Tag
The paper adapts classical statistical estimators to deep learning for tracing epistemic uncertainty into aleatoric and data-scarcity sources, leveraging approximate Fisher Information Matrices for scalability.
This paper proposes a unified definition of uncertainty as pointwise posterior risk and introduces a theory-backed benchmark using semi-synthetic datasets to directly compute oracle epistemic and aleatoric uncertainty, enabling fine-grained evaluation beyond proxy tasks.
This paper introduces the Hybrid-to-NeSy (H2N) framework, which systematically translates hybrid mechanistic-data-driven models into neuro-symbolic AI designs, enabling the derivation of metrics for structural violation and belief dispersion as measures of epistemic uncertainty in the mechanistic part.
This paper introduces Heckman-corrected epistemic uncertainty to address selection on unobservables in machine learning, demonstrating that importance weighting fails when selection depends on unobservables correlated with outcomes. The method restores calibration in controlled experiments and real data, outperforming standard UQ baselines.
This paper introduces the concept of the 'Affectosphere' and argues that emotion AI cannot fully determine the meaning of an individual's emotion due to irreducible uncertainty, leading to the norm of 'affective sovereignty' where the experiencing subject retains final interpretive authority.
This paper explores Large Language Models' inability to recognize their knowledge limits on structured clinical data, proposing a cross-model attribution divergence method to detect epistemic blind spots. The approach improves calibration and accuracy without training by combining few-shot examples and SHAP-derived feature evidence.
This paper proposes a quantile Bayesian risk-aware MDP framework for online RL that adaptively balances robustness and exploration over time, providing theoretical regret bounds and demonstrating strong empirical performance.
This paper investigates Neutrosophic Logic as a framework for modeling epistemic states in Large Language Models, demonstrating that it can capture 'hyper-truth' states beyond traditional probability constraints, leading to more transparent and ethically aware AI systems.
This paper develops a PAC-Bayesian framework for test-time adaptation that uses MMD-balls as credal sets, providing formal generalization bounds and separating epistemic from aleatoric uncertainty under distribution shift.
This paper introduces GOEN, a pipeline combining multi-scale features, L2 normalization, and Mahalanobis distance for OOD detection, and finds that CenterLoss regularization actually degrades OOD performance despite improving classification accuracy.
MIT researchers developed a new method for identifying overconfident LLMs by measuring cross-model disagreement across similar models, rather than relying solely on self-consistency metrics. This approach better captures epistemic uncertainty and more accurately identifies unreliable predictions in high-stakes applications.