AAAI 2026 Papers — Page 5
AAAI Conference on Artificial Intelligence · 4149 papers
BokehFlow: Depth-Free Controllable Bokeh Rendering via Flow Matching
Yachuan Huang (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
GenerationData SynthesisVision Language ModelFlow-based ModelAuto EncoderImageText
🎯 What it does: Propose BokehFlow, a depth map input-free Bokeh rendering framework that achieves focus and blur intensity control through text prompts.
Bonsai: Interpretable Tree-Adaptive Grounded Reasoning
Kate Sanders (Johns Hopkins University), Benjamin Van Durme (Johns Hopkins University)
Explainability and InterpretabilityTransformerLarge Language ModelMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the BONSAI system, which can generate interpretable probabilistic reasoning trees based on multimodal evidence.
Boomda: Balanced Multi-objective Optimization for Multimodal Domain Adaptation
Jun Sun, Xiang Gao (Zhejinag Lab)
Domain AdaptationMultimodality
🎯 What it does: This paper proposes an algorithm called Boomda for heterogeneous multi-modal domain adaptation, aiming to address the domain shift problem caused by the scarcity of labeled multi-modal data.
Boosting Adversarial Transferability via Ensemble Non-Attention
Yipeng Zou (Hunan University), Guanghui Ye (University of Electronic Science and Technology of China)
ClassificationAdversarial AttackMeta LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a novel integrated adversarial attack method called NAMEA, which enhances the transferability of adversarial samples across architectures (CNN and ViT) by utilizing the model's non-attention regions during iterative gradient optimization and combining them with meta-learning to fuse gradients.
Boosting ASR Robustness via Test-Time Reinforcement Learning with Audio-Text Semantic Rewards
Linghan Fang (Hong Kong University of Science and Technology), Li Liu (Hong Kong University of Science and Technology)
RecognitionDomain AdaptationTransformerReinforcement LearningPrompt EngineeringContrastive LearningTextAudio
🎯 What it does: Proposed the ASR-TRA framework, inserting learnable prompts into the decoder of the Whisper model, generating multiple candidate transcriptions via temperature-controlled stochastic decoding, and using CLAP-based semantic reward-driven reinforcement learning (RL) updates to achieve unsupervised test-time adaptation.
Boosting Cross-problem Generalization in Diffusion-Based Neural Combinatorial Solver via Inference Time Adaptation
Haoyu Lei (Chinese University of Hong Kong), Farzan Farnia (Huawei Technologies Co., Ltd.)
OptimizationDiffusion modelGraphStochastic Differential Equation
🎯 What it does: This paper proposes a training-free inference-time adaptive framework (DIFU-Ada), achieving zero-shot transfer of trained diffusion-based neural combinatorial optimization (NCO) models to new problems (e.g., Prize-Collecting TSP, Orienteering Problem) and larger-scale instances through energy-guided sampling and recursive renoising-denoising journeys (Guided Langevin Dynamics).
Boosting Fine-Grained Urban Flow Inference via Lightweight Architecture and Focalized Optimization
Yuanshao Zhu (Southern University of Science and Technology), James Jianqiao Yu (Jilin University)
Super ResolutionOptimizationConvolutional Neural NetworkImageTime Series
🎯 What it does: Propose a lightweight Progressive Local-Global Fusion (PLGF) structure and a novel DualFocal Loss to infer fine-grained urban traffic maps from coarse-grained flow maps, addressing issues of large model scale and suboptimal optimization focus.
Boosting Noisy Correspondence Discrimination via Dynamic Neighborhood Semantic Verification
Yu Wang (Tongji University), Jianyu Wang (Tongji University)
RetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a dynamic neighborhood semantic verification framework, DNS, to identify and eliminate noise correspondences in cross-modal retrieval.
Boosting Resolution Generalization of Diffusion Transformers with Randomized Positional Encodings
Liang Hou (Kling Team, Kuaishou Technology), Kun Gai (Kling Team, Kuaishou Technology)
GenerationTransformerDiffusion modelImage
🎯 What it does: Propose a two-dimensional random position encoding (RPE-2D) to address the mismatch in position encoding when scaling up resolution, thereby enhancing the quality of extended generation.
Boosting the Robustness-Accuracy Trade-off of SNNs by Robust Temporal Self-Ensemble
Jihang Wang (Chinese Academy of Sciences), Yi Zeng (Chinese Academy of Sciences)
ClassificationAdversarial AttackSpiking Neural NetworkImage
🎯 What it does: Propose the Robust Temporal Self-Ensemble (RTE) training framework, which treats the temporal sub-networks as an ensemble model, jointly enhancing the robustness of each sub-network and suppressing adversarial vulnerability propagation across time steps, thereby improving the robustness-accuracy trade-off of SNNs.
Bootstrapping LLMs via Preference-Based Policy Optimization
Chen Jia (SI-TECH Information Technology)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Propose a preference-based policy optimization (PbPO) framework that enhances large language models (LLMs) through online iterative RLHF (reinforcement learning with human feedback), transforming the training of reward models and policies into a min-max game involving reward-irrelevant and reward-related exploration.
Bot Meets Shortcut: How Can LLMs Aid in Handling Unknown Invariance OOD Scenarios?
Shiyan Zheng (Xi'an Jiaotong University), Junhang Huang (Beijing Institute of Technology)
ClassificationDomain AdaptationGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: This paper addresses the "shortcut" problem in social robot detection by systematically evaluating the impact of text feature shortcomings on model robustness, and proposes an adversarial data augmentation (CDA) strategy based on large language models to alleviate the shortcomings effect.
BrainHGT: A Hierarchical Graph Transformer for Interpretable Brain Network Analysis
Jiajun Ma (Anhui University), Shengbing Pei (Anhui University)
Anomaly DetectionExplainability and InterpretabilityGraph Neural NetworkTransformerBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Propose a BrainHGT model based on hierarchical graph Transformers for interpretable brain network analysis from local brain regions to global functional communities.
BrainLMM: A Label-Free Framework for Mapping Multi-Semantic Representation in the Human Visual Cortex
Tan Gao (Beijing Institute of Technology), Guoyuan Yang (Beijing Institute of Technology)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerDiffusion modelBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposed the BrainLMM framework, achieving label-free, voxel-based multi-semantic mapping to explore how neurons in the human visual cortex respond to multiple semantic concepts.
Branch, or Layer? Zeroth-Order Optimization for Continual Learning of Vision-Language Models
Ziwei Liu (Sichuan University), Jun Luo (Nanyang Technological University)
OptimizationComputational EfficiencyRepresentation LearningTransformerMixture of ExpertsVision Language ModelImage
🎯 What it does: Propose and verify a hybrid strategy combining zeroth-order optimization (ZO) with first-order optimization (FO) for parameter-efficient visual-language continual learning (VLCL);
BraSTORM: A Dual-Branch Self-Supervised Framework for EEG Representation Learning via Input-Level Spatio-Temporal Decomposition
Yifan Wang (Zhejiang University), Bruce X.B. Yu (Zhejiang University)
Representation LearningConvolutional Neural NetworkTransformerAuto EncoderContrastive LearningBiomedical Data
🎯 What it does: Proposed the BraSTORM framework, achieving EEG representation learning through input-level spatiotemporal decomposition with a dual-branch self-supervised learning approach.
Breadth-First Search vs. Restarting Random Walks for Escaping Uninformed Heuristic Regions
Daniel Platnick (Flybits Labs), Richard Valenzano (Toronto Metropolitan University)
OptimizationBenchmark
🎯 What it does: This paper investigates escape strategies in uninformed heuristic regions (UHR) from both theoretical and experimental perspectives. It compares the expected runtime of breadth-first search (BrFS) with restart random walk (RRW), integrates RRW into the Escalating Hill Climbing (EHC) algorithm, and proposes two variants: EHC-RRW Cℓ and EHC-RRW L. Experimental evaluations are conducted on PDDL planning benchmarks.
Break the Tie: Learning Cluster-Customized Category Relationships for Categorical Data Clustering
Mingjie Zhao (Hong Kong Baptist University), Yiu-ming Cheung (BNU-HKBU United International College)
Tabular
🎯 What it does: This paper proposes a method called DISC, specifically designed for clustering tasks involving discrete (categorical) data, by improving distance metrics through learning subspace class relationships for each cluster, thereby achieving more accurate clustering.
Breaking Alignment Barriers: TPS-Driven Semantic Correlation Learning for Alignment-Free RGB-T Salient Object Detection
Lupiao Hu, Haojie Li (Dalian Minzu University)
Object DetectionTransformerContrastive LearningMultimodality
🎯 What it does: This paper proposes a lightweight salient object detection method for misaligned RGB-T images.
Breaking Barriers, Finding Boundaries: Not Obviously Manipulable Budget-Feasible Mechanism Design
Bart de Keijzer (University of Amsterdam), Carmine Ventre (University of Amsterdam)
Optimization
🎯 What it does: This paper relaxes the traditional strategyproof (SP) requirement in budget feasible mechanism design by adopting the concept of 'non-obviously manipulable (NOM)', and designs a mechanism that achieves the optimal approximation ratio.
Breaking Down Market Barriers: Distilled Prompt-Tuning Approach for Cross-Market Recommendation
Leqi Zhang (Xi'an Jiaotong-Liverpool University), Jia Wang (Xi'an Jiaotong-Liverpool University)
Domain AdaptationRecommendation SystemKnowledge DistillationGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: Propose a distillation-based prompt tuning method (DCMPT) for cross-market recommendation.
Breaking Measurement Barriers: From Compressed Sensing to Deep Reconstruction
Gang Qu (Westlake University), Xin Yuan (Westlake University)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose an asymmetric Kronecker compressive sensing (AKCS) model and a measurement-aware cross-attention (MACA) module, integrating them into MEUNet to achieve high-quality image compressive sensing reconstruction.
Breaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space
Cheng Yan, Yanyong Zhang (University Of Science And Technology Of China)
OptimizationComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the ZeroRouter framework to address the model lock problem in LLM routing, enabling zero-shot rapid integration of new models and intelligent query allocation across multiple model pools.
Breaking One-Size-Fits-All: Revisiting Out-of-Distribution Detection on Graphs Under Diverse Distribution Shifts
Chuancheng Song (Chinese Academy of Sciences), Yanan Cao (Chinese Academy of Sciences)
Anomaly DetectionGraph Neural NetworkGraphStochastic Differential Equation
🎯 What it does: Proposed a unified graph OOD detection framework called UniGOD, which can adaptively select geometric spaces and achieve efficient discrete detection by leveraging neural SDE dynamic uncertainty.
Breaking Task Boundaries: A Unified Model for 3D Medical Image Fusion and Segmentation Guided by Manifold Perspective
Zeyu Wang (Dalian Minzu University), Haiyu Song (Dalian Minzu University)
SegmentationConvolutional Neural NetworkTransformerBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes a unified model M²-CoFS to achieve joint training for 3D medical image fusion and segmentation.
Breaking the Adversarial Robustness-Performance Trade-off in Text Classification via Manifold Purification
Chenhao Dang (Renmin University of China), Jing Ma (China Electronics Technology Group Corporation)
ClassificationAdversarial AttackMixture of ExpertsFlow-based ModelText
🎯 What it does: Proposed the MC F 2 defense framework, which detects anomalies in the sentence embedding space using hierarchical Riemannian continuous normalizing flows, and then purifies the anomalies back to the clean data manifold via geodesic projection;
Breaking the Aggregation Bottleneck in Federated Recommendation: A Personalized Model Merging Approach
Jundong Chen (Key Laboratory of Big Data and Artificial Intelligence in Transportation Ministry of Education China), Yidong Li (Jilin University China)
Recommendation SystemFederated LearningTabular
🎯 What it does: The study addresses the aggregation bottleneck in federated recommendation by proposing FedEM, an elastic model merging approach to enhance personalization.
Breaking the Dyadic Barrier: Rethinking Fairness in Link Prediction Beyond Demographic Parity
Joao Mattos (Rice University), Arlei Silva (Rice University)
Graph Neural NetworkGraph
🎯 What it does: This paper addresses the fairness issue in link prediction by proposing an evaluation framework based on exposure fairness and a post-processing method.
Breaking the Modality Barrier: Generative Modeling for Accurate Molecule Retrieval from Mass Spectra
Yiwen Zhang (Zhejiang University), Huajun Chen (Zhejiang University)
RetrievalDrug DiscoveryTransformerLarge Language ModelContrastive LearningGraphBiomedical Data
🎯 What it does: Proposed a two-stage molecular retrieval framework called GLMR based on generative language models, first retrieving candidate molecules through contrastive learning, then generating and re-ranking molecules using a context-aware generative model.
Breaking the Passive Learning Trap: An Active Perception Strategy for Human Motion Prediction
Juncheng Hu (Jilin University), Kedi Lyu (Jilin University)
Pose EstimationTransformerGenerative Adversarial NetworkTime SeriesSequential
🎯 What it does: Propose an active perception strategy (APS) for 3D human motion prediction, comprising a data perception module (mapping poses to quotient space) and a network perception module (enhancing active learning through noise/occlusion).
Breaking the Stealth-Potency Trade-off in Clean-Image Backdoors with Generative Trigger Optimization
Binyan Xu (Chinese University of Hong Kong), Kehuan Zhang (Zhejiang University)
Adversarial AttackGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Propose a clean image backdoor attack framework named GCB that optimizes the trigger using conditional InfoGAN.
Breaking the Trade-Off Between Faithfulness and Expressiveness for Large Language Models
Chenxu Yang (Chinese Academy of Sciences), Zheng Lin (Chinese Academy of Sciences)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose a collaborative decoding method named CoDe, which breaks the trade-off between reliability and expressiveness in knowledge-driven conversations by dynamically fusing internal parameter distributions with external knowledge conditional distributions.
BREPS: Bounding-Box Robustness Evaluation of Promptable Segmentation
Andrey Moskalenko (Lomonosov Moscow State University), Vlad Shakhuro (Lomonosov Moscow State University)
SegmentationAdversarial AttackPrompt EngineeringImageBiomedical Data
🎯 What it does: This paper evaluates the robustness of prompt-based segmentation models using real user-drawn bounding box annotations and proposes a white-box bounding box adversarial attack method called BREPS to measure the model's sensitivity to variations in box inputs.
BRIC: Bridging Kinematic Plans and Physical Control at Test Time
Dohun Lim (Jeonbuk National University), Sungchan Kim (Jeonbuk National University)
Robotic IntelligenceReinforcement LearningDiffusion modelTextSequential
🎯 What it does: This paper proposes the BRIC framework, achieving test-time adaptive physical controllers to track motion plans generated by diffusion models, and enhancing the physical feasibility and consistency of long-term human motion through lightweight signal space guidance.
BridgeShape: Latent Diffusion Schrödinger Bridge for 3D Shape Completion
Dequan Kong (Nanjing University of Aeronautics and Astronautics), Mingqiang Wei (Nanjing University of Aeronautics and Astronautics)
RestorationGenerationDiffusion modelAuto EncoderPoint CloudMeshStochastic Differential Equation
🎯 What it does: High-quality completion from incomplete 3D shapes to complete shapes is achieved by constructing a Schrodinger bridge model in the latent space; during this process, Depth-Enhanced VQ-VAE is first used to compress 3D shapes into structured latent vectors, followed by optimal transport via potential diffusion Schrodinger bridge in the latent space; finally, the latent codes are decoded back into complete 3D surfaces.
Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment
Henglin Liu (Tsinghua University), Xiangyang Ji (National Cheng Kung University)
TransformerLarge Language ModelVision Language ModelScore-based ModelImageTextMultimodality
🎯 What it does: Proposed the ArtQuant framework, addressing data scarcity and model fragmentation in art image aesthetic evaluation through hierarchical aesthetic description generation and auxiliary description tasks.
Bridging Day and Night: Target-Class Hallucination Suppression in Unpaired Image Translation
Shuwei Li (National University of Singapore), Robby T. Tan (National University of Singapore)
Image TranslationDomain AdaptationTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Propose a framework for day-night unpaired image translation that enhances semantic consistency by detecting and suppressing hallucinations in target categories.
Bridging Granularity Gaps: Hierarchical Semantic Learning for Cross-domain Few-shot Segmentation
Sujun Sun (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)
SegmentationDomain AdaptationConvolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: Propose a Hierarchical Semantic Learning (HSL) framework for cross-domain few-shot segmentation, adapting to differences in segmentation granularity in the target domain by learning hierarchical semantic features;
Bridging Modalities via Progressive Re-alignment for Multimodal Test-Time Adaptation
Jiacheng Li (Beijing Jiaotong University), Songhe Feng (Beijing Jiaotong University)
Domain AdaptationPrompt EngineeringContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Propose a multimodal test-time adaptation framework called BriMPR based on prompt adjustment, which first calibrates monomodal feature distributions using modality-specific prompts, and then enhances modality alignment through cross-modal mask recombination and instance-level contrastive learning.
Bridging Optimization and Neural Networks for Efficient Multi-view Clustering
Huilang Xu (Fuzhou University), Xing Chen (Fuzhou University)
OptimizationComputational EfficiencyRepresentation LearningContrastive LearningMultimodality
🎯 What it does: Propose a lightweight and interpretable BONE framework that utilizes optimization-guided low-level feature extraction and high-level feature learning with a small number of learnable parameters to accomplish multi-view clustering tasks.
Bridging Scale Discrepancies in Robotic Control via Language-Based Action Representations
Yuchi Zhang (Harbin Institute of Technology), Ting Liu (Harbin Institute of Technology)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelVideoTextMultimodality
🎯 What it does: By introducing language-based action representations (motion verbs), normalizing actions in robot control, and employing two-stage pre-training plus fine-tuning to enhance cross-platform and cross-task generalization capabilities.
Bridging Synthetic and Real Routing Problems via LLM-Guided Instance Generation and Progressive Adaptation
Jianghan Zhu (Singapore Management University), Xiaoli Li (Nanyang Technological University)
Data SynthesisDomain AdaptationOptimizationLarge Language ModelSupervised Fine-TuningPrompt EngineeringGraph
🎯 What it does: This paper proposes the EvoReal framework, which generates synthetic data with structures similar to real VRP instances using LLM-driven evolutionary algorithms, and transfers pre-trained neural combinatorial optimization (NCO) models from uniformly distributed synthetic data to real TSPLib and CVRPLib instances through a phased stepwise fine-tuning approach;
Bridging the Copyright Gap: Do Large Vision-Language Models Recognize and Respect Copyrighted Content?
Naen Xu (Zhejiang University), Shouling Ji (Zhejiang University)
RecognitionLarge Language ModelVision Language ModelMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: This study evaluates the detection and compliance capabilities of large vision-language models in handling copyrighted content and constructs a benchmark dataset of 50,000 multimodal question-answer pairs.
Bridging the Language Gap: Uncovering and Aligning Shared Circuits for Multi-Hop Reasoning in Multilingual LLMs
Chenghao Sun (University of Science and Technology of China), Jieping Ye (Independent Researcher)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Track causal paths in multilingual LLMs for multi-hop reasoning through a mechanism-interpretable framework, revealing that factual knowledge is stored in shared language-agnostic semantic neurons, while cross-lingual gaps arise from misaligned attention paths; subsequently propose repairing this gap by fine-tuning only key attention heads.
Bridging the Modality Reliability Gap in Drug-Target Interaction Prediction via a Confidence-aware Multimodal Fusion Framework
Jie Yang (ShanghaiTech University), Zhen Cheng (Shanghai Institute of Materia Medica)
Drug DiscoveryTransformerMultimodalityBiomedical Data
🎯 What it does: Designed a confidence-aware multimodal fusion framework named DrugCMF to address the modality reliability gap in drug-target interaction prediction;
Bridging the Tokenizer Gap: Semantics and Distribution-aware Knowledge Transfer for Unbiased Cross-Tokenizer Distillation
Huazheng Wang (Beijing University of Posts and Telecommunications), Dacheng Tao (AGH University of Krakow)
Knowledge DistillationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper proposes a cross-tokenizer knowledge distillation framework called SEDI, which can efficiently transfer knowledge between teacher and student models even when they use different tokenizers.
Bridging Vision and Language for Robust Context-Aware Surgical Point Tracking: The VL-SurgPT Dataset and Benchmark
Rulin Zhou (Chinese University of Hong Kong), Hongliang Ren (Chinese University of Hong Kong)
Object TrackingTransformerVideoTextMultimodalityBiomedical DataBenchmark
🎯 What it does: Proposed the multimodal surgical point tracking dataset VL-SurgPT and its text-guided tracking model TG-SurgPT, achieving joint visual and linguistic point tracking.
Bring Your Dreams to Life: Continual Text-to-Video Customization
Jiahua Dong (Mohamed bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Mohamed bin Zayed University of Artificial Intelligence)
GenerationConvolutional Neural NetworkDiffusion modelVideoText
🎯 What it does: This paper proposes the Continuous Text-to-Video Customization (CTVC) problem and introduces a Continuous Customization Video Diffusion (CCVD) model capable of continuously learning new concepts while avoiding catastrophic forgetting and concept neglect.
Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO2 Storage
Haoyue Bai (Arizona State University), Yanjie Fu (Arizona State University)
OptimizationReinforcement LearningAuto EncoderContrastive LearningTime SeriesPhysics RelatedStochastic Differential Equation
🎯 What it does: This paper proposes a Brownian bridge-based enhanced agent simulation and injection planning framework for geological CO2 storage.
BTPG-max: Achieving Local Maximal Bidirectional Pairs for Bidirectional Temporal Plan Graphs
Yifan Su (Carnegie Mellon University), Jiaoyang Li (Carnegie Mellon University)
OptimizationGraphBenchmark
🎯 What it does: Propose the BTPG-max algorithm for post-processing the Bidirectional Temporal Plan Graph (BTPG) in multi-agent path planning, maximizing the number of reversible conflict pairs (bi-pairs) to enhance robustness and efficiency during execution.
BugSweeper: Function-Level Detection of Smart Contract Vulnerabilities Using Graph Neural Networks
Uisang Lee (Seoul National University), Soo-Mook Moon (Seoul National University)
ClassificationAnomaly DetectionGraph Neural NetworkGraph
🎯 What it does: Construct a function-level abstract syntax graph (FLAG) and use a two-stage graph neural network to detect vulnerabilities in Solidity contracts, completely without rule-based preprocessing.
BuildingWorld: A Structured 3D Building Dataset for Urban Foundation Models
Shangfeng Huang (University of Calgary), Xin Wang (University of Calgary)
GenerationData SynthesisPoint CloudMeshBenchmark
🎯 What it does: Constructed the BuildingWorld dataset, containing approximately five million LOD2 building models from 44 cities across five continents, along with corresponding field-measured and simulated aerial LiDAR point clouds, and provided the Cyber City procedural generator.
BulletTime4D: Towards High Spatio-Temporal Resolution Dynamic Scene Rendering via Spike-Guided Stereo Vision
Yiqian Chang (Harbin Institute of Technology), Peixi Peng (Peng Cheng Laboratory)
GenerationData SynthesisSpiking Neural NetworkTransformerGaussian SplattingVideo
🎯 What it does: Propose the BulletTime4D framework, which integrates spiking cameras with stereo RGB cameras to achieve dynamic scene rendering at high spatiotemporal resolution;
Burst Image Quality Assessment: A New Benchmark and Unified Framework for Multiple Downstream Tasks
Xiaoye Liang (Beijing University of Aeronautics and Astronautics), Mai Xu (Beijing University of Aeronautics and Astronautics)
Knowledge DistillationTransformerPrompt EngineeringImageBenchmark
🎯 What it does: Propose the Burst Image Quality Assessment (BuIQA) task, construct the first BuIQA benchmark dataset, and design a unified task-aware Prompt-tuning framework to evaluate the quality contribution of each frame in a burst sequence for multiple downstream tasks.
C-GNN-PRUNE: A Unified Graph-Based Framework for Structure-Aware Pruning of Mixture-of-Experts Models
Lin Li (Inner Mongolia University), Zhuopeng Wang (Inner Mongolia University)
Computational EfficiencyGraph Neural NetworkMixture of ExpertsTextBenchmark
🎯 What it does: Propose a unified graph-based structural-aware pruning framework called C-GNN-PRUNE for efficient compression of Mixture-of-Experts (MoE) models while preserving structural and functional diversity.
C3RL: Rethinking the Combination of Channel-independence and Channel-mixing from Representation Learning
Shusen Ma, Yu Kang (University Of Science And Technology Of China)
Representation LearningContrastive LearningTime Series
🎯 What it does: This paper proposes a representation learning framework called C3RL, which jointly utilizes channel mixing (CM) and channel independent (CI) strategies, aligning positive and negative samples in time series data through a SimSiam-style two-branch network;
C³TG: Conflict-aware, Composite, and Collaborative Controlled Text Generation
Yu Li (Southeast University), Guilin Qi (Southeast University)
GenerationTransformerLarge Language ModelAgentic AIPrompt EngineeringText
🎯 What it does: Propose a two-stage C3TG framework that leverages the collaborative work of a large language model and a lightweight attribute classifier to achieve fine-grained control over 17-dimensional attributes, including sentiment, style, tone, topic, and toxicity, and resolves attribute conflicts through iterative optimization.
CABTO: Context-Aware Behavior Tree Grounding for Robot Manipulation
Yishuai Cai (National University of Defense Technology), Minglong Li (National University of Defense Technology)
Robotic IntelligenceLarge Language ModelVision Language ModelMultimodality
🎯 What it does: This paper proposes a framework for automatically constructing a complete and consistent behavior tree (BT) system, formally defines the BT grounding problem, and designs the CABTO method to achieve automatic matching and verification of high-level planning and low-level control strategies for robotic manipulation tasks.
CAD-VAE: Leveraging Correlation-Aware Latents for Comprehensive Fair Disentanglement
Chenrui Ma (University of California, Irvine), Yanning Shen (University of California, Irvine)
GenerationRepresentation LearningAuto EncoderGenerative Adversarial NetworkImageBenchmark
🎯 What it does: Proposes CAD-VAE, a method within the variational autoencoder framework that achieves fair disentanglement between target attributes and sensitive attributes by introducing a co-related latent variable z_R and directly minimizing conditional mutual information, while supporting the generation of fair counterfactuals and fine-grained image editing.
CADiff: Context-Aware Diffusion for Controllable Anomaly Generation in Anomaly Detection
Xuan Tong (Fudan University), Wenqiang Zhang (Fudan University)
Anomaly DetectionDiffusion modelContrastive LearningImage
🎯 What it does: Generate high-quality, diverse industrial anomaly images through the context-aware diffusion framework CADiff, which are used to enhance anomaly detection, localization, and classification.
CADTrack: Learning Contextual Aggregation with Deformable Alignment for Robust RGBT Tracking
Hao Li (Army Engineering University Of Pla), Huchuan Lu (Dalian University Of Technology)
Object TrackingTransformerMixture of ExpertsMultimodality
🎯 What it does: Propose a novel RGBT object tracking framework CADTrack to address issues of modal differences, feature fusion, and spatial mismatch, thereby enhancing all-weather tracking performance.
CAFU: Constrained Alignment and Filtered Uniformity for Denoising Recommendation
Xinzhe Jiang (Anhui University), Yiwen Zhang (Anhui University)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes a recommendation framework named CAFU, specifically designed for AU methods based on alignment and uniformity, achieving denoising and performance improvement in scenarios with missing negative samples and data sparsity.
CAG-GS: Consistent Anchor Guided Gaussian Splatting for Large-scale Scene Rendering
Shijie Xu (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences), Qiulei Dong (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences)
GenerationComputational EfficiencyGaussian Splatting
🎯 What it does: Proposes the Consistent Anchor Guided Gaussian Splatting (CAG-GS) method, which combines learnable anchors with semantic features from the pre-trained semantic segmentation model SAM2 to address the global consistency issue in chunk-based rendering for large-scale scenes.
Calibrating and Rotating: A Unified Framework for Weight Conditioning in PEFT
Da Chang (Pengcheng Laboratory), Shixun Zhang (Pengcheng Laboratory)
OptimizationComputational EfficiencyTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose a unified weight adjustment framework, and based on this framework, introduce two parameter-efficient fine-tuning methods: Pre-Diag and SORA.
CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge
Lei Zan (Huawei Noah's Ark Lab), Lujia Pan (Huawei Noah's Ark Lab)
Explainability and InterpretabilityAI Code AssistantTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: By constructing a mathematical causal graph (MCG) and embedding it into LLM, the accuracy of LLMs in complex mathematical reasoning tasks is enhanced.
CAMAR: Continuous Actions Multi-Agent Routing
Artem Pshenitsyn (CogAI Lab), Alexey Skrynnik (CogAI Lab)
OptimizationReinforcement LearningBenchmark
🎯 What it does: Proposed the CAMAR environment, a multi-agent path planning benchmark that supports continuous actions and runs efficiently on GPUs;
CAMERA: Multi-Matrix Joint Compression for MoE Models via Micro-Expert Redundancy Analysis
Yuzhuang Xu (Harbin Institute of Technology), Wanxiang Che (Harbin Institute of Technology)
CompressionMixture of ExpertsText
🎯 What it does: This paper proposes three compression methods for Mixture-of-Experts (MoE) large language models: CAMERA, CAMERA-P, and CAMERA-Q, which can achieve structured pruning and mixed-precision quantization at the micro-expert level without additional training.
Can Editing LLMs Inject Harm?
Canyu Chen (Illinois Institute of Technology), Kai Shu (Emory University)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose the concept of 'Editing Attack', viewing knowledge editing as a security threat to large language models (LLMs), and construct the EditAttack dataset to systematically evaluate misinformation injection and bias injection.
Can Humans Teach Machines to Code?
Celine Hocquette, Ute Schmid (University Of Bamberg)
AI Code AssistantTransformerLarge Language ModelText
🎯 What it does: In the experiments, human participants (with and without computer backgrounds) were asked to generate input-output examples for six list operation concepts on their own, and these examples were evaluated for their ability to help five program synthesis systems (ILP, Bayesian learning, LLM, etc.) learn programs that achieve high accuracy on unseen data.
Can Molecular Evolution Mechanism Enhance Molecular Representation?
Kun Li (Wuhan University), Jia Wu (Macquarie University)
Drug DiscoveryGraph Neural NetworkTransformerBiomedical Data
🎯 What it does: Construct a molecular evolutionary network MEvoN, and enhance molecular representations using its evolutionary paths and label information for molecular property prediction.
Can Protective Watermarking Safeguard the Copyright of 3D Gaussian Splatting?
Wenkai Huang (Shanghai Jiao Tong University), Tiejun Huang (Peking University)
Safty and PrivacyComputational EfficiencyGaussian SplattingPoint Cloud
🎯 What it does: This paper systematically evaluates the vulnerability of 3D Gaussian Splatting (3DGS) watermarks and proposes the GSPure framework, which utilizes perspective-aware Gaussian weight accumulation and geometric feature clustering to precisely prune watermark-related Gaussian primitives in 3DGS models, thereby removing watermarks while preserving scene visual quality.
Can Pseudo-Label Be More Reliable? A Simple yet Effective Topology-Aware Graph Self-Training Method
Gen Liu, Qingtian Zeng (Shandong University of Science and Technology)
ClassificationKnowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Proposed a topology-aware self-training method called TA-GST, which selects reliable pseudo-labels by leveraging node classification scores and neighbor label distributions.
Can You Tell the Difference? Contrastive Explanations for ABox Entailments
Patrick Koopmann (Vrije Universiteit Amsterdam), Balram Tiwari (Paderborn University)
Explainability and InterpretabilityComputational EfficiencyContrastive LearningGraphBenchmark
🎯 What it does: Proposes a contrastive ABox explanations framework for ABox reasoning in description logic knowledge bases, aiming to answer questions like 'Why is a an instance of C while b is not?', and provides both semantic and syntactic explanation forms;
Cancer Survival Prediction by Cyclic Generation and Multi-grained Alignment
Yongqi Bu (Shandong University), Guoxian Yu (Shandong University)
ClassificationTransformerContrastive LearningBiomedical Data
🎯 What it does: Developed the CIMA framework, utilizing cyclic modal reconstruction and multi-granularity alignment to achieve cancer survival prediction
CANDI: Curated Test-Time Adaptation for Multivariate Time-Series Anomaly Detection Under Distribution Shift
HyunGi Kim (Seoul National University), Sungroh Yoon (Seoul National University)
Domain AdaptationAnomaly DetectionConvolutional Neural NetworkAuto EncoderTime SeriesBenchmark
🎯 What it does: Propose the CANDI framework, integrating false positive mining and spatiotemporal adaptive modules to achieve test-time adaptation in multivariate time series anomaly detection.
CANVAS: A Benchmark for Vision-Language Models on Tool-Based User Interface Design
Daeheon Jeong (KAIST), Juho Kim (KAIST)
Vision Language ModelVision-Language-Action ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed the CANVAS benchmark to evaluate the multi-round interaction capabilities of vision-language models (VLMs) in tool-driven UI design, covering two task categories: design replication and design modification.
CapeNext: Rethinking and Refining Dynamic Support Information for Category-Agnostic Pose Estimation
Yu Zhu (Sun Yat-sen University), Bo Tang (Guilin University of Technology)
Pose EstimationGraph Neural NetworkTransformerVision Language ModelImageText
🎯 What it does: Proposes CapeNext, which dynamically improves keypoint embeddings by leveraging query images and category descriptions to enhance category-agnostic pose estimation.
CaPro: Curvilinear-aware Prompt Learning with Single Unlabeled Image for Cost-effective Curvilinear Structure Segmentation
Zhuangzhuang Chen (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
SegmentationTransformerPrompt EngineeringAuto EncoderImageBiomedical Data
🎯 What it does: Propose a two-stage self-supervised Curvilinear-aware Prompt Learning (CaPro) framework without fine-tuning, which adapts the Segment Anything Model (SAM) to curve structure segmentation tasks under the condition of only being given a single unlabeled image.
Capturing Dynamic User Interests Under Modality Imbalance for Multimodal Sequential Recommendation
Zilong Li (Zhejiang Normal University), Jianxia Ling (Zhejiang Normal University)
Recommendation SystemTransformerMixture of ExpertsGenerative Adversarial NetworkContrastive LearningMultimodalitySequential
🎯 What it does: Proposed the DuAF-MAT framework to address dynamic user interests, modality imbalance, and data sparsity in multimodal sequence recommendation.
CareCom: Generative Image Composition with Calibrated Reference Features
Jiaxuan Chen (Shanghai Jiao Tong University), Li Niu (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelAuto EncoderImage
🎯 What it does: Propose CareCom, a multi-reference image generative image synthesis framework, enhancing foreground detail preservation and pose/view alignment through global and local feature calibration.
Careful Queries, Credible Results: Teaching RAG Models Advanced Web Search Tools with Reinforcement Learning
Yuqin Dai (Tsinghua University), Shuai Lu (Tsinghua University)
RetrievalTransformerLarge Language ModelReinforcement LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed the WebFilter framework, modeling the retrieval process as a Markov Decision Process (MDP), enabling large language models (LLMs) to autonomously generate precise queries with advanced search operators (e.g., site:, after:, AND, OR), filtering online rumors and improving retrieval accuracy.
CART: Compositional AutoRegressive Transformer for Image Generation
Siddharth Roheda (Samsung Research Institute Bangalore), Rohan Jaiswal (Samsung Research Institute Bangalore)
GenerationSuper ResolutionTransformerAuto EncoderImage
🎯 What it does: Proposed a hierarchical decomposition-based autoregressive image generation framework (CART), which achieves image synthesis by first generating a structured base image and then progressively overlaying details.
CASL: Curvature-Augmented Self-supervised Learning for 3D Anomaly Detection
Yaohua Zha (Tsinghua Shenzhen International Graduate School, Tsinghua University), Shu-Tao Xia (Tsinghua Shenzhen International Graduate School, Tsinghua University)
ClassificationSegmentationAnomaly DetectionRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningAuto EncoderPoint Cloud
🎯 What it does: Proposed a curvature-enhanced self-supervised learning framework named CASL for 3D point cloud anomaly detection and general point cloud tasks;
CasMoE: A Cascaded Framework for Efficient MoE Inference on Resource-constrained Devices
Chengcheng Wang (University of Electronic Science and Technology of China), Shaohua Wan (University of Electronic Science and Technology of China)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Propose the CasMoE framework, which uses offline calibration + online cascading to pre-predict and pre-fetch MoE experts on resource-constrained devices, thereby improving inference efficiency.
CAST-LUT: Tokenizer-Guided HSV Look-Up Tables for Purple Flare Removal
Pu Wang (Shandong University), Zhuoran Zheng (Zaozhuang University)
RestorationData SynthesisTransformerAuto EncoderImageBenchmark
🎯 What it does: Propose the CAST-LUT network, which removes purple flare by leveraging the HSV color space and token-guided 1D LUT;
CastX: Cohort-Level Causal Inference Meets Statistical Testing for Faithful and Reliable GNN Explanations
Guanyuan Yu, Gang Kou (Southwestern University Of Finance And Economics)
Explainability and InterpretabilityGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Propose the CastX framework, combining collective layer causal inference with non-parametric permutation tests, using reinforcement learning to iteratively remove edges and generate credible and concise GNN explanation subgraphs.
CaT-Diff: Cascaded Text-enhanced Diffusion Model for Time-Series Imputation
Changjian Xu (University of Electronic Science and Technology of China), Kexin Li (University of Electronic Science and Technology of China)
RestorationMixture of ExpertsDiffusion modelMultimodalityTime Series
🎯 What it does: Proposes a multimodal information-integrated diffusion model, CaT-Diff, for probabilistic imputation of multivariate time series under MNAR (Missing Not At Random) scenarios.
CAT-Net: A Cross-Attention Tone Network for Cross-Subject EEG-EMG Fusion Tone Decoding
Yifan Zhuang (Sony Interactive Entertainment), Jiawei Ju (Shanghai Center for Brain Science and Brain-Inspired Technology)
ClassificationRecognitionDomain AdaptationRecurrent Neural NetworkTransformerMultimodalityBiomedical Data
🎯 What it does: This paper proposes a Chinese four-tone recognition framework based on EEG and EMG bimodal fusion (CAT-Net), achieving tone classification under both audible and silent speech conditions;
CATAL: Causally Disentangled Task Representation Learning for Offline Meta-Reinforcement Learning
Shan Cong (Sun Yat-sen University), Xiangyuan Lan (Sun Yat-sen University)
Meta LearningReinforcement LearningAuto EncoderSequential
🎯 What it does: This paper proposes the CATAL method, which learns causally disentangled task representations in context-free unsupervised offline meta-reinforcement learning to enhance cross-task generalization.
Catastrophic Forgetting in Kolmogorov-Arnold Networks
Mohammad Marufur Rahman (Wake Forest University), Fan Yang (Wake Forest University)
TransformerLarge Language ModelImageTextBenchmark
🎯 What it does: Systematic theoretical analysis and experimental validation of catastrophic forgetting in Kolmogorov-Arnold networks (KAN) for continual learning, and the proposal of the KAN-LoRA adapter for continual editing of language models.
CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation
Rui Ke (Chinese University of Hong Kong), Haizhou Li (Chinese University of Hong Kong)
Recommendation SystemTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextSequentialFinance Related
🎯 What it does: Proposes the controllable topic detection framework CATCH, achieving cross-dialogue consistent and personalized topic identification through context-aware topic segmentation, user preference-enhanced clustering, and hierarchical topic generation.
CaTFormer: Causal Temporal Transformer with Dynamic Contextual Fusion for Driving Intention Prediction
Sirui Wang (Beijing Jiaotong University), Jie Liu (Beijing Jiaotong University)
Autonomous DrivingTransformerVideoTime Series
🎯 What it does: Proposes CaTFormer, a causal temporal Transformer for driving intent prediction, explicitly modeling the mutual causal relationships between driver behavior and the external environment.
CATP: Contextually Adaptive Token Pruning for Efficient and Enhanced Multimodal In-Context Learning
Yanshu Li (Brown University), Ruixiang Tang (Rutgers University)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Designed and implemented a two-stage, no-training image token pruning method called Contextually Adaptive Token Pruning (CATP) for eliminating image token redundancy in multi-modal in-context learning (ICL).
CATS: Category-Aware Token-level Steering for Training-Free Redundancy Reduction in Large Reasoning Models
Mengfei Zhang (Zhejiang University), Zhenglin Wang (Southeast University)
CompressionExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose CATS, a training-agnostic, lightweight category-based token-level activation-driven method for compressing redundant reasoning chains in large inference models.
Causal Decoupling Domain Generalization for Remote Sensing Change Detection
Jiaqi Zhao (China University of Mining and Techology), Rui Yao (China University of Mining and Techology)
SegmentationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Proposed a causal disentanglement-based remote sensing change detection network, CDDGNet, for unsupervised change detection in cross-domain environments;
Causal Discovery from Interval-Based Event Sequences
Lénaïg Cornanguer (CISPA Helmholtz Center for Information Security), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
OptimizationExplainability and InterpretabilityData-Centric LearningTime SeriesSequentialBiomedical DataElectronic Health Records
🎯 What it does: For event sequences with durations, discover causal relationships from observational data and build interpretable structured causal models.
Causal Inference Under Threshold Manipulation: Bayesian Mixture Modeling and Heterogeneous Treatment Effects
Kohsuke Kubota (NTT DOCOMO, INC.), Shonosuke Sugasawa (Keio University)
Tabular
🎯 What it does: This paper proposes two methods based on Bayesian mixture models, BMTM and HBMTM, to estimate the causal effect of thresholds on consumer spending behavior in marketing scenarios with threshold manipulation, and further estimates heterogeneous effects across different consumer subgroups through a hierarchical structure.
Causal Reward Adjustment: Mitigating Reward Hacking in External Reasoning via Backdoor Correction
Ruike Song (Institute of Software Chinese Academy of Sciences), Wenwen Qiang (Institute of Software Chinese Academy of Sciences)
Explainability and InterpretabilityAuto EncoderText
🎯 What it does: Proposes the Causal Reward Adjustment (CRA) method using a causal inference framework and backdoor correction, leveraging a sparse autoencoder to extract interpretable semantic features and eliminate reward hacking issues, thereby improving the accuracy of external reasoning systems.
Causal Structure Learning for Dynamical Systems with Theoretical Score Analysis
Nicholas Tagliapietra (Bosch Center for Artificial Intelligence), Osman Mian (Bosch Center for Artificial Intelligence)
Score-based ModelTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: Proposed an algorithm named CADYT that uses continuous-time Gaussian process models and information-theoretic methods for causal structure learning in dynamic systems.
Causal Tracing of Object Representations in Large Vision Language Models: Mechanistic Interpretability and Hallucination Mitigation
Qiming Li (Harbin Institute of Technology), Xiachong Feng (Harbin Institute of Technology)
Explainability and InterpretabilityTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a fine-grained cross-modal causal tracing framework (FCCT) to quantify the causal effects of visual objects in large audio-visual language models, and designs a training-agnostic inference-time intermediate representation injection method (IRI) to alleviate object hallucinations.