing observation as the feature representation are complete
• multi time scale
  • a mixture of RBF kernels with different timescale
  • a Gamma distribution over the length scale to compute infinite mixture of RBF -> Cauchy kernel
• efficient inference
  • precision matrix is parameterized in terms of a product of bidiagonal matrices
• encoder (CNN) convolve over time dimension of the input, outputs a tensor of size T×3k, corresponding to timestep t and 3k parameters, where k is the dimensionality of the latent space

Transformable Bottleneck Networks

Multi-Angle Point Cloud-VAE: Unsupervised Feature Learning for 3D Point Clouds from Multiple Angles by Joint Self-Reconstruction and Half-to-Half Prediction