Research map / Probabilistic and scientific modeling / Variational and generative sampling
The Boomerang Sampler
Joris Bierkens, Sebastiano Grazzi, Kengo Kamatani, Gareth Roberts
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 control-variate subsampling instead of full-data gradients | Boomerang Sampler · Logistic regression, large n | ESS per second | Improved |
| Use control-variate subsampling instead of the naive unbiased gradient estimator | Subsampled Boomerang Sampler · Large-data scaling | event rate | Improved |
| Use factorised component-wise switching | Boomerang Sampler · Sparse conditional-dependence models | computational cost | Improved |
| Use Boomerang dynamics rather than constant-velocity dynamics | PDMC samplers · Logistic regression | ESS per second | Improved |
| Increase dimension while using subsampled algorithms | Subsampled Boomerang Sampler · Logistic regression | ESS per second | Hurt |
| Use Boomerang rather than MALA | Logistic regression samplers · Dimensions d ≤ 32 | ESS per second: d ≤ 32 | Improved |
| Increase dimension when comparing Boomerang with MALA | Boomerang Sampler and MALA · Logistic regression | scaling with dimension | Hurt |
| Use the Boomerang Sampler instead of BPS | Boomerang Sampler · Standard-normal target with perturbed reference measure | ESS per second | Improved |
| Perturb the reference mean or covariance away from the target Gaussian | Boomerang Sampler · Standard-normal target | ESS per second | Mixed |
| Use Boomerang rather than Zig-Zag for high-level diffusion-bridge coefficients | Factorised Boomerang Sampler · Diffusion bridges, N=10 | number of reflections: N=10 | Improved |
| Increase the number of observations to n ≥ 10,000 | Boomerang Sampler · Logistic regression, d=2 | accepted switches: n ≥ 10,000; d = 2 | No clear effect |
Closest papers by meaning
- Stochastic Bouncy Particle Sampler (ICML 2017 · similarity 0.88)
- Chain of Log-Concave Markov Chains (ICLR 2024 · similarity 0.85)
- Piecewise deterministic generative models (NeurIPS 2024 · similarity 0.85)
- Reverse Diffusion Sequential Monte Carlo Samplers (NeurIPS 2025 · similarity 0.85)
- Multi-Marginal Couplings for Metropolis--Hastings (NeurIPS 2026 · similarity 0.84)
- Zeroth-Order Sampling Methods for Non-Log-Concave Distributions: Alleviating Metastability by Denoising Diffusion (NeurIPS 2024 · similarity 0.84)
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.
