bayesian-inference

An essay arguing that typical LLM-based agent pipelines built as LLM → tool → action lack proper uncertainty handling, and proposing a belief-state architecture with Bayesian updates and value-of-information policies. The LLM should act as investigator/translator, while the system enforces permissions and calibrated beliefs.

This paper introduces a Bayesian approach to conditional causal discovery, where the posterior over causal graphs and parameters is conditioned on user-specified causal-effect constraints (e.g., a large causal effect). They adapt rare-event estimation techniques to handle events with small posterior mass and validate the method on synthetic data and the Sachs protein dataset.

This paper proposes a hybrid nested search framework that decouples structural sketching (by an LLM) from numeric parameter optimization (by traditional solvers like CMA-ES) in LLM-driven evolutionary optimization, and validates it across meta-optimization, code-based policies, and Bayesian inference tasks.

This paper investigates whether tabular foundation models (TFMs) like TabPFN, TabICL, TabDPT, and TabFM produce predictions consistent with any joint distribution. It demonstrates that all evaluated TFMs violate both marginalization and factorization consistency for classification and regression, questioning their Bayesian inference claims.

This paper introduces Bayesian Data Reweighting, a probabilistic framework that adaptively weights query-document pairs to mitigate false negatives in contrastive training, improving multimodal retrieval across multiple knowledge-based VQA benchmarks.

This paper connects counterfactual explanations to generalized Bayes inference, showing that distance-minimization CEs are MAP estimates of a Gibbs posterior, and introduces new decision rules and evaluation metrics.

This paper extends ad-hoc teamwork to multi-task settings by introducing CE-CM, an approximate Bayesian method for inferring task-invariant partner capabilities, and CE-CM-Div, which accounts for human behavioral diversity. Experiments show improved capability estimation and adaptation to human partners.

Proposes the Verbalized Particle Posterior (VPP), treating verbalized learning as Bayesian inference by maintaining a population of natural-language hypotheses as particles, updated via Metropolis-Hastings or Sequential Monte Carlo, and improving over single-hypothesis VML on benchmarks.

This paper presents QANTIS, a method that uses IBM Heron quantum hardware to perform calibrated belief updates for Partially Observable Markov Decision Processes (POMDPs), focusing on reusing a quantum belief-update primitive across sequential decision steps without corrupting the posterior. The authors validate the approach with a controlled case study on the Tiger POMDP, showing that the hardware posterior selects the same actions as exact Bayesian inference.

This paper presents a sparse Gaussian process framework for quantile regression that uses a Laplace approximation for posterior inference and variance-based mechanisms for adaptive inducing-input placement and data acquisition.

该论文提出贝叶斯上下文实验者（Bayesian in-context experimenters），通过训练Transformer模仿贝叶斯后验Neyman教师策略，实现自适应平均处理效应（ATE）估计，并采用混合专家Transformer处理未知平滑性，理论证明可通过监督预训练学习该策略。

This paper presents a method to reduce the size of chemical reaction networks (CRNs) implementing probabilistic inference by leveraging factor graph reduction techniques, resulting in smaller CRNs while preserving belief propagation fixed points on surviving variables.

An educational thread explaining the mathematical foundations used by quantitative trading firms like Renaissance Technologies, covering concepts from Bernoulli to Brownian motion.

This paper presents PC-MCMC-CIGP, a gray-box workflow that combines spike-and-slab topology sampling with physical constraints and a Chemical-Informed Gaussian Process for reaction network discovery. The method demonstrates improved yield on styrene epoxidation and distinguishes elementary pathways from deceptive fits on a hydrogen-bromine benchmark.

This paper compares joint and stepwise approaches for estimating learning under competing cognitive diagnostic models, using data from reading games. It shows that the choice of approach can change conclusions about learner progress, and joint analysis is more reliable when item-skill structure is uncertain.

Bayesian-Agent presents a framework that treats reusable skills and SOPs as hypotheses, using Bayesian inference to guide agent behavior and improve task performance through posterior-guided harness optimization. It achieves significant improvements on multiple benchmarks with deepseek-v4-flash.

This paper proposes an extended family of structured spatial priors combining total variation (TV) with ℓ_p norms for Bayesian T1 mapping, enabling uncertainty quantification. The method is evaluated on synthetic and real MRI datasets, showing improved spatial coherence and reduced uncertainty.

This paper develops a scaling limit theory for SGLD-Gibbs to provide principled hyperparameter tuning guidance for meaningful uncertainty quantification in large-scale latent variable models.

This paper proposes new discrete-time approximations for stochastic gradient Langevin dynamics (SGLD) with and without momentum, enabling accurate predictions of stationary covariance, iterate average covariance, and integrated autocorrelation time. The method provides improved tuning guidance for large-sample uncertainty quantification, especially under model misspecification.

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.