Research map / Probabilistic and scientific modeling / Variational and generative sampling
LSB: Local Self-Balancing MCMC in Discrete Spaces
Emanuele Sansone
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 |
|---|---|---|---|
| Learn the balancing function with the more expressive LSB 2 parametrization | LSB · Ising model, especially dependent-clean Case 4 | convergence speed / target likelihood evaluations | Improved |
| Learn the balancing function with LSB 2 instead of using fixed balancing functions | LSB · Ising Case 3 | convergence speed: twice time faster | Improved |
| Use LSB 2 instead of fixed balancing functions | LSB · UAI Markov networks MN 1 and MN 2 | target likelihood evaluations to converge | Improved |
| Use a learned LSB proposal rather than fixed sqrt(t) proposal | RBM sampler · RBM | target likelihood evaluations to converge | Improved |
| Use the more expressive FLSB 2 rather than FLSB 1 (ablation) | RBM sampler · RBM | ESS/sec | Hurt |
| Use FLSB 1 rather than the fixed sqrt(t) GWG proposal (ablation) | RBM sampler · RBM | ESS/sec | No clear effect |
| Use a learned balancing function rather than fixed sqrt(t) | LSB 1 · Ising Case 1 | ESS: 2.29±0.23 -> 2.32±0.27 | No clear effect |
| Use a learned balancing function rather than fixed sqrt(t) | LSB 1 · Ising Case 2 | ESS: 2.89±0.28 -> 2.85±0.28 | No clear effect |
| Use a learned balancing function rather than fixed sqrt(t) | LSB 1 · UAI MN 1 and MN 2 | ESS | No clear effect |
| Use the more expressive LSB 2 rather than LSB 1 (ablation) | LSB · Ising Case 4 | ESS: 2.52±1.11 -> 20.11±15.58 | Improved |
| Use the more expressive LSB 2 rather than LSB 1 (ablation) | LSB · UAI MN 2 | ESS: 3.52±0.42 -> 3.44±0.44 | No clear effect |
| Use LSB 2's more complex balancing function rather than LSB 1's smaller parametrization (ablation) | FLSB · RBM | ESS/sec | Hurt |
Closest papers by meaning
- AutoStep: Locally adaptive involutive MCMC (ICML 2025 · similarity 0.87)
- Sampling from Binary Quadratic Distributions via Stochastic Localization (ICML 2025 · similarity 0.86)
- Parallel Affine Transformation Tuning of Markov Chain Monte Carlo (ICML 2024 · similarity 0.86)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.85)
- Discrete Neural Flow Samplers with Locally Equivariant Transformer (NeurIPS 2025 · similarity 0.85)
- Tuning Sequential Monte Carlo Samplers via Greedy Incremental Divergence Minimization (ICML 2025 · similarity 0.85)
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.
