identifiability

The paper proves that the closed-form bounds proposed by Shu et al. for probabilities of causation are sharp in every dimension, and shows that PNS, PSub, PRep, and PN are all special cases of a single unified theorem for discrete PoCs.

This paper proves the Ho–Nguyen conjecture on the order of density cancellation in over-fitted Gaussian mixtures for all cluster sizes M≥2, deriving exact maximum-likelihood exponents and local minimax rates, and applies the coefficient framework to identifiability issues in mixtures of experts.

Meta-iLaD 提出一种通过跨动力学环境元学习实现可识别潜在动力学的新框架，利用少量样本的条件先验快速提取环境变量，并同时对潜在状态、环境变量与动力学函数建立可识别性，从而提升对未见环境的泛化与预测能力。

该论文研究中心化二次神经网络的谱可辨识性与学习曲线辨识问题，构造了谱摘要一致但损失曲线不同的输入分布，并在高斯输入下证明单条精确标量损失曲线可辨识初始预测器与教师谱。

LLM-IDEA 论文提出一个由可识别性驱动的实验智能体，用于闭环自主发现机制性世界模型，其可识别性引擎能在能力上限、设计类内可解、认证耗尽三种平台期判定间做出选择，从而让自主智能体计算出平台期需要更多搜索、更优同类实验还是换一类实验。

DSReg提出通过依赖稀疏性正则化（Structural Diversity条件），在无需重建、解码器或标签的情况下可证明地恢复单个世界隐变量，建立了首个完全可识别的JEPA表示学习方法。

This paper establishes that one intervention per strongly connected component suffices to recover parameters of linear stochastic dynamics from steady-state data, and proposes a recursive learning algorithm.

This paper establishes identifiability conditions for causal DAGs with mixed ordinal and exponential family nodes under linear parametric models, proving that edge orientations can be determined from the joint distribution alone with sufficient categories and support points.

This paper establishes a spectral identifiability threshold for recovering dissipation rates in open quantum systems, showing how many Liouvillian modes are needed for amplitude damping and dephasing recovery in a six-qubit Lindblad model.

This paper introduces a theoretical framework for identifying model-intrinsic structure in mechanistic interpretability using Koopman operator theory, proving the first identifiability theorem for a mechanistic-interpretability primitive with empirical validation on GPT-2, Gemma-2-2B, and Qwen3-8B-Base.

This paper introduces GeoID-PINN, a physics-informed neural network for regional epidemic inference that incorporates geographic coupling via a spatial source-composition matrix. It demonstrates that accurate trajectory forecasts do not guarantee correct recovery of regional dependence structures, using simulations and COVID-19 data from Louisiana counties.

This paper introduces PCINN, a physics-chemistry-informed neural network that acts as a hybrid AI surrogate for real-time spatial atomic layer deposition (SALD) coverage prediction, achieving CFD-level accuracy in ~77ms and enabling reliable kinetics inversion via identifiability analysis.

This paper studies identifiability in matrix tri-factorisation, showing that sparsity constraints can induce unique solutions for the factorisation problem.

This paper shows that the smallest eigenvalue of the invariant-measure moment matrix determines the identifiability ceiling for system discovery methods like SINDy and PySR, based on attractor geometry. It introduces a framework that predicts discovery limits from a short reference trajectory without refitting.

This paper studies the inverse problem of recovering load magnitudes from stress-intensity-factor profiles along a crack front, using SIFBench finite-element data. It characterizes when the combined load is identifiable and provides an estimator with calibrated uncertainty for ill-posed regimes.

This paper analyzes how Bayesian causal discovery fails under latent confounding in linear Gaussian networks, deriving a correlation threshold that causes spurious edges to be favored by the score function and characterizing two distinct posterior failure regimes.

This paper introduces textual latent states and factorized GRPO (fGRPO) to enforce strict mediation in text-based world models, addressing the identifiability problem and achieving up to 57% gains in representation quality and 98% improvements in rollout performance.

This paper introduces concept modulation models (CMMs), a unified framework for identifiability and extrapolation in conditional generative models. It shows that feature agreement on observed attributes induces constraints through attribute potentials, enabling algebraic extrapolation criteria that recover and generalize existing results.

The paper provides a geometric interpretation of last-layer model stealing attacks on transformers using exterior differential systems, showing that recovery of the projection matrix is governed by the polar space of a quadric. It also characterizes an identifiability wall below the last layer, revealing what can and cannot be extracted.

This paper develops a measure-theoretic framework analyzing when contrastive learning recovers meaningful latent geometry, introducing a 'diversity condition' on positive-pair sampling and a support-corrected InfoNCE variant, with experiments validating that sampling diversity and architectural inductive bias interact critically in contrastive representation learning.