Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hamiltonian Variational Auto-Encoder
Anthony L Caterini, Arnaud Doucet, Dino Sejdinovic
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add Hamiltonian dynamics with free tempering (K=15, epsilon fixed across layers) | Convolutional VAE · Binarized MNIST test | NLL: 83.20 -> 82.78 | Improved |
| Use fixed rather than no tempering at K=5 with epsilon fixed across layers (ablation) | HVAE · Binarized MNIST test | NLL: 83.68 -> 83.26 | Improved |
| Use fixed rather than no tempering at K=10 with epsilon fixed across layers (ablation) | HVAE · Binarized MNIST test | NLL: 83.40 -> 83.26 | Improved |
| Use fixed rather than no tempering at K=15 with epsilon fixed across layers (ablation) | HVAE · Binarized MNIST test | NLL: 83.82 -> 83.56 | Hurt |
| Use free rather than fixed tempering with epsilon fixed across layers (ablation) | HVAE · Binarized MNIST test | NLL: 83.26 -> 83.09 (K=5); 83.26 -> 82.97 (K=10); 83.56 -> 82.78 (K=15); 83.18 -> 82.93 (K=20) | Improved |
| Vary leapfrog step size across layers rather than keeping it fixed (free tempering) (ablation) | HVAE · Binarized MNIST test | NLL: 83.09 -> 83.01 (K=5); 82.97 -> 82.62 (K=10); 82.78 -> 82.62 (K=15); 82.93 -> 82.83 (K=20) | Improved |
| Increase leapfrog steps from K=15 to K=20 under free tempering with fixed epsilon (ablation) | HVAE · Binarized MNIST test | NLL: 82.78 -> 82.93 | Hurt |
| Add time-inhomogeneous tempering instead of fixing beta_0 to 1 (ablation) | HVAE · Gaussian model parameter estimation | squared parameter error | Improved |
| Use HVAE rather than mean-field variational Bayes | Gaussian model · Gaussian model parameter estimation | squared parameter error | Improved |
| Use HVAE rather than planar normalizing flow | Gaussian model · Gaussian model parameter estimation | squared parameter error | Mixed |
Closest papers by meaning
- Monte Carlo Variational Auto-Encoders (ICML 2021 · similarity 0.92)
- Cauchy–Schwarz Regularized Autoencoder (JMLR 2022 · similarity 0.89)
- Variational Autoencoder with Implicit Optimal Priors (AAAI 2019 · similarity 0.88)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.88)
- Embrace the Gap: VAEs Perform Independent Mechanism Analysis (NeurIPS 2022 · similarity 0.88)
- Simple and Effective VAE Training with Calibrated Decoders (ICML 2021 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
