Statistical inference · Statistical learning theory / Open question
Learning Interactions from Fluctuations
Observed correlations reflect both interactions and the dynamics used to observe them. Inference near slow or critical regimes must account for both.
Seed catalog · Showing the bundled record.
Problem statement
For an -spin Ising model of maximum degree , with and every nonzero coupling satisfying , observe stationary rate-one single-site heat-bath dynamics only at times . Each observed spin is independently flipped with known probability . Determine sample, elapsed-time and computational requirements for graph recovery and parameter error at most , retaining dependence on , noise and relaxation time.
Assumptions and scope
- Fix graph size/degree, parameter bounds, minimum nonzero coupling and a specified reversible update generator with known or explicitly estimated rates.
- Observe states at declared time gaps, possibly with bounded measurement noise; complete flip times and failed update attempts are not available unless stated.
- Separate structure recovery from parameter estimation and distinguish temperature-dependent information loss from the computational cost of producing independent samples.
What would settle it?
Matching observation-time and sample bounds with an efficient estimator for a specified subsampled regime, plus an indistinguishability or computational barrier when temporal resolution is insufficient.
Selected for a theoretical or computational target. A fitted scaling curve or improved benchmark score alone does not establish the stated learning-theoretic guarantee.
Status and evidence
Bresler and the interaction-screening line establish independent-sample Ising recovery. Dutt et al. analyze dynamical data; Gaitonde et al. learn from actual continuous-time evolution. Jayakumar et al. recover stationary models from independent strongly metastable samples under local conditions. The target is coarsened/noisy trajectory complexity, not basic low-temperature, dynamical or metastable learnability. This is an editorial, source-motivated research target. The preliminary review through 2026-11-19 does not certify novelty or exclude an unlocated resolution.
Preliminary source check; not an expert-verified open status or an independent audit of proposed proofs. The compilation date is not a literature-review cutoff.
Sources and further reading
- Bresler — Efficiently Learning Ising Models on Arbitrary Graphs ↗partial theory · source checked 2026-11-19
Primary abstract checked: efficient bounded-degree graph recovery from independent samples, including low-temperature and nonuniform cases under parameter-identifiability assumptions. Slow physical sampling is separate from learning with independent samples.
- Gaitonde, Moitra and Mossel — Learning Ising Models from Evolutions ↗partial theory · source checked 2026-11-19
Publisher abstract checked: efficient recovery from actual continuous-time state evolution without observing failed update attempts, for a broad reversible single-site dynamics class. Temporal dependence alone is not an unresolved barrier.
- Vuffray, Misra, Lokhov and Chertkov — Interaction Screening: Efficient and Sample-Optimal Learning of Ising Models ↗partial theory · source checked 2026-11-19
Primary abstract checked: convex regularized local screening gives graph recovery with logarithmic system-size dependence and exponential coupling/degree dependence under full independent samples. Basic inverse-Ising recovery is established.
- Lokhov, Vuffray, Misra and Chertkov — Optimal Structure and Parameter Learning of Ising Models ↗partial theory · source checked 2026-11-19
PDF p.2 and Discussion p.9 checked: RISE/logRISE and thresholded pseudolikelihood; distinguish proved bounds from empirical exponents. The paper's 'optimal regime' does not determine the exact temperature exponent. Dense-model scaling was posed as a direction, but has subsequent results.
- Dutt, Lokhov, Vuffray and Misra — Exponential Reduction in Sample Complexity with Learning of Ising Model Dynamics ↗partial theory · source checked 2026-11-19
Primary abstract checked: screening/conditional-likelihood estimators learn from dynamical observations; far-from-equilibrium data can reduce sample requirements exponentially in the studied regimes. The observation protocol is part of the guarantee.
- Jayakumar, Lokhov, Misra and Vuffray — Discrete Distributions Are Learnable from Metastable Samples ↗partial theory · source checked 2026-11-19
PDF Definitions 1–2, Condition 1 and Theorem 1 checked: strong metastability bounds a detailed-balance defect; summed average conditional TV is at most eta/omega_P. Learning analysis uses independent draws from the metastable law; arbitrary correlated trajectories or weak stationarity alone are not covered.