Research map / Learning theory and algorithms / Sample complexity
BR-SNIS: Bias Reduced Self-Normalized Importance Sampling
Gabriel Cardoso, Sergey Samsonov, Achille Thin, Eric Moulines, Jimmy Olsson
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 |
|---|---|---|---|
| recycle i-SIR samples (BR-SNIS) instead of standard SNIS | SNIS · Gaussian mixture, M=16384, N in {129,513} | bias: by a factor of almost 10 | Improved |
| recycle i-SIR samples (BR-SNIS) instead of standard SNIS | SNIS · Gaussian mixture, M=16384 | MSE: around 20% | Hurt |
| use burn-in k0 = floor(0.625 kmax) instead of kmax-1 (ablation) | BR-SNIS · Gaussian mixture | bias and MSE: 3 times less bias, 10% more MSE | Mixed |
| BR-SNIS instead of SNIS, M=32 | SNIS · Bayesian logistic regression, CoverType | TV distance: 0.0028 -> 0.0014 | Improved |
| BR-SNIS instead of SNIS, M=512 | SNIS · Bayesian logistic regression, CoverType | TV distance: 0.0026 -> 0.0013 | Improved |
| BR-SNIS instead of SNIS, M=32 | SNIS · Bayesian logistic regression, Breast | TV distance: 0.00011 -> 7.9e-5 | Improved |
| BR-SNIS instead of SNIS, M=512 | SNIS · Bayesian logistic regression, Heart | TV distance: 7.8e-5 -> 4.9e-5 | Improved |
| BR-SNIS instead of SNIS | SNIS · Bayesian logistic regression, three datasets | posterior mean bias vs variance | Mixed |
| train IWAE decoder with BR-SNIS gradient (k=4) instead of plain IWAE, d=20 | IWAE · binarized MNIST | mean log-likelihood: -91.5 -> -90.5 | Improved |
| train IWAE decoder with BR-SNIS gradient (k=4) instead of plain IWAE, d=50 | IWAE · binarized MNIST | mean log-likelihood: -92.9 -> -90.1 | Improved |
| increase bootstrap rounds k from 4 to 8, d=20 (ablation) | BR-IWAE · binarized MNIST | mean log-likelihood: -90.5 -> -90.4 | No clear effect |
| increase bootstrap rounds k from 4 to 8, d=50 (ablation) | BR-IWAE · binarized MNIST | mean log-likelihood: -90.1 -> -90.1 | No clear effect |
Closest papers by meaning
- Differentiable Annealed Importance Sampling Minimizes The Jensen-Shannon Divergence Between Initial and Target Distribution (ICML 2024 · similarity 0.84)
- Understanding the difficulties of posterior predictive estimation (ICML 2025 · similarity 0.83)
- One Sample Fits All: Approximating All Probabilistic Values Simultaneously and Efficiently (NeurIPS 2024 · similarity 0.82)
- Fearless Stochasticity in Expectation Propagation (NeurIPS 2024 · similarity 0.82)
- Estimating the Permanent by Nesting Importance Sampling (ICML 2024 · similarity 0.81)
- A Black-Box Debiasing Framework for Conditional Sampling (NeurIPS 2025 · similarity 0.81)
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.
