Research map / Probabilistic and scientific modeling / Variational and generative sampling
Importance Weighted Hierarchical Variational Inference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the learned inverse proposal τ(ψ|z,x) instead of the hyperprior q(ψ|x) for estimating the hierarchical posterior density | IWHVI versus SIVI-like · MNIST-trained VAE | marginal log-likelihood bound | Helped |
| Increase K during training (ablation) | IWHVI-VAE · MNIST | marginal log-likelihood | Helped |
| Increase K during training (ablation) | IWHVI-VAE · MNIST | KL(τ(ψ|z,x) || q(ψ)) | Helped |
| Use the IWHVI-DReG gradient estimator (ablation) | IWHVI-VAE · MNIST validation | marginal log-likelihood | Helped |
| Use the IWHVI-DReG gradient estimator (ablation) | IWHVI-VAE · MNIST | KL(τ(ψ|z) || q(ψ)) | Helped |
| Replace the learned τ(ψ|z,x) with the prior q(ψ|x) | IWHVI · MNIST-trained VAE | marginal log-likelihood bound | Hurt |
| Use SIVI sample reuse (ablation) | SIVI · MNIST-trained VAE | marginal log-likelihood bound | No clear effect |
| Increase K in the marginal log-density upper bound (ablation) | IWHVI · 50-dimensional factorized Laplace distribution | negative differential entropy bound gap | Helped |
| Use IWHVI instead of HVM | VAE · MNIST | KL(τ(ψ|z,x) || q(ψ)): extremely close to zero -> approximately 6.2 nats | Helped |
| Use IWHVI instead of HVM | VAE · OMNIGLOT | KL(τ(ψ|z,x) || q(ψ)): ≈ 1.98 nats -> ≈ 9.97 nats | Helped |
| Omit the ψ₀-based B term from IWHVI-DReG (ablation) | IWHVI-DReG · Toy gradient-SNR task | gradient signal-to-noise ratio | Helped |
| Use the standard autodiff gradient instead of IWHVI-DReG (ablation) | IWHVI · Toy gradient-SNR task | gradient signal-to-noise ratio as K grows | Hurt |
Closest papers by meaning
- Variational Inference with Mixtures of Isotropic Gaussians (NeurIPS 2025 · similarity 0.90)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.89)
- Provably Scalable Black-Box Variational Inference with Structured Variational Families (ICML 2024 · similarity 0.89)
- Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians (ICML 2024 · similarity 0.89)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · similarity 0.89)
- Kernel Semi-Implicit Variational Inference (ICML 2024 · 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.