Sampling-Optimisation-Geometry-Bench
SamplingBench
Laws, chains and sampling guarantees.
Public-dev · shared task identities · Dataset manifest
Choose a task
From atomic balance to reversibilitysogb-001 · NL L3 · Lean L3→A quadratic envelope from one midpointsogb-003 · NL L2 · Lean L2→Null atoms in a random-scan balance proofsogb-006 · NL L3→Adaptive hybrid error does not certify expected costsogb-007 · NL L4→A log-concave shape is not yet a probability densitysogb-008 · NL L1→What kind of guarantee does a sampling claim describe?sogb-009 · NL L1→One random-site step is not one sweepsogb-013 · NL L1→Invariant, reversible and convergent are different claimssogb-014 · NL L1→Averaging a conditional event-error boundsogb-017 · NL L3 · Lean L3→Gluing the right joint law before applying Minkowskisogb-018 · NL L4 · Lean L4→Why retaining one coordinate and conditionally resampling the other is invariantsogb-021 · NL L3 · Lean L2→One null set for every event, or a different null set for each event?sogb-024 · NL L3→From Gibbs Poincaré to a directional covariance boundsogb-025 · NL L3 · Lean L3→Can the SPHMC smoothing argument survive weak convexity?sogb-026 · NL L4→Tensorizing a two-point entropy bound without losing the flip constantsogb-027 · NL L3→From the Boolean cube to compact Gaussian log-Sobolevsogb-028 · NL L4 · Lean L3→Scope and difficulty
These are views of the same catalog, not new datasets. A cross-listed core keeps its identity, family, split and source binding. NL/Lean levels are provisional, not empirically calibrated.
Topics: Log-Concave Sampling, SampleWiki research problems, Discrete Sampling, Markov Chain Monte Carlo.