Tag
提出了一种基于条件归一化流的摊销变分推断方法，构建加权似然的局部后验来处理模型校准中的模型偏差与上下文不确定性，并在Branin-Hoo孪生实验及单柱海洋模型的k-ε湍流闭合校准中验证了其相较于全局后验的误差降低效果。
论文提出 KPSA（核化参数敏感度图谱），通过经验神经切线核的块分解构建紧凑表示，以一次前向计算估计未见输入的参数敏感度，替代逐样本反向传播归因，在视觉与蛋白质建模任务中以极小参数预算保留 39–85% 的扰动敏感度并大幅提升查询吞吐量。
The paper proposes PFN-OPE, a prior-data fitted network that amortizes off-policy evaluation across contextual bandit tasks, handling both policy shift and reward shift in LLM deployment. On HelpSteer2 and UltraFeedback with Qwen, Llama, and Gemma policies, it achieves 2.0 to 9.3 times lower error than the best baselines in reward-shifted settings.
LAVOIR introduces a method for single-pass decision models to ask questions based on amortized value of information, improving accuracy and efficiency in handling underspecified inputs.
This paper formulates cognitive experiment design as a Bayesian Experimental Design problem, introducing an amortized framework that efficiently identifies maximally informative environments for inferring latent planning parameters, with validated performance on the Mouselab-MDP paradigm.
This paper presents a hybrid amortized inference method that accelerates hierarchical sparse predictive coding by combining a fast initial estimate with corrective refinement steps, achieving better efficiency than pure iterative or amortized approaches.
Introduces Amortized Factor Inference Networks (AFINs), a family of encode-merge-decode inference networks that generalize across varying priors, likelihoods, and dimensionality, achieving posterior accuracy comparable to NUTS with much less compute.
SurvivalPFN is a prior-data fitted network that amortizes Bayesian inference for survival analysis via in-context learning, achieving strong predictive performance across 61 datasets without task-specific training or hyperparameter tuning.