Singapore

Personalization in Large Language Models (LLMs) often relies on user-specific soft prompts. However, these prompts become obsolete when the foundation model is upgraded, necessitating costly, full-scale retraining. To overcome this limitation, we propose the \textbf{P}rompt-level \textbf{U}ser \textbf{M}igration \textbf{A}dapter (\textbf{PUMA}), a lightweight framework that efficiently migrates personalized prompts across incompatible models. PUMA utilizes a parameter-efficient adapter to bridge the semantic gap, combined with a group-based user selection strategy to drastically reduce training costs. Experiments on three large-scale datasets show our method matches or even surpasses the performance of retraining from scratch, reducing computational cost up to 98\%. The framework demonstrates strong generalization across diverse model architectures and robustness in advanced scenarios like chained and aggregated migrations, offering a practical path for the sustainable evolution of personalized AI by decoupling user assets from the underlying models.

AAAI 2026

Don’t Start Over: A Cost-Effective Framework for Migrating Personalized Prompts Between LLMs

technical paper

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.<br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>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.

Explainable AI (XAI) as a field has traditionally focused
and succeeded at providing explanations for single-shot
decision-making systems like classifiers.
However, there is a growing consensus that some of the most
challenging problems within explainable AI lie not within
single-shot decision-making settings, but rather in
sequential decision-making ones. Moreover, by switching to
a sequential decision-making setting, we also see novel
opportunities to build explainable agents that are actually
able to reason about when and how to provide explanations
to maximize their effectiveness. As of right now, most of
the work within explainable sequential decision-making
tends to happen in isolation within subcommunities like
planning, reinforcement learning, and robotics. While this
has led to the development of useful explanation methods,
the fragmentation has also made it harder to adopt and
transfer advances and insights from one subcommunity to
another. Addressing this requires us to take a more
holistic view of explainable sequential decision-making,
one that allows us to account for and understand the
contributions from every subfield in a single unified
framework. In this talk, we will talk about the efforts
that have been made towards this end and some of the most
important challenges that remain.

Explanations for Sequential Decision-Making – an Overview

Existing reinforcement learning (RL) methods struggle with long-horizon robotic manipulation tasks, particularly those involving sparse rewards. While action chunking is a promising paradigm for robotic manipulation, using RL to directly learn continuous action chunks in a stable and data-efficient manner remains a critical challenge. 
This paper introduces AC3 (Actor-Critic for Continuous Chunks), a novel RL framework that learns to generate high-dimensional, continuous action sequences. To make this learning process stable and data-efficient, AC3 incorporates targeted stabilization mechanisms for both the actor and the critic. First, to ensure reliable policy improvement, the actor is trained with an asymmetric update rule, learning exclusively from successful trajectories. Second, to enable effective value learning despite sparse rewards, the critic's update is stabilized using intra-chunk n-step returns and further enriched by a self-supervised module providing intrinsic rewards at anchor points aligned with each action chunk. We conducted extensive experiments on 25 tasks from the BiGym and RLBench benchmarks. 
Results show that by using only a few demonstrations and a simple model architecture, AC3 achieves superior success rates on most tasks, validating its effective design.

Actor-Critic for Continuous Action Chunks: A Reinforcement Learning Framework for Long-Horizon Robotic Manipulation with Sparse Reward

Control in high-dimensional action spaces remains a fundamental challenge in reinforcement learning (RL), primarily due to inefficient exploration of the action space. While recent methods attempt to guide exploration, they often fall short of achieving the agility and coordination exhibited in biological motor control. Inspired by how organisms exploit muscle synergies for efficient movement, we propose Explore to Learn (ETL), a two-stage framework that first discovers fundamental synergy patterns and then leverages them for task-specific policy learning.
In the first stage, ETL discovers underlying synergy patterns by deploying a targeted exploration policy. These patterns are modeled as latent directions in a low-dimensional space, along which the agent is guided to collect diverse and structured muscle activation trajectories. A variational autoencoder (VAE) is then trained to encode high-dimensional actions into a latent space whose dimensions correspond to the synergy patterns.
In the second stage, the policy is trained entirely in this synergy-aware latent space, producing synergy coefficients that the decoder maps back to full-dimensional muscle actions. This structured representation significantly reduces the complexity of learning, while the decoder is further fine-tuned to enhance expressiveness and generalization across downstream tasks.
Extensive experiments across musculoskeletal environments and the DMControl suite demonstrate that ETL consistently outperforms prior methods in both exploration efficiency and control performance, achieving superior scalability and generalization in overactuated control tasks.

Explore to Learn: Latent Exploration Through Disentangled Synergy Patterns for Reinforcement Learning in Overactuated Control

Video captions play a crucial role in text-to-video generation tasks, as their quality directly influences the semantic coherence and visual fidelity of the generated videos. Although large vision-language models (VLMs) have demonstrated significant potential in caption generation, existing benchmarks inadequately address fine-grained evaluation, particularly in capturing spatial-temporal details critical for video generation. To address this gap, we introduce the Fine-grained Video Caption Evaluation Benchmark (\textbf{VCapsBench}), the first large-scale fine-grained benchmark comprising 5,677 (5K+) videos and 109,796 (100K+) question-answer pairs. These QA-pairs are systematically annotated across 21 fine-grained dimensions (\eg, camera movement, and shot type) that are empirically proven critical for text-to-video generation. We further introduce three metrics (Accuracy (\textbf{AR}), Inconsistency Rate (\textbf{IR}), Coverage Rate (\textbf{CR})), and an automated evaluation pipeline leveraging a large language model (LLM) to verify caption quality via contrastive QA-pairs analysis. Our benchmark can advance the development of robust text-to-video models by providing actionable insights for caption optimization. The dataset and codes are available at the website: \url{https://github.com/anoycode22/VCapsBench}.

VCapsBench: A Large-scale Fine-grained Benchmark for Video Caption Quality Evaluation

Medication recommendation systems aim to provide personalized and safe medication options based on individual patient records. However, existing approaches often face challenges related to inadequate modeling of complex relationships within Electronic Health Records (EHRs), data sparsity, and a lack of explainability for recommendations. In this paper, we present a Knowledge-enhanced Explainable HyperGraph Convolution Network (KEHGCN) that constructs a hierarchical hypergraph structure to capture the multi-level relationships within EHR data. By incorporating external knowledge graphs, our approach introduces additional positive relations that help alleviate the impact of data sparsity on model learning. Furthermore, by performing generalized metapath construction and selection on the knowledge graph, our approach achieves effective knowledge filtering and extracts semantically meaningful metapaths, thereby further enhancing the explainability of the recommendation results. We also explicitly introduce negative relations present in the domain knowledge to improve the safety of medication recommendation. Extensive experiments on different hospital departments of MIMIC-III and MIMIC-IV datasets demonstrate that KEHGCN outperforms other state-of-the-art baselines.

Knowledge-Enhanced Explainable Hypergraph Convolution Network for Medication Recommendation

Real-world conversations are rich with pragmatic elements, such as entity mentions, references, and implicatures. Understanding such nuances is a requirement for successful natural communication, and often requires building a local _world model_ which encodes such elements and captures the dynamics of their evolving states. However, it is not well-understood whether language models (LMs) construct or maintain a robust implicit representation of conversations. In this work, we evaluate the ability of LMs to encode and update their internal world model in dyadic conversations and test their _malleability_ under linguistic alterations. To facilitate this, we apply seven minimal linguistic alterations to conversations sourced from popular datasets and construct two benchmarks comprising yes-no questions. We evaluate a wide range of open and closed source LMs and observe that they struggle to maintain robust accuracy. Our analysis unveils that LMs struggle to memorize crucial details, such as tracking entities under linguistic alterations to conversations. We then propose a dual-perspective interpretability framework which identifies transformer layers that are _useful_ or _harmful_ and highlights linguistic alterations most influenced by harmful layers, typically due to encoding spurious signals or relying on shortcuts. Inspired by these insights, we propose two layer-regularization based fine-tuning strategies that suppress the effect of the harmful layers.

PRAGWORLD: A Benchmark Evaluating LLMs’ Local World Model Under Minimal Linguistic Alterations and Conversational Dynamics

While adversarial attacks can effectively deceive deep neural networks, their real-world applicability is often limited by complex and conspicuous patterns that reveal their attack intent to human observers. To overcome this limitation, we propose UYE, a novel camouflage framework designed to simultaneously mislead DNNs and evade human perception. UYE incorporates two key components: an attention refiner leveraging a pre-trained vision encoder to optimize adversarial patterns for robust attacks across diverse environments, and a perception evaluator trained on CAMO-Critic—a human preference dataset curated using tailored prompts from human-aligned large multimodal models—to ensure natural and unobtrusive camouflage generation. Extensive experiments demonstrate that UYE outperforms state-of-the-art methods in achieving an optimal balance between human stealth and model deception while maintaining effectiveness in real-world scenarios.

Unnoticed Yet Effective: A Hybrid Physical Camouflage Framework Against DNNs and Human Perception

Fairness-aware machine learning aims to build predictive models that comply with fairness requirements, particularly concerning sensitive attributes such as race, gender, and age. Among causality-based fairness notions, counterfactual fairness is widely adopted for its individual-level guarantees, requiring that an individual’s predicted outcome remains unchanged in a counterfactual world where its sensitive attribute is altered. However, existing methods critically assume that the true causal graph is fully known, which is rarely the case in practice. Moreover, counterfactual fairness suffers from inherent identifiability limitations, as counterfactual quantities cannot always be uniquely estimated from observational data, especially under incomplete causal knowledge. To address these challenges, we propose a principled framework (CF-ICG) for counterfactual fairness under imperfectly known causal graphs, e.g., Partially Directed Acyclic Graphs (PDAGs). We first introduce a criterion to determine the identifiability, and bound the counterfactual quantities under PDAGs. Building upon this, we develop an efficient local algorithm that avoids the exhaustive enumeration of all DAGs, ensuring robustness against worst-case fairness violations. Experimental results on synthetic and real-world datasets demonstrate the practical effectiveness and theoretical soundness of CF-ICG.

Counterfactual Fairness with Imperfect Causal Graphs

The proliferation of large language models has intensified demands for reliable content attribution, yet existing watermarking techniques face a fundamental trilemma: they cannot simultaneously optimize for robustness against attacks, minimal text quality degradation, and detection efficiency. To resolve this challenge, we propose \textbf{ARGH-Mark}, a novel watermarking framework that integrates three synergistic innovations: (1) anchor-synchronized phase recovery for maintaining detection integrity under insertion/deletion attacks, (2) RG-balanced vocabulary modulation that dynamically partitions lexicons via contextual hashing to preserve generation quality, and (3) Hamming-based error correction enabling single-bit error rectification through algebraic coding. 
Comprehensive evaluations across question answering (ELI5), summarization (CNN/DM), and code generation (HumanEval) demonstrate state-of-the-art performance: ARGH-Mark achieves 96.2\% detection F1 under 30\% deletion attacks—surpassing MPAC by 21.9\%—while maintaining negligible quality degradation ($\Delta_{\text{KL}} = 0.07$ vs. 0.21 baseline). The framework reduces detection latency to 40ms (48$\times$ faster than hash-seed methods) and boosts bit recovery by 17.9$\times$ at BER=0.2 through Hamming correction. Crucially, watermarked text preserves 99.3\% of baseline utility across all tasks.
Theoretically, we establish linear-time detection complexity and quantifiable robustness bounds, proving that ARGH-Mark achieves a new Pareto frontier in the watermarking design space. By open-sourcing both implementation and the \textsc{WatErMark} benchmark with 10K adversarial samples, this work advances trustworthy deployment of generative AI in alignment-critical applications.

ARGH-Mark: Anchor-Synchronized Watermarking with Hamming Correction for Robust and Quality-Preserving LLM Attribution

Professional photography face retouching requires achieving a balance between texture preservation and quality increase, a problem that conventional automated methods find difficult to effectively handle. We provide a new lightweight neural architecture that converts conventional dodge-and-burn photography methods into predictions for learnable additive blend maps. Instead of rebuilding whole images, our method uses a small U-Net that predicts pixel-level changes, allowing for exact brightness adjustments while maintaining the original skin texture. With a 6MB model that operates effectively on common hardware, the technique produces high-quality results while preserving texture fidelity, which is crucial for professional applications. Experimental validation offers significant computing advantages while demonstrating competitive performance with current methods.

Downloads

Next from AAAI 2026

Explanations for Sequential Decision-Making – an Overview

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES