Symbolic regression (SR) explores mathematical functions to capture relationships in datasets, but its adoption is limited by a lack of uncertainty quantification (UQ). This survey is the first to address UQ in SR, reviewing current literature across frequentist, Bayesian, and model selection approaches.
This content explores the critical aspect of evaluating the quality of uncertainty quantification specifically within Bayesian Neural Network inference. It likely delves into methodologies for assessing and enhancing the reliability of predictive uncertainty estimates in AI models.
MixAtlas introduces an uncertainty-aware method for optimizing data mixtures in multimodal LLM midtraining by decomposing corpora along image concepts and task supervision. Using proxy models and a Gaussian-process surrogate, it finds better-performing data recipes for improved sample efficiency and generalization.
Neural Bayesian Sequential Routing (NBSR) is a new framework modeling neural inference as active evidence accumulation over a hierarchical Directed Acyclic Graph. It utilizes a Dirichlet–Categorical conjugate framework to update a Dirichlet belief state, enabling path-dependent routing and uncertainty quantification.
This research paper proposes a new method to quantify uncertainty in language models by tracing the cumulative path of per-layer MLP updates. By extracting eleven scale-invariant geometric features, a sparse linear probe is shown to outperform maximum softmax probability in evaluating uncertainty, especially with baseline miscalibration.
This research addresses the critical challenge of quantifying uncertainty in Large Reasoning Models (LRMs), noting the limitations of traditional and existing Conformal Prediction (CP) methods. It aims to develop a statistically rigorous approach that considers logical connections, interprets uncertainty origins, and disentangles reasoning quality from answer correctness.
This paper proposes a novel bootstrap-based framework for uncertainty quantification (UQ) in Convolutional Neural Networks (CNNs), addressing the lack of theoretically consistent UQ tools. The method utilizes convexified neural networks to establish theoretical consistency, offers significantly less computational load, and explores a novel transfer learning approach.
This paper argues that current Uncertainty Quantification (UQ) methods for LLMs are essentially unsupervised clustering algorithms, measuring internal consistency rather than external correctness. Consequently, these methods fail to detect "confident hallucinations" and may create a deceptive sense of safety when deploying LLMs in high-stakes domains.
Este artigo propõe SELFDOUBT, uma estrutura de passagem única para quantificar a incerteza em LLMs de raciocínio, especialmente para APIs proprietárias. Utiliza o Hedge-to-Verify Ratio (HVR) para identificar marcadores de incerteza e autoavaliação diretamente do rastro de raciocínio, superando métodos caros de amostragem.
Este artigo explora o uso de busca evolucionária impulsionada por LLMs para desenvolver automaticamente métodos de Quantificação de Incerteza (UQ) não supervisionados. Os métodos evoluídos superam baselines manuais em verificação de alegações, demonstrando generalização robusta e estratégias distintas entre diferentes modelos de LLM.