Singapore

Human cognition excels at transcending sensory input and forming latent representations that structure our understanding of the world. Although Large Language Models (LLMs) can produce chain‐of‐thought reasoning, they lack a principled framework to capture latent structures and model uncertainty, especially in compositional reasoning tasks. In this work, we explore for the first time how to bridge LLMs with probabilistic graphical models (PGMs) to address LLM reasoning under uncertainty. To this end, we introduce Verbalized Probabilistic Graphical Modeling (vPGM), a LLM-based Bayesian framework that (i) guides LLMs in following key principles of PGMs through natural language and (ii) refines the resulting posterior distributions via numerical Bayesian inference. Unlike many traditional probabilistic methods requiring substantial domain expertise, vPGM bypasses expert‐driven model design, making it well‐suited for scenarios with limited assumptions. We evaluated our model on several compositional reasoning tasks, both close-ended and open-ended. Our results indicate that the model effectively enhances confidence calibration and text generation quality.

AAAI 2026

BayesAgent: Bayesian Agentic Reasoning Under Uncertainty via Verbalized Probabilistic Graphical Modeling

bayesian inference

uncertainty quantification

large language models

We are pleased to announce the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25), which will be held in Philadelphia, Pennsylvania at the Pennsylvania Convention Center from February 25 to March 4, 2025.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

### [Invited Speakers](https://aaai.org/conference/aaai/aaai-25/aaai-25-invited-speakers/)

Register [here](https://aaai.org/conference/aaai/aaai-25/registration/)

AAAI 2025

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.



poster

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Case-based reasoning networks are machine-learning models that make predictions based on similarity between the input and prototypical parts of training samples, called prototypes. Such models are able to explain each decision by pointing to the prototypes that contributed the most to the final outcome. As the explanation is a core part of the prediction, they are often qualified as "interpretable by design". While promising, we show that such explanations are sometimes misleading, which hampers their usefulness in safety-critical contexts. In particular, several instances may lead to different predictions and yet have the same explanation. Drawing inspiration from the field of formal eXplainable AI (formal XAI), we propose Abductive Latent Explanations (ALEs), a formalism to express sufficient conditions on the intermediate (latent) representation of the instance that imply the prediction. Our approach combines the inherent interpretability of case-based reasoning models and the guarantees provided by formal XAI. We propose a solver-free and scalable algorithm for generating ALEs based on three distinct paradigms, compare them, and present the feasibility of our approach on diverse datasets for both standard and fine-grained image classification.

Formal Abductive Latent Explanations for Prototype-Based Networks

We study the problem of fairly allocating a set of $m$ goods among $n$ agents in the asymptotic setting, where each item's value for each agent is drawn from an underlying joint distribution. Prior works have shown that if this distribution is well-behaved, then an envy-free allocation exists with high probability when $m=\Omega(n\log{n})$ (Dickerson et. al., AAAI'14). Under the stronger assumption that item values are independently and identically distributed (i.i.d.) across agents, this requirement improves to $m=\Omega(n\log{n}/\log{\log{n}})$, which is tight (Manurangsi and Suksompong, SIDMA'21). However, these results rely on non-strategyproof mechanisms, such as maximum-welfare allocation or the round-robin algorithm, limiting their applicability in settings with strategic agents. 

In this work, we extend the theory to a broad and more realistic class of joint value distributions, allowing for correlations between agents, atomicity, and unequal probabilities of having the highest value for an item. We show that envy-free allocations continue to exist with high probability when $m=\Omega(n\log{n})$. More importantly, we give a new randomized mechanism that is truthful in expectation, efficiently implementable in polynomial time, and outputs envy-free allocations with high probability, answering an open question posed by Manurangsi and Suksompong, MSS'17. We further extend our mechanism to the settings of asymptotic weighted fair division, and multiple agent types and good types, proving new results in each of these cases.

Designing Truthful Mechanisms for Asymptotic Fair Division

In this paper, we investigate the complexity of determining if various restricted forms of hierarchical task network (HTN) planning have a plan. We perform a systematic analysis of new restrictions formed by applying symmetries and relaxations to two existing restrictions called regularity and tail-recursiveness. By doing so, we find that many variations on common restrictions do not affect the complexity of the plan existence problem at all, but we also obtain the counter-intuitive result that combining some of these seemingly inert relaxations together renders the plan existence problem undecidable. We also unearth a critical difference in definitions between an early paper in HTN planning and modern formalisms that appears to have gone unnoticed.

Symmetries and Other Variations of “End-Recursive” HTN Problems: Mapping the Border Between Decidable and Undecidable Restrictions

Remote sensing (RS) image–text retrieval faces significant challenges in real-world datasets due to the presence of Pseudo-Matched Pairs (PMPs)—semantically mismatched or weakly aligned image–text pairs—which hinder the learning of reliable cross-modal alignments. To address this issue, we propose a novel retrieval framework that leverages Cross-Modal Gated Attention and a Positive–Negative Awareness Attention mechanism to mitigate the impact of such noisy associations. The gated module dynamically regulates cross-modal information flow, while the awareness mechanism explicitly distinguishes informative (positive) cues from misleading (negative) ones during alignment learning. Extensive experiments on three benchmark RS datasets—RSICD, RSITMD, and RS5M—demonstrate that our method consistently achieves state-of-the-art performance, highlighting its robustness and effectiveness in handling real-world mismatches and PMPs in RS image–text retrieval tasks.

PMPGuard: Catching Pseudo-Matched Pairs in Remote Sensing Image–Text Retrieval

We study a sequential selling problem in which an agent receives daily offers to sell a good, incurs a holding cost each day, and is subject to sunk cost bias---allowing past, irrecoverable costs to influence present decisions. We introduce a formal model parameterizing the degree of sunk cost bias and distinguish between three behavioral types: optimistic (who ignore future bias), naive (who assume their current bias persists), and sophisticated (who anticipate the evolution of their own bias). For each type, we characterize the optimal selling strategy and precisely quantify the worst-case gap in expected objective profit compared to an unbiased agent. Our results show that optimistic agents can suffer a quadratic loss in profit due to excessive waiting, naive agents perform identically to unbiased agents, and sophisticated agents limit their losses to a linear function of the time horizon. These findings clarify how different anticipations of sunk cost bias affect sequential decision-making and suggest targeted interventions to mitigate inefficiency.

Sequential Selling with Sunk Cost Bias

In-context learning-based medical segmentation (ICLM) enables foundation models to generalize to unseen cases without retraining. To enhance performance on test queries, existing methods typically follow a two-stage process: (1) using a retrieval encoder (RE) to map both queries and training samples into a shared feature space, and (2) retrieving and utilizing the top-$k$ most similar training samples. While current methods fix the RE and focus on optimizing stage (2), we show that the choice of RE in stage (1) alone can account for over 70% of the performance variation, highlighting RE selection as a critical yet often overlooked factor in ICLM.
In this paper, we conduct an analysis of the RE selection and make two main findings: (1) dynamically selecting the RE for each query outperforms selecting a fixed RE for the entire task; and (2) feature-space heuristics (e.g., intra-class compactness and inter-class separability) fail to predict RE quality. To this end, we propose the \emph{instance-adaptive retrieval encoder selection} (IRES) method that can select the optimal RE for each query based on output predictions. IRES is based on the intuition that a good RE retrieves relevant demonstrations, helping the ICL model generate more accurate and stable segmentation masks. Thus, we introduce the shape stability score (S$^3$), which evaluates the morphological stability of predicted masks under iterative erosion. Experiments show S$^3$ correlates strongly with true RE quality (Pearson $>$ 0.8), serving as a reliable selection proxy. To reduce S$^3$’s per-query cost, we propose parallel prediction with reciprocal neighbor reuse (P2R), which accelerates inference by parallelizing encoding and reusing encoder selections across reciprocal neighbors, avoiding redundant computation. Built on S$^3$ and P2R, IRES improves ICLM performance across FUNDUS, Brain MRI, and Chest X-ray datasets, with up to 10.6% gain on fundus segmentation.

Retriever Encoder Selection Matters for In-Context Learning-based Medical Segmentation

In goal-oriented dialogue tasks, the main challenge is to steer the interaction towards a given goal within a limited number of turns. Existing approaches either rely on elaborate prompt engineering, whose effectiveness is heavily dependent on human experience, or integrate policy networks and pre-trained policy models, which are usually difficult to adapt to new dialogue scenarios and costly to train. Therefore, in this paper, we present Nested Rollout Policy Adaptation for Goal-oriented Dialogue (NRPA-GD), a novel dialogue policy planning method that completely avoids specific model training by utilizing a Large Language Model~(LLM) to simulate behaviors of user and system at the same time. Specifically, NRPA-GD constructs a complete evaluation mechanism for dialogue trajectories and employs an optimization framework of nested Monte Carlo simulation and policy self-adaptation to dynamically adjust policies during the dialogue process. The experimental results on four typical goal-oriented dialogue datasets show that NRPA-GD outperforms both existing prompt engineering and specifically pre-trained model-based methods. Impressively, NRPA-GD surpasses ChatGPT and pre-trained policy models with only a 0.6-billion-parameter LLM. The proposed approach further demonstrates the advantages and novelty of employing planning methods on LLMs to solve practical planning tasks.

A General Highly Accurate Online Planning Method Integrating Large Language Models into Nested Rollout Policy Adaptation for Dialogue Tasks

Geological CO$_2$ storage (GCS) involves injecting captured CO$_2$ into deep subsurface formations to support climate goals. The effective management of GCS relies on adaptive injection planning to dynamically control injection rates and well pressures to balance both storage safety and efficiency. Prior literature, including numerical optimization methods and surrogate-optimization methods, is limited by real-world GCS requirements of smooth state transitions and goal-directed planning within limited time. To address these limitations, we propose a Brownian Bridge–augmented framework for surrogate simulation and injection planning in GCS and develop two insights (i) Brownian bridge as smooth state regularizer for better surrogate simulator; (ii) Brownian bridge as goal-time-conditioned planning guidance for better injection planning. Our method has three stages: (i) learning deep Brownian bridge representations with contrastive and reconstructive losses from historical reservoir and utility trajectories, (ii) incorporating Brownian bridge-based next state interpolation for simulator regularization (iii) guiding injection planning with Brownian utility-conditioned trajectories to generate high-quality injection plans. Experimental results across multiple datasets collected from diverse GCS settings demonstrate that our framework consistently improves simulation fidelity and planning effectiveness while maintaining low computational overhead.

Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO2 Storage

The potential of Generalized Category Discovery (GCD) lies in its ability to identify previously undiscovered patterns in both labeled and unlabeled data by leveraging insights from partially labeled training samples. However, interference can arise due to the model's dual focus on discovering both novel and known categories, often leading to conflicts that obscure true patterns in the dataset. This paper presents a divide-and-conquer framework, Foundation-Adaptive Integrated Refinement (FAIR), which fine-tunes pretrained foundational weights for various purposes, divided into $\texttt{Foundation}$ (pretrained weights), $\texttt{Adaptive}$ (weights fine-tuned with an anticlimactic loss), and $\texttt{Integrated}$ (weights adjusted for both labeled and unlabeled data). The $\texttt{Adaptive}$ utilizes a newly proposed adaptive contrastive loss that introduces variances within classes to preserve the individuality of representations. The $\texttt{Integrated}$ addresses inherent estimation errors while dynamically estimating the number of categories, incorporating a cosine-based perturbation mechanism as a margin to ensure the ground truth falls within it, rather than relying on biased estimates. 
Extensive experiments on six benchmark datasets demonstrate our method's effectiveness, outperforming state-of-the-art algorithms, especially on fine-grained datasets.

Foundation-Adaptive Integrated Refinement for Generalized Category Discovery

Large Reasoning Models (LRMs) have advanced automated multi-step reasoning, but their ability to generate complex Chain-of-Thought (CoT) trajectories introduces severe privacy risks, as sensitive information may be deeply embedded throughout the reasoning process. Existing Large Language Models (LLMs) unlearning approaches that typically focus on modifying only final answers are insufficient for LRMs, as they fail to remove sensitive content from intermediate steps, leading to persistent privacy leakage and degraded security. To address these challenges, we propose Sensitive Trajectory Regulation (STaR), a parameter-free, inference-time unlearning framework that achieves robust privacy protection throughout the reasoning process. Specifically, we first identify sensitive content via semantic-aware detection. Then, we inject global safety constraints through secure prompt encoder. Next, we perform trajectory-aware suppression to dynamically block sensitive content across the entire reasoning chain. Finally, we apply token-level adaptive filtering to prevent both exact and paraphrased sensitive tokens during generation. Furthermore, to overcome the inadequacies of existing evaluation protocols, we introduce two metrics: Multi-Decoding Consistency Assessment (MCS), which measures the consistency of unlearning across diverse decoding strategies, and Multi-Granularity Membership Inference Attack (MIA) Evaluation, which quantifies privacy protection at both answer and reasoning-chain levels. Experiments on the R-TOFU benchmark demonstrate that STaR achieves comprehensive and stable unlearning with minimal utility loss, setting a new standard for privacy-preserving reasoning in LRMs.

Downloads

Next from AAAI 2026

Formal Abductive Latent Explanations for Prototype-Based Networks

Similar lecture

Reasoning over Uncertain Text by Generative Large Language Models

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES