English(EN)Accurate Large-sample Uncertainty Quantification using Stochastic Gradient Markov Chain Monte Carlo
新方法增强大型AI模型的不确定性量化能力
作者PulseAugur 编辑部·[8 个来源]·
研究人员正在开发新方法来改进大型模型中的不确定性量化。一种方法，语义高斯过程不确定性（SGPU），分析答案嵌入的几何结构，以估计语义一致性，而无需脆弱的聚类。另一个框架，方差门控集成（VGE），使用信噪比门将认知敏感性注入不确定性估计。这些方法旨在为高风险应用中的决策提供更可靠、更准确的不确定性估计。
AI
arXiv:2512.14177v3 Announce Type: replace Abstract: Large Vision-Language Models (LVLMs) often produce plausible but unreliable outputs, making robust uncertainty estimation essential. Recent work on semantic uncertainty estimates relies on external models to cluster multiple sam…
arXiv stat.ML
TIER_1English(EN)·H. Martin Gillis, Isaac Xu, Thomas Trappenberg·
arXiv:2602.08142v2 Announce Type: replace-cross Abstract: Machine learning applications require fast and reliable per-sample uncertainty estimation. A common approach is to use predictive distributions from Bayesian or approximation methods and additively decompose uncertainty in…
arXiv stat.ML
TIER_1English(EN)·H. Martin Gillis, Isaac Xu, Thomas Trappenberg·
arXiv:2509.08846v2 Announce Type: replace-cross Abstract: Evaluation of per-sample uncertainty quantification from neural networks is essential for decision-making involving high-risk applications. A common approach is to use the predictive distribution from Bayesian or approxima…
arXiv stat.ML
TIER_1English(EN)·Yu Wang, Jie Ding, Jonathan H. Huggins·
arXiv:2606.00293v1 Announce Type: cross Abstract: Tuning algorithms such as stochastic gradient descent (SGD) and stochastic gradient Langevin dynamics (SGLD) for approximate sampling and uncertainty quantification remains challenging, particularly in the practically relevant set…
arXiv stat.ML
TIER_1English(EN)·Xiaoyu Wang, Jonathan H. Huggins·
arXiv:2606.00309v1 Announce Type: cross Abstract: Stochastic gradient Langevin dynamics combined with Gibbs updates (SGLD--Gibbs) provides a highly scalable approach to approximate Bayesian inference in latent variable models. However, it remains unclear how to tune the algorithm…
arXiv stat.ML
TIER_1English(EN)·Hugh Dance, Peter Orbanz, Arthur Gretton·
arXiv:2410.14483v3 Announce Type: replace Abstract: Reliable uncertainty quantification for causal effects is crucial in high-stakes applications, but remains challenging when the target is an entire function rather than a scalar estimand. In this work, we introduce a GP-based ap…
arXiv stat.ML
TIER_1English(EN)·Jonathan H. Huggins·
Stochastic gradient Langevin dynamics combined with Gibbs updates (SGLD--Gibbs) provides a highly scalable approach to approximate Bayesian inference in latent variable models. However, it remains unclear how to tune the algorithm's hyperparameters in a principled manner to ensur…
arXiv stat.ML
TIER_1English(EN)·Jonathan H. Huggins·
Tuning algorithms such as stochastic gradient descent (SGD) and stochastic gradient Langevin dynamics (SGLD) for approximate sampling and uncertainty quantification remains challenging, particularly in the practically relevant settings when the batch size is large or the model is…
