Research map / Probabilistic and scientific modeling / Variational and generative sampling
Improving Explorability in Variational Inference with Annealed Variational Objectives
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 |
|---|---|---|---|
| Add AVO training with a=0.2 to HVI (T=5) (ablation) | VAE inference on Binarized MNIST · Binarized MNIST test | -LL_te: 92.48 -> 90.01 | Helped |
| Add AVO training with a=0.2 to HVI (T=5) (ablation) | VAE inference on Binarized MNIST · Binarized MNIST test | -log p(D_te): 87.62 -> 86.06 | Helped |
| Increase AVO weight from a=0.2 to a=0.5 (ablation) | HVI on Binarized MNIST · Binarized MNIST test | -LL_te: 90.01 -> 90.26 | Hurt |
| Add AVO training with a=0.5 to HVI (T=5) (ablation) | VAE inference on Omniglot · Omniglot test | -LL_te: 113.19 -> 111.94 | Helped |
| Add AVO training with a=0.5 to HVI (T=5) (ablation) | VAE inference on Omniglot · Omniglot test | -log p(D_te): 106.25 -> 105.04 | Helped |
| Use HVI-AVO rather than HVI-ELBO | Hierarchical variational inference · Four-mode mixture toy energy | mode coverage | Helped |
| Use HVI-AVO rather than HVI-ELBO during beta annealing | Hierarchical variational inference · Six toy energy functions | negative KL | Helped |
| Use a unimodal Gaussian approximate posterior instead of the more exploratory AVO posterior | Amortized inference on the bimodal toy posterior · Biased noise model | posterior mode coverage | Hurt |
| Use AVO rather than the 500-sample IWAE posterior | Amortized inference on the bimodal toy posterior · Biased noise model | posterior approximation | No clear effect |
| Apply beta annealing that discourages exploration (ablation) | HVI-ELBO · Toy energy functions | mode coverage | Hurt |
| Use deterministic transitions rather than stochastic transitions (ablation) | Deep Sigmoidal Flows · Toy energy functions | exploration | Hurt |
Closest papers by meaning
- Robust, Accurate Stochastic Optimization for Variational Inference (NeurIPS 2020 · similarity 0.91)
- The Thermodynamic Variational Objective (NeurIPS 2019 · similarity 0.90)
- Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference (NeurIPS 2026 · similarity 0.88)
- Annealing in variational inference mitigates mode collapse: a theoretical study on Gaussian mixtures (NeurIPS 2026 · similarity 0.88)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · similarity 0.88)
- Globally Convergent Variational Inference (NeurIPS 2024 · similarity 0.87)
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.