Research map / Large language models / Agents, code and math
BayesAgent: Bayesian Agentic Reasoning Under Uncertainty via Verbalized Probabilistic Graphical Modeling
Hengguan Huang, Xing Shen, Guang-Yuan Hao, Songtao Wang, Lingfa Meng, Dianbo Liu, David Alejandro Duchene, Hao Wang, Samir Bhatt
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add verbalized PGM latent-variable reasoning (N=3) instead of Chameleon | Chameleon agent (Llama3-8B-Instruct) · ScienceQA | ECE (x10^2): 9.62 -> 1.67 | Improved |
| add verbalized PGM latent-variable reasoning (N=3) instead of Chameleon | Chameleon agent · ScienceQA | accuracy: 85.29 -> 86.38 | Improved |
| add numerical Bayesian inference (Dirichlet posterior, learned lambda) to vPGM | vPGM N=3 · ScienceQA | ECE (x10^2): 1.67 -> 1.05 | Improved |
| add numerical Bayesian inference to vPGM | vPGM N=3 · ScienceQA | accuracy: 86.38 -> 86.38 | No clear effect |
| increase latent variables from 3 to 4 (ablation) | vPGM · ScienceQA | ECE (x10^2): 1.67 -> 2.15 | Hurt |
| increase latent variables from 2 to 3 (ablation) | vPGM · ScienceQA | accuracy: 85.49 -> 86.38 | Improved |
| use vPGM instead of GCoT | GCoT · ChatCoach detection | BERTScore: 72.4 -> 76.3 | Improved |
| use vPGM instead of GCoT | GCoT · ChatCoach correction | BLEU-2 / Rouge-L: BLEU-2 1.6 -> 1.7; BERTScore 65.4 -> 68.3; ROUGE-L 2.0 -> 2.0 | Mixed |
| use vPGM instead of Chameleon+ on misinformation inputs | Chameleon+ · A-OKVQA-noisy | accuracy: 59.04 -> 61.03 | Improved |
| add BayesVPGM inference on top of vPGM | vPGM · A-OKVQA-noisy | ECE (x10^2): 10.54 -> 9.85 | Improved |
| latent variable Z2 flags rationale mismatch | BayesVPGM · A-OKVQA-clean vs noisy | P(Z2): 0.86 clean vs 0.42 noisy | Improved |
| use latent variables when no misinformation is present | BayesVPGM · A-OKVQA-clean | calibration | Hurt |
Closest papers by meaning
- Position: Uncertainty Quantification Needs Reassessment for Large Language Model Agents (ICML 2025 · similarity 0.88)
- From Assumptions to Actions: Turning LLM Reasoning into Uncertainty-Aware Planning for Embodied Agents (ICLR 2026 · similarity 0.88)
- Improving Cooperation in Language Games with Bayesian Inference and the Cognitive Hierarchy (AAAI 2025 · similarity 0.88)
- Sequential Probabilistic Uncertainty Estimation for Parallel Multi-Agent Reasoning Systems (NeurIPS 2026 · similarity 0.87)
- Narrowing the Collaboration Gap, Probably (NeurIPS 2026 · similarity 0.87)
- AutoToM: Scaling Model-based Mental Inference via Automated Agent Modeling (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
