427 papers
Distribird is an agentic web application that automates the creation of literature-informed prior distributions for Bayesian model calibration by deploying a multi-agent pipeline to extract, weight, and fit scientific data, thereby offering a traceable and validity-checked alternative to the commonly used uniform priors or ungrounded single-prompt LLM baselines.
This paper retrospectively validates a machine-learning model for New York City housing cost burdens, revealing systematic under-prediction due to static economic assumptions and subsequently recalibrating the framework with uncertainty-aware methods to generate more robust, scenario-based forecasts for 2025–2028.
This paper proposes a new matrix variate skew-normal distribution and its finite mixture model to effectively capture asymmetry and perform clustering in matrix-valued data, offering improved flexibility and performance over traditional symmetric models through derived estimators and an ECM algorithm.
This paper proposes a Seemingly Unrelated Autoregressive Distributed Lag (SUR-ARDL) framework for small-N panels that enhances estimation efficiency and cointegration testing power by leveraging cross-equation error correlations while avoiding the finite-sample distortions of long-run covariance matrix estimators, as demonstrated through theoretical proofs, Monte Carlo simulations, and an application to the renewable energy-growth nexus in ASEAN countries.
This paper proposes a closed-loop, human-AI collaborative framework that iteratively refines knowledge-based causal models (DAGs) by integrating process mining of empirical event logs with expert adjudication to enhance the accuracy and transparency of causal inference in complex chronic diseases.
This study demonstrates that when changepoint detectors for biopharmaceutical quality monitoring are compared under matched false-positive rates and search-space ablations, a standard variance-based penalty outperforms robust alternatives across various sample sizes and distributions, revealing that previous preferences for robust methods were driven by uncontrolled experimental conditions rather than inherent superiority.
This paper presents a practical, question-driven framework and accompanying tools to guide researchers in selecting the most appropriate randomized design family—such as parallel cluster trials, stepped-wedge, factorial, or adaptive trials—for implementation strategy studies based on their specific decision-making goals rather than methodological preference.
This paper demonstrates that one-step prediction error is a poor predictor of long-term statistical fidelity in learned chaotic surrogates and introduces a system-specific response operator that can efficiently bound long-run statistical errors without requiring computationally expensive rollouts.
This paper proposes a robust log-contrast regression framework that combines Huber loss and Lasso penalty to handle outliers and ensure sparsity in high-dimensional compositional microbiome data, while providing theoretical guarantees and an efficient symmetric Gauss-Seidel ADMM algorithm for implementation.
This paper introduces Penalized Multinomial Logistic Tree Regression (PMLTR) as a hybrid model that balances the predictive power of data-driven methods like RUMBoost with the interpretability of classical MNL models, demonstrating through extensive benchmarking that while manual specification remains competitive, purely data-driven approaches excel in prediction whereas PMLTR uniquely offers readable utilities essential for optimization and inference.