Research map / Probabilistic and scientific modeling / Variational and generative sampling
Gradient Estimation for Mixture Variational Inference
Javier Burroni, Daniel Sheldon
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 |
|---|---|---|---|
| use reparameterization-based estimator (J&K) instead of score function | mixture VI · Deep Markov model K=2 | test ELBO: −11.26 -> −7.72 | Improved |
| use post-stratified instead of score function | mixture VI · Deep Markov model K=5 | test ELBO: −11.24 -> −11.20 | Improved |
| use J&K or implicit reparameterization instead of post-stratified | mixture VI · Deep Markov model K=2 | test ELBO: −10.64 -> −7.72 | Improved |
| use a mixture (K>1) instead of single Gaussian | variational posterior · Deep Markov model | test ELBO: −7.61 -> −7.75 | Hurt |
| add Sticking-the-Landing to J&K (ablation) | J&K estimator · Deep Markov model K=2 | ELBO: +0.16 | Improved |
| add Sticking-the-Landing to post-stratified (ablation) | post-stratified estimator · Deep Markov model K=3 | ELBO: +0.01 | No clear effect |
| add Sticking-the-Landing (ablation) | estimators with single particle · MNIST-like datasets, diagonal Gaussians | ELBO: 24 of 32 configurations (75 %) | Mixed |
| add Sticking-the-Landing (ablation) | stratified, post-stratified, J&K · Baseball K=5 single particle | test ELBO | Hurt |
| use implicit reparameterization instead of score function | mixture VI · Deep Markov model running time | running time increase (%): 64 % to 68 % longer | Hurt |
| use J&K instead of score function | mixture VI · Deep Markov model running time | running time increase (%): 20 % longer | Hurt |
| use stratified instead of single-sample estimators | mixture VI · Baseball | ELBO | Improved |
| use stratified estimator in chains of mixtures | stratified estimator · Deep Markov model | feasibility | Hurt |
Closest papers by meaning
- Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inference (NeurIPS 2017 · similarity 0.92)
- Efficient Mixture Learning in Black-Box Variational Inference (ICML 2024 · similarity 0.91)
- Universal Boosting Variational Inference (NeurIPS 2019 · similarity 0.91)
- Variational Inference with Mixtures of Isotropic Gaussians (NeurIPS 2025 · similarity 0.91)
- Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians (ICML 2024 · similarity 0.89)
- End-to-end Learning of Gaussian Mixture Priors for Diffusion Sampler (ICLR 2025 · similarity 0.89)
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.
