Specialty
Simulations found a clearer advantage for changing privacy spending over time in longer website experiments; across an experiment portfolio, regret-based allocation often produced more simulated clicks than equal splitting at intermediate budgets.
Scientific publication3 min read
An arXiv preprint gives a closed-form upper boundary for Spearman’s rho when Spearman’s footrule is prescribed, identifies a unique optimizer, and maps the exact attainable region. It extends the analysis to finite rankings and mixability, but leaves the exact relationship between ξ and η unresolved.
Scientific publication6 min read
A new preprint evaluates MCES, a system that combines 11 analytical methods into one score for ranking candidate driver–outcome pairs. It performed strongly in several tested benchmarks, but the authors describe it as a hypothesis-prioritisation tool rather than proof of cause and effect.
Scientific publication4 min read
A methods preprint classifies covariance-polynomial denominators used in rational causal estimators. It reports exact self-normalization for products built from nested covariance volumes, while weak denominators can produce non-Gaussian ratio behavior.
Scientific publication3 min read
A preprint describes a Bayesian method that borrows information across related brain connections. Applied to multisite autism data, it produced a predominantly negative but non-uniform association pattern; the authors stress that the result is observational and conditional on model choices.
Scientific publication6 min read
A conceptual preprint links the development of spatial statistics with Spatial AI, arguing for flexible learning alongside explicit spatial structure and careful validation.
Scientific publication3 min read
A methods preprint proposes D3 and D3op, which use logistic Lasso screening and post-Lasso inference to test whether groups share the same mean vector across many variables. D3 was faster in the reported benchmark, but the findings are conditional on the study’s assumptions and selected settings.
Scientific publication4 min read
A preprint presents an assumption-dependent central-limit theorem for higher-order Markov chains whose order grows with sequence length, with a binary variable-length example showing one sufficient growth condition.
Scientific publication5 min read
A statistical methods preprint proposes fitting a working model and then calibrating its decision threshold on separate data. Theory and simulations report stronger control of false inclusions and smaller disagreement between estimated and true reliability sets under the tested conditions.
Scientific publication4 min read
A new graph-based AI system performed strongly in simulated epidemic forecasts, especially when predicting hospitalisation peaks. Its real-world accuracy remains unknown.
Scientific publication4 min read
A methods preprint tests propensity-score approaches for case-control studies. Weighting was close to target in simulations, while matching depended strongly on its settings. In a Montreal ovarian-cancer application, regular aspirin-use estimates were below 1 but varied by method and uncertainty.
Scientific publication3 min read