Research map / Probabilistic and scientific modeling / Variational and generative sampling
Discretely Relaxing Continuous Variables for tractable Variational Inference
Trefor Evans, Prasanth Nair
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 exact DIRECT ELBO gradients instead of REINFORCE gradient estimates | discrete-prior GLM regression · synthetic random Fourier feature regression | optimization iterations and wall-clock time | Improved |
| Discretely relax a Gaussian prior and train with DIRECT mean-field inference instead of a continuous Gaussian prior with reparameterization | GLM regression · UCI regression datasets | test RMSE | Mixed |
| Discretely relax a Gaussian prior and train with DIRECT mean-field inference instead of a continuous Gaussian prior with reparameterization | GLM regression · UCI regression datasets | training time | Improved |
| Use a 5-component mixture variational distribution instead of mean-field (ablation) | DIRECT GLM regression · UCI regression datasets | test RMSE | Improved |
| Train the mixture variational distribution using an ELBO lower bound instead of unbiased SGD (ablation) | DIRECT mixture GLM regression · UCI regression datasets | test RMSE | Hurt |
| Train the mixture variational distribution using an ELBO lower bound instead of unbiased SGD (ablation) | DIRECT mixture GLM regression · UCI regression datasets | training time | Improved |
| Use a mixture variational distribution instead of mean-field (ablation) | DIRECT GLM regression · UCI regression datasets | training time | Hurt |
| Represent posterior samples using discrete low-precision values | DIRECT GLM regression · Gas | posterior sample sparsity: over 84% | Improved |
| Represent posterior samples using discrete low-precision values | DIRECT GLM regression · Electric | posterior sample sparsity: over 99.6% | Improved |
Closest papers by meaning
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.90)
- VarGrad: A Low-Variance Gradient Estimator for Variational Inference (NeurIPS 2020 · similarity 0.89)
- Quantized Variational Inference (NeurIPS 2020 · similarity 0.89)
- Least squares variational inference (NeurIPS 2025 · similarity 0.87)
- Globally Convergent Variational Inference (NeurIPS 2024 · similarity 0.87)
- EigenVI: score-based variational inference with orthogonal function expansions (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.
