Singapore

This perspective article discusses an emerging research direction: to what extent can foundation models yield usable structure for modeling the physical world? We offer a Markovian formulation of structured world models and outline the notion of multi-level hybrid world representations that support compositional structure. We then review and suggest possible discovery paradigms, spanning distillation, interaction-driven continual learning, and ensemble learning.

AAAI 2026

Discovering Hybrid World Representations with Co-Evolving Foundation Models

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.

Video-to-Music generation seeks to generate musically appropriate background music that enhances audiovisual immersion for videos. However, current approaches suffer from two critical limitations: 1) incomplete representation of video details, leading to weak alignment, and 2) inadequate temporal and rhythmic correspondence, particularly in achieving precise beat synchronization. To address the challenges, we propose Video Echoed in Music (VeM), a latent music diffusion that generates high-quality soundtracks with semantic, temporal, and rhythmic alignment for input videos. To capture video details comprehensively, VeM employs a hierarchical video parsing that acts as a music conductor, orchestrating multi-level information across modalities. Modality-specific encoders, coupled with a storyboard-guided cross-attention mechanism (SG-CAtt), integrate semantic cues while maintaining temporal coherence through position and duration encoding. For rhythmic precision, the frame-level transition-beat aligner and adapter (TB-As) dynamically synchronize visual scene transitions with music beats. We further contribute a novel video-music paired dataset sourced from e-commerce advertisements and video-sharing platforms, which imposes stricter transition-beat synchronization requirements. Meanwhile, we introduce novel metrics tailored to the task. Experimental results demonstrate superiority, particularly in semantic relevance and rhythmic precision.

Video Echoed in Music: Semantic, Temporal, and Rhythmic Alignment for Video-to-Music Generation

In the federated clustering task, structural heterogeneity across clients inevitably impedes effective multi-source information sharing. To solve this issue, Personalized Federated Learning (PFL) has emerged as a potentially effective solution for image and text clustering. Unlike Euclidean data, graph-structured data exhibits diverse and fragile local patterns, which widely exist in real-world scenarios. Multi-graph data analysis in the federated learning setting is challenging and important, yet remains underexplored. This motivates us to propose a novel **PER**sonalized **F**ederated graph-**E**vel **C**lustering **T**work (PERFECT), which generates a specialized aggregation strategy for each client by uploading key model parameters and representative samples without sharing private information. Specifically, on each client, we first reconstruct privacy-preserving representative samples in a min-max optimization manner, and then upload these samples to the server for subsequent personalized parameter aggregation. On the server, we first extract graph-level embeddings from the uploaded data, and then estimate affinities among multiple learned embeddings to formulate a personalized aggregation strategy for each client. Subsequently, to help each local model better identify the cluster boundaries, we utilize clustering-wise gradient to update the key components in the personalized model parameters from the server. Extensive experimental results have demonstrated the effectiveness and superiority of PERFECT over its competitors.

Personalized Federated Graph-Level Clustering Network

Current image fusion methods struggle to adapt to real-world environments encompassing diverse degradations with spatially varying characteristics. To address this challenge, we propose a robust fusion controller (RFC) capable of achieving degradation-aware image fusion through fine-grained language instructions, ensuring its reliable application in adverse environments. Specifically, RFC first parses language instructions to innovatively derive the functional condition and the spatial condition, where the former specifies the degradation type to remove, while the latter defines its spatial coverage. Then, a composite control priori is generated through a multi-condition coupling network, achieving a seamless transition from abstract language instructions to latent control variables. Subsequently, we design a hybrid attention-based fusion network to aggregate multi-modal information, in which the obtained composite control priori is deeply embedded to linearly modulate the intermediate fused features. To ensure the alignment between language instructions and control outcomes, we introduce a novel language-feature alignment loss, which constrains the consistency between feature-level gains and the composite control priori. Extensive experiments on publicly available datasets demonstrate that our RFC is robust against various composite degradations, particularly in highly challenging flare scenarios.

Robust Fusion Controller: Degradation-Aware Image Fusion with Fine-Grained Language Instructions

Unsupervised multimodal semantic discovery aims to learn discriminative representations from multimodal data. However, existing methods suffer from two key limitations. First, they only align instances across modalities without modeling semantic-level consistency, which fails to mitigate semantic bias caused by the gaps among feature distributions of multiple modalities. Second, they inevitably generate incorrect negative pairs during contrastive learning, pushing semantically similar samples apart.
To address these challenges, we propose GLAD (Global and Local semantic Alignment for unsupervised multimodal semantic Discovery), which aligns multimodal data at both global and local semantic levels. At the global level, GSA integrates multi-modal features into a shared space and employs joint clustering via optimal transport to capture common semantic patterns while mitigating cross-modality semantic bias. At the local level, LSA adaptively weights samples within each cluster based on their semantic importance, alleviating the effect of incorrect negative pairs.
Through the joint optimization of GSA and LSA, GLAD effectively captures both the global semantic structure and the local semantic nuances of multimodal data. Extensive experiments on three benchmark datasets demonstrate GLAD significantly outperforms state-of-the-art methods, with an average improvement of 3.22\%.

Unsupervised Semantic Discovery via Global and Local Semantic Alignment in Multimodal Clustering

Emotion recognition from EEG signals is essential for affective computing and has been widely explored using deep learning. While recent deep learning approaches have achieved strong performance on single EEG emotion datasets, their generalization across datasets remains limited due to the heterogeneity in annotation schemes and data formats. Existing models typically require dataset-specific architectures tailored to input structure and lack semantic alignment across diverse emotion labels. To address these challenges, we propose EMOD: A Unified EEG Emotion Representation Framework Leveraging Valence–Arousal (V–A) Guided Contrastive Learning. EMOD learns transferable and emotion-aware representations from heterogeneous datasets by bridging both semantic and structural gaps. Specifically, we project discrete and continuous emotion labels into a unified V–A space and formulate a soft-weighted supervised contrastive loss that encourages emotionally similar samples to cluster in the latent space. To accommodate variable EEG formats, EMOD employs a flexible backbone comprising a Triple-Domain Encoder followed by a Spatial-Temporal Transformer, enabling robust extraction and integration of temporal, spectral, and spatial features. We pretrain EMOD on 8 public EEG datasets and evaluate its performance on three benchmark datasets. Experimental results show that EMOD achieves the state-of-the-art performance, demonstrating strong adaptability and generalization across diverse EEG-based emotion recognition scenarios.

EMOD: A Unified EEG Emotion Representation Framework Leveraging V-A Guided Contrastive Learning

Identifying suitable reaction conditions is critical for chemical synthesis, as they directly affect yield, selectivity, and transformation feasibility. While recent methods have shown promising results, most approaches rely on static molecular representations and fail to explicitly capture the structural transformations between reactants and products, which are essential for understanding reaction mechanisms and predicting conditions. In this work, we propose TRACE, a transformation-aware graph refinement framework for reaction condition prediction. TRACE constructs an atom-level graph that captures reaction-specific structural changes by integrating information from both reactants and products. A structure-aware encoder enriches atom features with chemical context, followed by a dynamic interaction refinement module that adaptively selects transformation-relevant edges. A mechanism-regularized graph encoder further incorporates reaction center information to guide learning toward condition-relevant interactions, enabling the model to better capture reaction patterns for accurate condition prediction. Experiments on benchmark datasets show that TRACE achieves state-of-the-art performance across multiple condition types. The incorporation of transformation-aware refinement improves predictive accuracy, enhances generalization, and supports robust performance in realistic synthesis planning scenarios.

TRACE: Transformation-Aware Graph Refinement for Reaction Condition Prediction

Evaluating reward models is a fundamental challenge in reinforcement learning, particularly in settings where the reward model is learned or manually designed. The standard paradigm for Reward Model Evaluation (RME) involves training an optimal policy via reinforcement learning (RL) on the given reward model and assessing model quality through the performance of the resulting policy. However, this approach conflates the quality of the reward model with the effectiveness of RL training, and is computationally expensive due to the need for policy optimization. Recent methods attempt to circumvent this issue by evaluating reward models directly, without RL, but often rely on impractical assumptions such as access to a ground-truth reward or fail to utilize available supervision in a fine-grained manner. To overcome these limitations, we propose the Policy Preference Alignment Coefficient (PPAC), a novel metric for RME that requires neither RL training nor ground-truth rewards. PPAC first generates a sequence of automatically ranked policy preferences that guarantee monotonic improvement in the policy value, and then quantifies the alignment between these generated preferences and those implied by the candidate reward model. Experimental results across gridworld and continuous control task demonstrate that PPAC yields preference sequences with consistently increasing policy values and outperforms existing metrics in evaluating reward model quality.

Reward Model Evaluation via Automatically-Ranked Policy Alignment

Recent studies have revealed Neural Collapse (NC) in deep classifiers, where last-layer weights and features align into an equiangular tight frame (ETF), concentrating class information along specific embedding directions. However, conventional fine-tuning typically disregards this structure, initializing task-specific classifier heads randomly. To explicitly leverage this phenomenon, we propose a simple yet effective method for metric learning: (1) initializing the classifier head along each class’s NC direction from a pretrained model to preserve the emergent structure, and (2) injecting small isotropic Gaussian noise during finetuning to boost generalization. In addition, we provide a theoretical bound proving that our method explicitly reduces cumulative weight drift from the NC-initialization, compared to standard finetuning. This suggests that our method better preserves the pretrained model’s class-specific structure. Empirically, this structural preservation yields Recall@K gains: reduced weight drift correlates with better performance. Concurrent decreases in the Neural Collapse 1 (NC1) measure confirm that stronger intra‐class cohesion underlies these improvements. Furthermore, we validate the effectiveness of our method on class‐imbalanced benchmarks.

Neural Collapse-Informed Initialization with Perturbation Injection in Classification-based Metric Learning

Link prediction is a fundamental task in network analysis with widespread applications, from social recommendation to knowledge graph completion. Fairness in this setting is critical, as biased predictions can propagate or exacerbate societal inequalities. Prior work adopts a dyadic perspective, enforcing fairness through demographic parity between intra-group and inter-group link predictions. However, this dyadic framing can obscure underlying disparities across subgroups, allowing systemic biases to go undetected. Moreover, we argue that demographic parity does not meet desired properties for fairness assessment in ranking-based tasks such as link prediction. We formalize the limitations of existing fairness evaluations and borrow a framework inspired by information retrieval that enables a more expressive assessment, addressing these limitations. Additionally, we propose a lightweight post-processing method combined with decoupled link predictors that effectively mitigates bias and achieves state-of-the-art fairness–utility trade-offs.

Breaking the Dyadic Barrier: Rethinking Fairness in Link Prediction Beyond Demographic Parity

Vision-Language Retrieval (VLR) aims to retrieve relevant visual or textual information from multimodal data using language or image queries. However, traditional VLR methods often rely on data-driven shallow semantic alignment and fail to understand the deeper structural and fine-grained entity features of queries, resulting in poor performance on multi-entity layouts and challenging entities. In this paper, we propose the Layout-Aware and Sketch-Enhanced (LASE) VLR framework, which refines query representations by incorporating multimodal layout and sketch knowledge. Specifically, layout knowledge encodes the spatial arrangement of entities, while sketch knowledge refines entity perception by capturing essential structural details. To extract these knowledge representations, we leverage Large Language Models' (LLMs) powerful semantic understanding for layout generation, and Diffusion Models' (DMs) fine-grained cross-modal generative capabilities for sketch generation. However, integrating knowledge into queries may introduce biases and query-specific preferences due to varying visual content and knowledge demands. To address this, we propose the Gated Dual-Stream Knowledge Module (GDKM), which consists of a multi-instance fusion network with a sample-aware gating network. The fusion network aggregates diverse knowledge using multi-head attention to reduce bias, while the gating network adjusts knowledge weights based on query characteristics. Extensive experiments demonstrate that the LASE significantly enhances VLR performance across multiple benchmarks, with superior generalization and transferability.

Downloads

Next from AAAI 2026

Video Echoed in Music: Semantic, Temporal, and Rhythmic Alignment for Video-to-Music Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2026

Video Echoed in Music: Semantic, Temporal, and Rhythmic Alignment for Video-to-Music Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads