AAAI 2026 Papers — Page 10
AAAI Conference on Artificial Intelligence · 4149 papers
DiffBench Meets DiffAgent: End-to-End LLM-Driven Diffusion Acceleration Code Generation
Jiajun Jiao (Advanced Micro Devices, Inc.), Emad Barsoum (Advanced Micro Devices, Inc.)
GenerationComputational EfficiencyLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: Proposes DiffBench benchmark and DiffAgent framework, enabling large language models (LLM) to automatically generate and optimize code for accelerating diffusion model inference.
Difference Vector Equalization for Robust Fine-tuning of Vision-Language Models
Satoshi Suzuki (NTT, Inc.), Ryo Masumura (NTT, Inc.)
ClassificationRetrievalRepresentation LearningSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Under the premise of maintaining the generalization ability of pre-trained vision-language models, perform robust fine-tuning to prevent distortion of the geometric structure of the embedding space during fine-tuning.
Differentiable Semantic Meta-Learning Framework for Long-Tail Motion Forecasting in Autonomous Driving
Bin Rao (University of Macau), Zhenning Li (University of Macau)
Autonomous DrivingExplainability and InterpretabilityComputational EfficiencySequential
🎯 What it does: Proposes SAML, a long-tail motion prediction framework based on differentiable semantic meta-learning, specifically addressing rare and safety-critical motion events in autonomous driving scenarios.
Differentiable Sparse Identification of Lagrangian Dynamics
Zitong Zhang (Renmin University of China), Hao Sun (Renmin University of China)
OptimizationTime SeriesSequentialPhysics Related
🎯 What it does: Proposed a new framework for differentiable sparse identification of Lagrangian dynamics, combining cubic B-spline approximation with sparse regression to jointly learn the equations of motion for mechanical systems.
Differentially Private Linear Programming: Reduced Sub-Optimality and Guaranteed Constraint Satisfaction
Alexander Benvenuti (Georgia Institute of Technology), Matthew Hale (Georgia Institute of Technology)
OptimizationSafty and PrivacyTabular
🎯 What it does: To address data sensitivity in linear programming (LP), this paper proposes a method that simultaneously provides differential privacy protection for the constraint matrix A, constraint vector b, and objective coefficient c, and proves that the method ensures the feasibility of the original non-private constraints while preserving privacy.
Differentially Private Subspace Fine-Tuning for Large Language Models
Lele Zheng (Xidian University), Yulong Shen (Institute of Science Tokyo)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Introduce differential privacy into the fine-tuning process of large language models, proposing the DP-SFT method that injects noise only within a task-specific low-dimensional subspace;
Difficulty Controlled Diffusion Model for Synthesizing Effective Training Data
Zerun Wang (CyberAgent), Toshihiko Yamasaki (University of Tokyo)
ClassificationData SynthesisSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Proposed a diffusion model based on difficulty control, using learning difficulty as a conditional signal to synthesize training data.
Difficulty Is Not Enough: Curriculum Learning for LLMs Fine-tuning Must Consider Utility
Zishang Jiang (Fudan University), Yanghua Xiao (Fudan University)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposes a dual-evaluation curriculum learning framework called DUCL for LLM fine-tuning, which prioritizes low-difficulty, high-benefit samples during early training and gradually introduces high-difficulty samples later.
Difficulty-Aware Label-Guided Denoising for Monocular 3D Object Detection
Soyul Lee (Ewha Womans University), Dongbo Min (Ewha Womans University)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: MonoDLGD provides explicit geometric supervision to enhance monocular 3D object detection by applying difficulty-aware perturbation and reconstruction to 3D object labels during training;
Difficulty-Aware Learning Curve Extrapolation
Mengyang Li (Tianjin Normal University), Pinlong Zhao (Hangzhou Dianzi University)
Data SynthesisOptimizationHyperparameter SearchConvolutional Neural NetworkTransformerDiffusion modelTime SeriesBenchmark
🎯 What it does: Proposed a task difficulty-aware learning curve extrapolation framework DA-LCE, which achieves precise extrapolation of learning curves in automated machine learning by dynamically quantifying difficulty through early curves, generating synthetic data via conditional diffusion, and employing Transformer predictors.
DiffMM: Efficient Method for Accurate Noisy and Sparse Trajectory Map Matching via One Step Diffusion
Chenxu Han (East China Normal University), Jilin Hu (East China Normal University)
Autonomous DrivingComputational EfficiencyTransformerDiffusion modelTime SeriesSequential
🎯 What it does: Propose the DiffMM framework, using single-step diffusion + shortcut model for GPS trajectory map matching;
DiffNR: Diffusion-Enhanced Neural Representation Optimization for Sparse-View 3D Tomographic Reconstruction
Shiyan Su (Monash University), Xuelian Cheng (Hong Kong Polytechnic University)
OptimizationConvolutional Neural NetworkSupervised Fine-TuningDiffusion modelNeural Radiance FieldAuto EncoderGaussian SplattingBiomedical DataComputed Tomography
🎯 What it does: Propose a fusion framework for sparse-view 3D CT reconstruction, DiffNR, combining neural representations with diffusion priors. Design a single-step diffusion model, SliceFixer, to repair slice artifacts in neural representation reconstructions. Subsequently, use the repaired pseudo-reference volume as 3D structural-aware regularization to enhance sparse-view reconstruction quality.
DiffOP: Reinforcement Learning of Optimization-Based Control Policies via Implicit Policy Gradients
Yuexin Bian (University of California, San Diego), Yuanyuan Shi (University of California, San Diego)
OptimizationReinforcement Learning
🎯 What it does: Propose the DiffOP framework, which employs implicit gradients to achieve reinforcement learning optimization control strategies without value function approximation, jointly learning cost and dynamics models;
DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving
Liuhan Yin (Polytechnic Institute, Zhejiang University), Erkang Cheng (Nullmax)
Autonomous DrivingTransformerDiffusion model
🎯 What it does: Propose DiffRefiner, a two-stage end-to-end autonomous driving trajectory prediction framework that first generates coarse-grained trajectory candidates using a Transformer, then refines them with a diffusion model to satisfy scene constraints.
Diffusion Distillation with Direct Preference Optimization for Efficient 3D LiDAR Scene Completion
An Zhao (Zhejiang University), Lingyun Sun (Zhejiang University)
GenerationAutonomous DrivingComputational EfficiencyKnowledge DistillationDiffusion modelScore-based ModelContrastive LearningPoint Cloud
🎯 What it does: Propose a framework combining score distillation with Direct Preference Optimization (Distillation-DPO) for efficient 3D LiDAR scene completion;
Diffusion Implicit Policy for Unpaired Scene-aware Motion Synthesis
Jingyu Gong (East China Normal University), Yuan Xie (East China Normal University)
GenerationData SynthesisReinforcement LearningDiffusion modelVideoMesh
🎯 What it does: Propose Diffusion Implicit Policy (DIP), achieving scene-aware action synthesis without requiring paired scene-action data;
Diffusion Model Based Signal Recovery Under 1-Bit Quantization
Youming Chen (University of Electronic Science and Technology of China), Zhaoqiang Liu (University of Electronic Science and Technology of China)
RestorationDiffusion modelImage
🎯 What it does: Propose a 1-bit quantized signal recovery method based on diffusion models, named Diff-OneBit, which can simultaneously address 1-bit compressed sensing and logistic regression problems;
Diffusion Once and Done: Degradation-Aware LoRA for All-in-One Image Restoration
Ni Tang (Xiamen University), Yanyun Qu (Xiamen University)
RestorationDiffusion modelImage
🎯 What it does: Proposes a one-stage, one-sampling stable diffusion model called Diffusion Once and Done (DOD) for universal image restoration across multiple degradation types.
Diffusion Reconstruction-based Data Likelihood Estimation for Core-Set Selection
Mingyang Chen (Hong Kong University of Science and Technology), Wei Wang (Dongguan University of Technology)
Data-Centric LearningDiffusion modelImage
🎯 What it does: Proposes a core set selection method based on diffusion reconstruction, improving the effectiveness of core set selection by estimating data likelihood through reconstruction bias.
Diffusion-Assisted Progressive Learning for Weakly Supervised Phrase Localization
Pengyue Lin (Beijing University of Posts and Telecommunications), Ruifan Li (Beijing University of Posts and Telecommunications)
Object DetectionGenerationVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposes the DAPO (Diffusion-Assisted Progressive Learning) framework, which first generates images of scenes with few objects using Stable Diffusion, enhances semantic alignment through attention guidance and diffusion-guided loss, and subsequently trains a weakly supervised phrase localization network in a progressive manner based on sample difficulty.
Diffusion-Based Contextual Reconstruction for Point Cloud Segmentation with Limited Annotations
Jiawei Lian (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
SegmentationConvolutional Neural NetworkDiffusion modelPoint Cloud
🎯 What it does: Proposed the DiCoSeg framework, which utilizes diffusion models to perform contextual semantic reconstruction on sparsely annotated point clouds, recovering complete segmentation information from noise.
Diffusion-based Personalized Pathology Disentanglement for Impaired Gait Analysis
Xiaoyue Wan (Shanghai Jiao Tong University), Xu Zhao (Shanghai Jiao Tong University)
ClassificationAnomaly DetectionTransformerDiffusion modelContrastive LearningBiomedical DataAlzheimer's DiseaseBenchmark
🎯 What it does: Propose a diffusion-based personalized pathology disentanglement model, DPPD, which can simultaneously perform visual gait scoring, dementia subtyping classification, and visualize abnormal joint localization.
Diffusion-calibrated Continual Test-time Adaptation
Xu Yang (Xidian University), Kun Wei (Xidian University)
Domain AdaptationDiffusion modelContrastive LearningImage
🎯 What it does: In continuous test-time adaptation (CTTA), the authors propose using diffusion models to generate calibration samples, selecting reliable pseudo-labels through cross-validation, and online optimizing the model via adaptive weighting and contrastive learning.
DiffusionPose: Markov-Optimized Diffusion Model for Human Pose Estimation
Zhigang Wang (Zhejiang University), Yingying Jiao (Zhejiang University of Technology)
Pose EstimationTransformerDiffusion modelVideo
🎯 What it does: Proposes a video human pose estimation framework called DiffusionPose, which combines diffusion models with a Mamba-Transformer hybrid encoder and a decoder based on Markov Random Fields (MRF), addressing issues such as slow convergence and unstable pose generation in traditional diffusion models;
DIFT: Protecting Contrastive Learning Against Data Poisoning Backdoor Attacks
Jiang Zhu (Hong Kong Polytechnic University), Haibo Hu (Hong Kong Polytechnic University)
Contrastive LearningImage
🎯 What it does: Proposed the DIFT framework, which actively identifies and removes poisoned samples during the Contrastive Learning training process, and reduces their backdoor effects on the model through fine-tuning;
DigimonGPT: An Evolvable Agent with Hierarchical Human-like Memory for Video Question Answering
Borui Li (Southeast University), Shuai Wang (Southeast University)
RetrievalTransformerLarge Language ModelVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Designed an evolutionary multi-modal video question-answering agent, DigimonGPT, which integrates video-internal declarative memory, cross-task procedural memory, and a hierarchical memory replay mechanism based on question complexity.
DiLO: Disentangled Latent Optimization for Learning Shape and Deformation in Grouped Deforming 3D Objects
Mostofa Rafid Uddin (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
GenerationExplainability and InterpretabilityComputational EfficiencyRepresentation LearningConvolutional Neural NetworkAuto EncoderPoint CloudMesh
🎯 What it does: Propose an unsupervised two-stage method called DiLO to decompose grouped deforming 3D objects into shape and deformation as two generative factors, achieving generator training through a self-decoder and AdaIN, followed by fast inference using a PointNet encoder;
DiM-TS: Bridge the Gap Between Selective State Space Models and Time Series for Generative Modeling
Zihao Yao (Tongji University), Yaying Zhang (Tongji University)
GenerationData SynthesisDiffusion modelTime SeriesFinance Related
🎯 What it does: Propose the DiM-TS framework, applying the selective state space model (Mamba) to unsupervised time series generation, and achieving joint modeling of temporal and channel dimension features through a dual-channel encoder-decoder structure.
DiMA: Distinguishing Resident and Tourist Preferences via Multi-Modal LLM Alignment for Out-of-Town Cross-Domain Recommendation
Fan Zhang (Beijing University of Posts and Telecommunications), Zhenye Yang (Jinan University)
Recommendation SystemKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposed DiMA, a re-ranking framework for city recommendation, which separates and re-ranks residents' and tourists' preferences through multimodal alignment, chain-of-thought reasoning, and teacher-student alignment.
Dimension-Aware Active Annotation for Aesthetic Perception via Multi-Agent Human–AI Collaboration
Ye Zhang (Northeast Normal University), Minghao Yin (Northeast Normal University)
OptimizationData-Centric LearningConvolutional Neural NetworkGraph Neural NetworkTransformerReinforcement LearningAgentic AIImage
🎯 What it does: Proposed the HAPA (Human-AI Collaborative Painting Annotation) framework, which utilizes a multi-agent active learning strategy to automatically identify and provide expert annotations only for the most difficult-to-predict aesthetic dimensions, thereby enhancing the accuracy of aesthetic evaluation models while significantly reducing manual annotation costs.
DIMM: Decoupled Multi-hierarchy Kalman Filter via Reinforcement Learning
Jirong Zha (Tsinghua University), Xinlei Chen (Tsinghua University)
Object TrackingAutonomous DrivingReinforcement LearningTime SeriesSequential
🎯 What it does: Proposes the DIMM framework, combining a 3D decoupled multi-level Kalman filter with an adaptive fusion network based on reinforcement learning, achieving high-precision tracking of unknown dynamic targets.
DIN: Dual Impulse Network for Multi-view Representation Learning
Yilin Wu, Shiping Wang (Fuzhou University)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Dual Impulse Network (DIN), which extracts information from both the attribute axis and channel axis, and generates an attention matrix through an integration network to achieve stronger multi-view representations;
DipGuava: Disentangling Personalized Gaussian Features for 3D Head Avatars from Monocular Video
Jeonghaeng Lee (Yonsei University), Sanghoon Lee (Yonsei University)
GenerationConvolutional Neural NetworkGaussian SplattingVideo
🎯 What it does: Proposes a two-stage 3D Gaussian head avatar method called DipGuava, capable of generating highly realistic and animatable 3D avatars with personalized details from monocular videos;
Directing Uncertainty-Aware Information Flow for Robust Diffusion Prediction
Weikang He (Chongqing University of Posts and Telecommunications), Qian Li (Chongqing University of Posts and Telecommunications)
Graph Neural NetworkTransformerContrastive LearningGraph
🎯 What it does: Propose the SIEVE framework, enhancing representation robustness through controlled uncertainty injection and contrastive learning, and improve information diffusion prediction by introducing uncertainty-aware directed aggregation.
Direction Sensitivity–Based Knowledge Distillation: Optimization-Aware Low-Rank Knowledge Transfer
Yongkai Liao (Huazhong University of Science and Technology), Jian Huang (Huazhong University of Science and Technology)
OptimizationKnowledge DistillationImageTextBenchmark
🎯 What it does: This paper proposes a direction-sensitivity-based knowledge distillation method called DSKD, which achieves more efficient low-rank knowledge transfer by dynamically selecting singular directions most sensitive to loss.
DISC: Dynamic Feature Selection for Cost-Sensitive Medical Diagnosis
Yu-sheng Li (Nanjing University), Han-jia Ye (Nanjing University)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsTextMultimodalityTabularElectronic Health Records
🎯 What it does: This paper proposes a dynamic feature selection framework called DISC, which gradually selects experiments during the medical diagnosis process according to the cost sensitivity of individual patients, thereby reducing resource consumption while maintaining diagnostic accuracy.
DisCo DETR: Distance-aware Multi-view Contrastive Learning for DETR Pre-training
Chao Ouyang, David Wenzhong Gao (Wuhan University)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: Propose the DisCo DETR self-supervised pre-training framework, enhancing DETR's localization and semantic feature learning through distance-aware multi-view object query fusion and contrastive learning.
DISCODE: Distribution-Aware Score Decoder for Robust Automatic Evaluation of Image Captioning
Nakamasa Inoue (Institute of Science Tokyo), Yusuke Sekikawa (Institute of Science Tokyo)
Large Language ModelVision Language ModelScore-based ModelMultimodalityBenchmark
🎯 What it does: Propose DISCODE, an adaptive decoder for LVLM inference that achieves robust evaluation scores using Gaussian prior through ATT loss; and introduce MCEval, a benchmark covering 18k image-text pairs across 6 visual domains;
Discounted Cuts: A Stackelberg Approach to Network Disruption
Pål Grønås Drange (University of Bergen), Danil Sagunov (ITMO University)
OptimizationGraph
🎯 What it does: This paper proposes the Discounted Cut model, studying the network disruption problem where the defender reroutes remaining traffic after the attacker deletes up to k edges in an attacker-defender Stackelberg game; and reduces it to finding the minimum/maximum cut price after removing the k most expensive or cheapest edges in the network.
Discovering Decoupled Functional Modules in Large Language Models
Yanke Yu (Hong Kong University of Science and Technology), Yi Zheng (Huawei Technologies Ltd)
Explainability and InterpretabilityLarge Language ModelText
🎯 What it does: Proposed an unsupervised LLM cross-layer module discovery framework (ULCMOD), aiming to simultaneously decouple neurons in large language models and discover input sample topics associated with these modules.
Discovering Latent Facts from Context to Construct Richer Open Knowledge Graphs
Jinpeng Li (National Deep Sea Center), Peng Qi (Shanghai University)
Adversarial AttackTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Construct an end-to-end open knowledge graph construction framework KG-DLF, capable of extracting facts from single-text segments, performing pattern normalization, discovering implicit facts across texts through clue retrieval, and conducting knowledge error correction via adversarial reasoning.
Discovering Mixture Skills for Unsupervised Reinforcement Learning
Nelson Ma (University of Technology Sydney), Jie Lu (University of Technology Sydney)
Representation LearningReinforcement LearningAuto EncoderTabularBenchmark
🎯 What it does: This paper proposes the DiMS method, which generates a clusterable mixture of skills through joint learning of an online-trained Gaussian Mixture Variational Auto-Encoder (GMVAE) and an unsupervised policy in reward-free environments.
Discrete Structure Augmentation for Graph Convolutional Networks
Jianxin Ren (Harbin Engineering University), Weining Wu (Harbin Engineering University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: Propose a graph convolutional network architecture named DSAGCN, which jointly learns discrete graph structures and label dependencies to achieve semi-supervised node classification under sparse or missing graph structures.
Discrete-Guided Diffusion for Scalable and Safe Multi-Robot Motion Planning
Jinhao Liang (University of Virginia), Ferdinando Fioretto (University of Virginia)
OptimizationRobotic IntelligenceDiffusion model
🎯 What it does: Integrate discrete multi-agent path planning with continuous diffusion models to propose the Discrete-Guided Diffusion (DGD) framework for large-scale multi-robot motion planning;
Discretization Is Not Always Better: Rethinking Deep Quantization for Asymmetric Image Retrieval
Xinze Liu (Institute of Information Engineering, Chinese Academy of Sciences), Pengwen Dai (Baidu Research)
RetrievalConvolutional Neural NetworkImage
🎯 What it does: In resource-constrained scenarios, a novel deep cross-modal alignment hashing framework (DCAH) is proposed, which enhances cross-modal image retrieval performance through implicit binary quantization of the query model via a correlation-aligned quantization strategy (CAQ).
Discriminative Graph Embedding Framework via Label-Free Marginal Fisher Analysis
Qianqian Wang (Xidian University), Bin Liu (Northwest A&F University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: Propose an unsupervised discriminative graph embedding framework called DGEF, which unifies unsupervised dimensionality reduction and clustering through center-free K-Means and MAF.
Disentangled Generation-Based Prototypical Alignment for Few-Shot Unsupervised Domain Adaptation in Graph-Level Anomaly Detection
Zhibin Ni (Tsinghua University), Xibin Zhao (Tsinghua University)
Domain AdaptationAnomaly DetectionGraph Neural NetworkGraphBenchmark
🎯 What it does: Proposes the DGPA method to address the few-shot unsupervised domain adaptation problem in graph-level anomaly detection, consisting of two modules: discretized sample generation and graph-based prototype self-supervised alignment;
Disentangled Hypergraph-Guided Mamba Scanning for Fine-Grained Visual Recognition
Zhongwei Xiong (Xiamen University), Taisong Jin (Xiamen University)
ClassificationRecognitionImage
🎯 What it does: A fine-grained visual recognition model named DHMamba based on Mamba was studied, which enhances the ability to capture fine-grained differences through hypergraph guidance and decoupled scanning.
Disentangling Adversarial Prompts: A Semantic-Graph Defense for Robust LLM Security
Xiang Fang (Huazhong University of Science and Technology), Wanlong Fang (Nanyang Technological University)
Adversarial AttackGraph Neural NetworkTransformerAuto EncoderText
🎯 What it does: Proposed a defense framework named APD, which can decompose input prompts into harmful and normal components and purify them before processing by LLMs;
Disentangling for Transfer: Boosting Limited Modalities via Information-Theoretic Regularization and Cross-Modal Reconstruction
Zhiyun Zhang (DAMO Academy, Alibaba Group), Minfeng Xu (Zhejiang University)
SegmentationConvolutional Neural NetworkAuto EncoderContrastive LearningMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose the DANTE framework, utilizing a two-stage decoupled and alignment learning approach to address the problem of missing key modalities in medical imaging, thereby improving segmentation performance when using limited modalities.
Distillation Dynamics: Towards Understanding Feature-Based Distillation in Vision Transformers
Huiyuan Tian (Zhejiang University), Shijian Li (Zhejiang University)
Knowledge DistillationTransformerImage
🎯 What it does: Investigated why feature-level knowledge distillation fails on Vision Transformers and proposed a multi-dimensional analysis framework called 'distillation dynamics'.
Distillation-Guided Structural Transfer for Continual Learning Beyond Sparse Distributed Memory
Huiyan Xue (Dalian University of Technology), Qiang Zhang (Dalian University of Technology)
Knowledge DistillationRepresentation LearningImage
🎯 What it does: This paper proposes a structure-oriented sparse subnetwork distillation framework called SSD, aimed at achieving cross-task knowledge transfer in sparse neural networks and alleviating catastrophic forgetting in continual learning.
Distilling Cross-Modal Knowledge via Feature Disentanglement
Junhong Liu (Beihang University), Renyu Yang (Beihang University)
Knowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerImageTextMultimodalityAudio
🎯 What it does: This paper proposes a frequency-domain decomposition-based cross-modal knowledge distillation method (FD-CMKD), which first decomposes the teacher and student features into low-frequency (cross-modal common information) and high-frequency (modality-specific information) components via Fourier transform, and then distills them using strong consistency (MSE) and weak consistency (logMSE) losses respectively, while incorporating feature normalization and a shared classifier to achieve scale and spatial alignment.
Distilling Future Temporal Knowledge with Masked Feature Reconstruction for 3D Object Detection
Haowen Zheng (Macau University of Science and Technology), Yanyan Liang (Macau University of Science and Technology)
Object DetectionAutonomous DrivingKnowledge DistillationTransformerImagePoint Cloud
🎯 What it does: Leverage an offline teacher model to distill knowledge from future frames, enabling the online student model to obtain richer temporal information without access to future frames, thereby improving the accuracy of 3D object detection and speed estimation.
Distribution Shift Is Key to Learning Invariant Prediction
Hong Zheng (Southwest Jiaotong University), Fei Teng (Southwest Jiaotong University)
Domain AdaptationImage
🎯 What it does: Investigates the role of distribution shift in domain generalization, proving that when the distribution shift between training domains is significant, empirical risk minimization (ERM) can approximate an invariant predictive model, supported by both theoretical and experimental validation.
Distribution-Based Feature Attribution for Explaining the Predictions of Any Classifier
Xinpeng Li (Nanjing University), Kai Ming Ting (Nanjing University)
Explainability and InterpretabilityComputational EfficiencyImageTextMultimodalityTabularBiomedical Data
🎯 What it does: Proposed a model-agnostic feature attribution method called DFAX based on data distribution differences, and provided a formal definition of the feature attribution problem.
Distributional Priors Guided Diffusion for Generating 3D Molecules in Low Data Regimes
Haokai Hong (Hong Kong Polytechnic University), Kay Chen Tan (Hong Kong Polytechnic University)
GenerationData SynthesisDrug DiscoveryGraph Neural NetworkDiffusion modelAuto EncoderGraphBiomedical Data
🎯 What it does: This paper proposes a geometric OOD diffusion model called GODD, based on distributed structural priors, to generate 3D molecules in data-scarce regions that satisfy validity, uniqueness, and novelty.
Distributionally Robust Online Markov Game with Linear Function Approximation
Zewu Zheng (Chinese University of Hongkong), Yuanyuan Lin (Chinese University of Hongkong)
Reinforcement LearningSequentialBenchmark
🎯 What it does: For multi-agent online distributed robust Markov games, a sample-efficient algorithm named DR-CCE-LSI based on linear function approximation is proposed to find an ε-approximate robust coarse correlated equilibrium.
Disturbance-based Discretization, Differentiable IDS Channel, and an IDS-Correcting Code for DNA-based Storage
Alan J.X. Guo (Tianjin University), Pengchen Zhang (Tianjin University)
TransformerAuto EncoderBiomedical Data
🎯 What it does: Developed a DNA storage error-correcting code (THEA-Code) based on autoencoders, achieving adaptive coding for different insertion-deletion-substitution (IDS) error channels through perturbation discretization and differentiable IDS channels.
DiTEA: Mixture-of-Experts for Vision-Language-Action Model in Robotic Manipulation
Chengxuan Li (Chinese Academy of Science), Xingwan Wang (University of Science and Technology of China)
Robotic IntelligenceTransformerMixture of ExpertsVision-Language-Action ModelDiffusion modelImageTextMultimodality
🎯 What it does: Propose DiTEA, a vision-language-action model that integrates Diffusion Transformer with Mixture-of-Experts (MoE), addressing skill forgetting and instruction following in robotic multi-task learning through a Task-Instruction Gate-enabled Action MoE.
DivControl: Knowledge Diversion for Controllable Image Generation
Yucheng Xie (Southeast University), Xin Geng (Southeast University)
GenerationTransformerMixture of ExpertsDiffusion modelImageText
🎯 What it does: Propose DivControl, which leverages knowledge diversion to decouple ControlNet into shared learngenes and condition-specific tailor, and achieves unified controllable image generation through dynamic gate routing based on conditional text embeddings, supporting zero-shot and few-shot rapid adaptation.
Diverse Human Driving Vehicle Simulation in Background Traffic for Autonomous Driving Tests
Wendi Li (Nanjing University), Sheng Zhong (Nanjing University)
GenerationData SynthesisAutonomous DrivingExplainability and InterpretabilityLarge Language ModelText
🎯 What it does: This paper proposes HDSim, an intelligent human-driven vehicle simulator that injects perception bias through a cognitive hierarchy model and LLM, generating diverse and realistic background traffic in AD testing platforms such as CARLA.
Diversifying Counterattacks: Orthogonal Exploration for Robust CLlP Inference
Chengze Jiang (Southeast University), Jie Gui (Southeast University)
Adversarial AttackTransformerVision Language ModelContrastive LearningImage
🎯 What it does: For test-time adversarial defense of the CLIP model, the Orthogonal Exploration Directional Adversarial method (DOC) is proposed, which enhances the diversity of adversarial examples by introducing orthogonal gradients and momentum, thereby improving the defense against adversarial attacks.
Diversity of Structured Domains via k-Kemeny Scores
Piotr Faliszewski (AGH University), Tomasz Wąs (Univeristy of Oxford)
🎯 What it does: This paper investigates the computational complexity and diversity evaluation of the k-Kemeny problem on structural domains such as single-peaked, single-crossing, group-separable, and Euclidean.
Diversity Recommendation via Causal Deconfounding of Co-purchase Relations and Counterfactual Exposure
Jingmao Zhang (Harbin Institute of Technology Shenzhen), Xiaofeng Zhang (Harbin Institute of Technology Shenzhen)
Recommendation SystemGraph Neural NetworkGraph
🎯 What it does: Proposes the Cadence framework, which enhances recommendation diversity through causal debiasing co-purchase relationships and adversarial exposure, based on unsupervised learning with LightGCN;
Divide-and-Conquer Decoupled Network for Cross-Domain Few-Shot Segmentation
Runmin Cong (Shandong University), Wei Zhang (Shandong University)
SegmentationDomain AdaptationAdversarial AttackContrastive LearningImageBiomedical Data
🎯 What it does: This paper proposes a method for cross-domain few-shot semantic segmentation by splitting encoder features into class-specific private features and domain-related shared features, and achieving cross-domain adaptation through matrix-guided dynamic fusion.
Divide, Conquer and Unite: Hierarchical Style-Recalibrated Prototype Alignment for Federated Medical Segmentation
Xingyue Zhao (Wuhan University), Mang Ye (Xinjiang University)
SegmentationFederated LearningContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a hierarchical style calibration prototype alignment framework called FedBCS, aimed at addressing feature bias caused by domain unevenness in federated medical segmentation.
Dividing Indivisible Items for the Benefit of All: It Is Hard to Be Fair Without Social Awareness
Argyrios Deligkas (Royal Holloway, University of London), Šimon Schierreich (Czech Technical University in Prague)
Optimization
🎯 What it does: This paper investigates the fair division problem of allocatable indivisible items by incorporating social influence factors, exploring the feasibility and computational complexity of maximizing social influence while satisfying various fairness constraints.
DLDA: Unified Dual-Level Domain Adaptation for Low-Light Object Detection
Jiayi Hu (Tongji University), Gang Li (Tongji University)
Object DetectionDomain AdaptationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Propose the DLDA framework, combining image-level photometric-aware contrastive translation and multi-scale conditional adversarial alignment to achieve unsupervised domain adaptation for low-light object detection.
DLVINet: Advancing Dual-Lens Video Inpainting Beyond Parallax Constraints
Zhiliang Wu (Zhejiang University), Yi Yang (Zhejiang University)
RestorationTransformerVideo
🎯 What it does: A comprehensive recovery framework is proposed for dual-camera video inpainting tasks, which simultaneously addresses both standard and non-standard dual-camera systems.
DMCAR: Disentangled Mixture-of-Experts with Context-Aware Routing for Multi-View Clustering
Baili Xiao (National University of Defense Technology), En Zhu (National University of Defense Technology)
Representation LearningMixture of ExpertsContrastive LearningBenchmark
🎯 What it does: Propose a deep multi-view clustering framework called DMCAR-MVC based on decoupled expert mixture networks and context-aware routing, which simultaneously learns view-shared and view-specific representations.
DMGIN: How Multimodal LLMs Enhance Large Recommendation Models for Lifelong User Post-click Behaviors
Zhuoxing Wei (Meituan), Jingsong Yu (PeKing University)
Recommendation SystemTransformerLarge Language ModelVision Language ModelMultimodalitySequential
🎯 What it does: Designed and implemented DMGIN, which utilizes a multimodal large language model to perform cross-modal clustering of shops, followed by compressing user life-cycle behavior sequences into interest groups for CTR prediction.
DMGINE: Day-Memory Guided Nighttime Image Enhancement for Dynamic Traffic Scenes
Ruizhou Liu (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
Image TranslationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageVideo
🎯 What it does: Generate a static background memory using daytime videos to guide nighttime image enhancement, while ensuring foreground information retention through foreground localization contrastive learning.
DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis
Yinghao Aaron Li (NewsBreak), Nima Mesgarani (Columbia University)
GenerationOptimizationSafty and PrivacyReinforcement LearningDiffusion modelAudio
🎯 What it does: Developed DMOSpeech 2, which uses reinforcement learning to optimize the duration predictor and introduces teacher-guided sampling to achieve end-to-end metric optimization;
DNOI-4DRO: Deep 4D Radar Odometry with Differentiable Neural-Optimization Iterations
Shouyi Lu (Tongji University), Xiao Tang (Tongji University)
Pose EstimationAutonomous DrivingOptimizationConvolutional Neural NetworkRecurrent Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposed a 4D millimeter-wave radar odometry model DNOI-4DRO that integrates traditional geometric optimization with end-to-end deep learning.
Do Audio-Visual Segmentation Models Truly Segment Sounding Objects?
Jia Li (University of Texas at Dallas), Yapeng Tian (University of Texas at Dallas)
SegmentationTransformerContrastive LearningVideoMultimodalityBenchmarkAudio
🎯 What it does: Investigate whether existing audio-visual segmentation models truly integrate audio information, propose the AVSBench-Robust benchmark, and design a classifier-guided similarity learning method to enhance model robustness against negative audio scenarios.
Do It for HER: First-Order Temporal Logic Reward Specification in Reinforcement Learning
Pierriccardo Olivieri (Politecnico di Milano), Matteo Papini (Universita degli Studi di Milano)
Reinforcement Learning
🎯 What it does: This paper proposes a non-Markovian reward specification framework based on LTLfMT (Linear Temporal Logic modulo Theory), which can directly evaluate first-order logic predicates in continuous control environments using SMT solvers, thereby avoiding the need for manually writing label functions.
Do Language Models Associate Sound with Meaning? A Multimodal Study of Sound Symbolism
Jinhong Jeong (Yonsei University), Youngjae Yu (Seoul National University)
Explainability and InterpretabilityRepresentation LearningData-Centric LearningTransformerLarge Language ModelMultimodalityAudio
🎯 What it does: Investigate the understanding of phonosemantic symbols (onomatopoeia and constructed pseudo-words) by multimodal large language models (MLLMs), construct the LEX-ICON dataset, and conduct semantic dimension prediction and internal attention analysis.
Do Large Language Models Reason About Uncertainty Like Humans? A Benchmark on Hurricane Forecast Visualization Comprehension
Le Liu (Northwestern Polytechnical University), Peng Wang (Northwestern Polytechnical University)
TransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Built and utilized the UnReason benchmark to systematically evaluate differences between humans and LLMs in visualizing uncertainties in hurricane prediction.
Do Large Language Models Think like the Brain? Sentence-Level Evidences from Layer-Wise Embeddings and fMRI
Yu Lei (Beijing University of Posts and Telecommunications), Bolei Ma (LMU Munich)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Compare the correspondence between 14 types of LLM hierarchical representations and human brain fMRI at the sentence level, and investigate the relationship between instruction fine-tuning, semantic understanding, brain alignment, and hemispheric asymmetry.
Do LLMs Feel? Teaching Emotion Recognition with Prompts, Retrieval, and Curriculum Learning
Xinran Li (Dalian University of Technology), Xiujuan Xu (Dalian University of Technology)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Propose the PRC-Emo framework, combining Prompt engineering, example retrieval, and curriculum learning, leveraging large language models to achieve explicit and implicit recognition of conversational emotions.
Do LLMs Really Struggle at NL-FOL Translation? Revealing Their Strengths via a Novel Benchmarking Strategy
Andrea Brunello (University of Udine), Nicola Saccomanno (University of Udine)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: A systematic evaluation of methods for assessing natural language to first-order logic (NL-FOL) translation was conducted, and a three-step benchmark based on ontology and logical translation splitting was proposed.
Do Not Merge My Model! Safeguarding Open-Source LLMs Against Unauthorized Model Merging
Qinfeng Li (Zhejiang University), Xuhong Zhang (Zhejiang University)
Safty and PrivacyTransformerLarge Language ModelText
🎯 What it does: Propose MergeBarrier, an active defense technique against model merging attacks, which utilizes shared orthogonal projections to disrupt the linear pattern connections in attention layers and expands the reparameterized FFN layer through activation functions, achieving the prevention of model merging without relying on external components.
Do Retrieval Augmented Language Models Know When They Don’t Know?
Youchao Zhou (Beijing Institute of Technology), Yang Deng (Singapore Management University)
Explainability and InterpretabilityTransformerSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Investigate whether retrieval-augmented language models (RALM) can correctly refuse to answer when no answer is available, and systematically evaluate the relationship between their calibration, refusal behavior, and external retrieval context; propose a post-rejection method leveraging uncertainty thresholds and contextual information to balance refusal rates and answer quality.
Do We Need Perfect Data? Leveraging Noise for Domain Generalized Segmentation
Taeyeong Kim (Kyung Hee University), MyeongAh Cho (Kyung Hee University)
SegmentationDomain AdaptationAutonomous DrivingDiffusion modelContrastive LearningImage
🎯 What it does: Propose a framework called FLEX-Seg for domain-generalized semantic segmentation using diffusion-generated synthetic data, focusing on solving the misalignment between synthetic images and semantic masks at edges.
Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute
Jianhao Chen (Nanjing University), Shuyue Hu (Shanghai Artificial Intelligence Laboratory)
Computational EfficiencyLarge Language ModelText
🎯 What it does: This paper proposes a test-time computational enhancement strategy called ModelSwitch, which leverages multi-model resampling and dynamically switches models based on consistency to improve answer accuracy while reducing sampling costs.
DoBlock: Blocking Malicious Association Propagation for Backdoor-Robust Federated Learning Under Domain Skew
Zhou Tan (Fuzhou University), Shouling Ji (National Interdisciplinary Research Center of Engineering Physics)
Domain AdaptationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: Propose the DoBlock scheme, which in federated learning aggregates domain information infusers and isolates local models to prevent the propagation of malicious trigger-label associations, thereby defending against backdoor attacks in domain skew environments.
DocR1: Evidence Page-Guided GRPO for Multi-Page Document Understanding
Junyu Xiong (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Reinforcement Learning from Human FeedbackTransformerReinforcement LearningTextMultimodality
🎯 What it does: Proposed a multi-page document understanding model DocR1 based on reinforcement learning, trained using the Evidence Page-Guided GRPO (EviGRPO) framework.
Does Question Really Matter? The Attribution of Answer Bias in LLM Evaluation
Boxi Cao (Chinese Information Processing Laboratory Institute of Software Chinese Academy of Sciences), Le Sun (Chinese Information Processing Laboratory Institute of Software Chinese Academy of Sciences)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelTextBenchmark
🎯 What it does: Systematically evaluate answer bias in multiple-choice question (MCQA) assessments, analyze its causes, and propose a pollution detection and dataset trimming method based on unique option evaluation.
DogFit: Domain-guided Fine-tuning for Efficient Transfer Learning of Diffusion Models
Yara Bahram (École de Technologie Supérieure), Eric Granger (École de Technologie Supérieure)
GenerationDomain AdaptationComputational EfficiencySupervised Fine-TuningDiffusion modelImage
🎯 What it does: Proposes DogFit, a training-time guidance mechanism used in diffusion model transfer learning;
DoKnowAD: Calibrating Normal Representations with Refined Domain Knowledge to Enhance Time Series Anomaly Detection
Shiwang Xing (Beihang University), Tao Ren (Chinese Academy of Sciences)
Anomaly DetectionAuto EncoderContrastive LearningTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: Propose the DoKnowAD framework, which leverages domain knowledge from auxiliary datasets through super-volume estimation to filter valuable data and calibrate the representation space of the target time-series data via contrastive learning, thereby enhancing unsupervised time-series anomaly detection performance.
Domain Adaptation Guided Infrared and Visible Image Fusion
Tianwei Guan (Chinese University of Hong Kong), Xingyuan Li (Zhejiang University)
Domain AdaptationImage
🎯 What it does: Propose the DAFusion fusion framework, combining domain adaptation with infrared-visible image fusion, and achieving adaptive fusion through dual-rank adapters and two-layer optimization.
Domain-Auxiliary Infrared Moving Small Target Detection by Learning to Overlook Domain Discrepancy
Shengjia Chen (University of Electronic Science and Technology of China), Mao Ye (University of Electronic Science and Technology of China)
Object DetectionDomain AdaptationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageVideo
🎯 What it does: This paper proposes the Loddis framework, which improves infrared small target detection by learning to ignore domain differences. It uses auxiliary large-sample domain data in the limited-sample task domain to avoid the negative shift degradation problem.
Domain-Aware Multi-View Contrastive Representation Learning for Protein Subcellular Localization Prediction
Qiang Zhang (Wuhan University), Juan Liu (Wuhan University)
ClassificationRepresentation LearningGraph Neural NetworkTransformerContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Proposed the DMVCL model for predicting protein subcellular localization based on domain-aware multi-view contrastive learning
Domain-Aware Suppression and Aggregation for Federated DG ReID
Zhixi Yu (Wuhan University of Science and Technology), Xin Xu (Wuhan University)
RetrievalDomain AdaptationFederated LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose the FedSupWA framework, combining domain-aware parameter suppression (DPS) and domain-invariant weighted aggregation (DWA), to enhance domain generalization performance for person ReID in federated learning environments.
DomainCQA: Crafting Knowledge-Intensive QA from Domain-Specific Charts
Yujing Lu (Zhejiang Lab), Qing Zhang (Zhejiang Lab)
Convolutional Neural NetworkLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodalityBenchmarkPhysics RelatedChain-of-Thought
🎯 What it does: Design the DomainCQA framework and construct the AstroChart benchmark in the astronomy field, providing 482 charts and 1,690 QA pairs covering two-tier tasks of FQA and AQA.
Dominance Pruning and Heuristics in Optimal Adversarial Non-Deterministic Planning
Rasmus G. Tollund (Aalborg University), Álvaro Torralba (Aalborg University)
OptimizationBenchmark
🎯 What it does: This paper proposes a domain-agnostic algorithm for adversarial FOND problems, utilizing single-outcome determinization to construct acceptable heuristics and introducing multiple dominance-based pruning methods to significantly reduce the search space and improve solving efficiency while guaranteeing worst-case cost minimization.
Don’t Start Over: A Cost-Effective Framework for Migrating Personalized Prompts Between LLMs
Ziyi Zhao (University of Science and Technology of China), Fuli Feng (Huawei Technologies Co Ltd)
Recommendation SystemLarge Language ModelPrompt EngineeringText
🎯 What it does: To address the issue of personalized soft prompts becoming ineffective after large language models (LLM) upgrades, the PUMA framework is proposed to achieve soft prompt transfer.
Don’t Stop the Multi-Party! On Generating Synthetic Written Multi-Party Conversations with Constraints
Nicolò Penzo (Fondazione Bruno Kessler), Sara Tonelli (Fondazione Bruno Kessler)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes using instruction-tuned large language models to generate synthetic written multi-party dialogues while satisfying constraints such as dialogue structure and speaker stance.
Dormant Backdoor: Weaponizing Model Finetuning for Feasible Backdoor Attacks Against Pretrained Models
Ruitao Li (Institute of Information Science Beijing Jiaotong University), Renshuai Tao (Institute of Information Science Beijing Jiaotong University)
Knowledge DistillationAdversarial AttackSupervised Fine-TuningImage
🎯 What it does: Propose the Dormant Backdoor attack, which activates a hidden backdoor through the fine-tuning process.