AAAI 2026 Papers — Page 9
AAAI Conference on Artificial Intelligence · 4149 papers
Decomposing Prompts, Composing Actions: A Multi-Granularity Prompting Approach for Incremental Action Learning
Xinyi Cheng (Xidian University), Yanhua Yang (Xidian University)
RecognitionGraph Neural NetworkPrompt EngineeringContrastive LearningVideoGraph
🎯 What it does: DPCA achieves incremental skeletal action recognition without replay through multi-grained prompts, task-agnostic prompts, and differential attention correction.
Decomposing the Neurons: Activation Sparsity via Mixture of Experts for Continual Test Time Adaptation
Rongyu Zhang (Nanjing University), Yuan Du (Nanjing University)
ClassificationSegmentationDomain AdaptationTransformerMixture of ExpertsImage
🎯 What it does: This paper proposes the MoASE module, which decomposes neural network activations into high-activation (domain-agnostic) and low-activation (domain-specific) categories, combining a teacher-student framework to achieve adaptation during continuous testing.
DECON: Reconstruction of Clothed-Geometric Multiple Humans from a Single Image via Geometry-Guided Decoupling
Yiming Jiang, Aimin Hao (Beihang University)
GenerationPose EstimationDiffusion modelNeural Radiance FieldImageMeshBenchmark
🎯 What it does: Propose the DECON framework to achieve disentanglement and reconstruction of multi-human full-body clothing geometry from a single RGB image, and restore realistic spatial relationships through perspective-aware position optimization.
Deconstructing Pre-training: Knowledge Attribution Analysis in MoE and Dense Models
Bo Wang, Xuming Hu (Hong Kong University of Science and Technology)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Investigate the knowledge acquisition dynamics between MoE and dense Transformers during pre-training, propose the Gated Log-Probability Increase (Gated-LPI) metric, and track the importance of neurons in model checkpoints
DeCoRL: Decoupling Reasoning Chains via Parallel Sub-Step Generation and Cascaded Reinforcement for Interpretable and Scalable RLHF
Ziyuan Gao (University College London), Minlong Peng (University College London)
Reinforcement Learning from Human FeedbackTransformerMixture of ExpertsTextChain-of-Thought
🎯 What it does: Propose the DeCoRL framework, which decomposes chained reasoning tasks into multiple parallel substeps generated by specialized modules in parallel, thereby eliminating the O(n) complexity bottleneck of traditional sequential decoding;
Decoupled Spatiotemporal Forecasting from Extreme Sparse Observations via Quantized Latent Space
Zhongnan Weng (Xiamen University), Xiangrong Liu (Xiamen University)
Representation LearningData-Centric LearningTransformerAuto EncoderPhysics Related
🎯 What it does: By decoupling spatial reconstruction from temporal extrapolation of extremely sparse observations, high-precision spatial reconstruction and long-term temporal prediction are achieved in the quantized latent space.
Decoupling Continual Semantic Segmentation
Yifu Guo (Sun Yat-sen University), Ruixuan Wang (Sun Yat-sen University)
SegmentationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelImageText
🎯 What it does: Proposed the DecoupleCSS two-stage framework, decoupling class-aware detection from class-agnostic segmentation to achieve continual semantic segmentation; using task-specific LoRA adapters for class detection driven by language, and leveraging SAM to generate positional prompts for segmentation.
Decoupling Knowledge and Reasoning in LLMs: An Exploration Using Cognitive Dual-System Theory
Mutian Yang (Tsinghua University), Ji Wu (Tsinghua University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This work proposes to split the reasoning process of large language models into two stages: knowledge retrieval (fast thinking) and reasoning adjustment (slow thinking), and quantifies the contribution of knowledge and reasoning to performance by comparing two prompting methods. Subsequently, systematic experiments were conducted on 15 models and three categories of datasets.
Decoupling Scene Perception and Ego Status: A Multi-Context Fusion Approach for Enhanced Generalization in End-to-End Autonomous Driving
Jiacheng Tang (Fudan University), Jian Pu (Fudan University)
Autonomous DrivingTransformerImagePoint Cloud
🎯 What it does: By designing the AdaptiveAD framework, a dual-branch structure is constructed to decouple scene perception and ego state reasoning, and a multi-context fusion module adaptively integrates the two decisions to enhance the generalization and robustness of end-to-end autonomous driving.
Decoupling Shared and Personalized Knowledge: A Dual-Branch Federated Learning Framework for Multi-Domain with Non-IID Data
Yiran Pang (Florida Atlantic University), Xiangnan Zhong (Florida Atlantic University)
Domain AdaptationFederated LearningImageBiomedical Data
🎯 What it does: Proposes a dual-branch personalized federated learning framework called pFedDB, which uses a two-phase training process to first learn expert models locally and then collaborates across multi-domain non-IID data through a shared branch to address catastrophic forgetting and negative transfer problems.
Decoupling Template Bias in CLIP: Harnessing Empty Prompts for Enhanced Few-Shot Learning
Zhenyu Zhang, Yuhua Li (Peking University)
ClassificationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes to eliminate the bias caused by template-sample similarity (TSS) in the CLIP model during few-shot learning by utilizing empty prompts and a template bias calibration loss, thereby improving classification accuracy and robustness.
Decoupling Understanding from Reasoning via Problem Space Mapping for Small-Scale Model Reasoning
Li Wang (Beihang University), Wenjun Wu (Beihang University)
Computational EfficiencyKnowledge DistillationRepresentation LearningLarge Language ModelReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Propose a framework that decouples understanding and reasoning by mapping natural language questions to a low-dimensional normalized question space, and implement a three-step alternating training algorithm called DURIT.
Decoupling What to Count and Where to See for Referring Expression Counting
Yuda Zou (Wuhan University), Yongchao Xu (Wuhan University)
Object DetectionTransformerVision Language ModelImageMultimodality
🎯 What it does: Designed the W2-Net framework, addressing the mismatch between annotation points and attribute-related visual regions in Referring Expression Counting through dual query mechanisms (what-to-count and where-to-see) and Subclass Separable Matching (SSM), achieving more fine-grained subclass counting and localization.
Deep (Predictive) Discounted Counterfactual Regret Minimization
Hang Xu (Institute of Automation, Chinese Academy of Sciences), Jian Cheng (Tencent AI Lab)
Reinforcement LearningBenchmark
🎯 What it does: Proposed two model-agnostic neural CFR variants, VR-DeepDCFR+ and VR-DeepPDCFR+, which approximate the updates of DCFR+ and PDCFR+ through bootstrapping, discounting, and truncating cumulative advantages, while employing a value baseline to reduce variance;
Deep Clustering Based on Sparse Kolmogorov-Arnold Network and Spectral Constraint
Zixuan Bi (Northwestern Polytechnical University), Ganchao Liu (Northwestern Polytechnical University)
OptimizationRepresentation LearningImageText
🎯 What it does: Proposes a deep clustering framework based on sparse Kolmogorov-Arnold Network (KAN) and spectral constraints, utilizing an adaptive adjacency matrix for unsupervised clustering.
Deep Incomplete Multi-View Clustering via Hierarchical Imputation and Alignment
Yiming Du (Old Dominion University), Lusi Li (Old Dominion University)
Representation LearningAuto EncoderContrastive LearningMultimodality
🎯 What it does: Propose a deep incomplete multi-view clustering framework DIMVC-HIA, achieving high-quality clustering through hierarchical missing value imputation and alignment.
Deep Inverse Shading: Consistent Albedo and Surface Detail Recovery via Generative Refinement
Jiacheng Wu (Hong Kong Baptist University), Jie Chen (Hong Kong Baptist University)
RestorationGenerationConvolutional Neural NetworkDiffusion modelImageVideoMesh
🎯 What it does: This paper proposes the DIS (Deep Inverse Shading) framework, which achieves geometric and material (chromatic) consistency recovery and high-quality relighting synthesis for human avatars through a SMPL-based sparse grid with regularization, deep generative prior, and inverse shading modules.
Deep Reinforcement Learning for Scalable Offline Three-Dimensional Packing
Hao Yin (Southwest Jiaotong University), Fan Chen (Southwest Jiaotong University)
OptimizationTransformerReinforcement LearningTabular
🎯 What it does: Propose a scalable deep reinforcement learning method to solve the offline 3D packing problem, capable of efficiently planning layouts for a large number of items (20–1000 pieces).
Deep Research Arena: The First Exam of LLMs’ Research Abilities via Seminar-Grounded Tasks
Haiyuan Wan (Shanghai Artificial Intelligence Laboratory), Dongzhan Zhou (Shanghai Artificial Intelligence Laboratory)
RetrievalLarge Language ModelAgentic AITextBenchmarkAudio
🎯 What it does: Proposes DeepResearch Arena, a deep research agent evaluation benchmark generated based on academic workshops, covering 12 domains and over 10,000 tasks.
DeepBooTS: Dual-Stream Residual Boosting for Drift-Resilient Time-Series Forecasting
Daojun Liang (Qilu University of Technology (Shandong Academy of Sciences)), Shuo Li (Case Western Reserve University)
Time SeriesBenchmark
🎯 What it does: Proposed the DeepBooTS dual-stream residual decrement boosting network to address concept drift in time series forecasting.
Deeper Treatment of the Bi-objective Search Framework
Shawn Skyler (Ben Gurion University Negev), Sven Koenig (Uppsala University)
OptimizationGraph
🎯 What it does: This paper investigates the impact of node classification and sorting functions on search efficiency in Bi-Objective Search (BOS), further refines the handling of non-expanded nodes (NENs), and proposes constant-time dominance check methods for MIN and MAX sorting functions.
Deeply Seeking Boundary for Lunar Regolith Segmentation
Yifeng Wang (Tsinghua University), Zongquan Deng (Harbin Institute of Technology)
SegmentationTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: This study addresses high-precision segmentation of lunar regolith particles by proposing the HiFi-LoRA and WEM frameworks to resolve spectral bias issues in deep learning models for capturing high-frequency details.
DeepOR: A Deep Reasoning Foundation Model for Optimization Modeling
Ziyang Xiao (Zhejiang University), Dongxiang Zhang (Zhejiang University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Proposes DeepOR, a deep reasoning foundation model designed for optimization modeling, which automatically generates expert flowcharts to construct long-chain reasoning data, and performs supervised fine-tuning and reinforcement learning on this data, explicitly visualizing intermediate reasoning steps to enhance model reasoning quality.
DeepPhy: Benchmarking Agentic VLMs on Physical Reasoning
Xinrun Xu (Taobao & Tmall Group of Alibaba), Bo Zheng (Taobao & Tmall Group of Alibaba)
Agentic AIVision Language ModelVision-Language-Action ModelWorld ModelMultimodalityBenchmarkPhysics Related
🎯 What it does: Propose the DeepPHY benchmark to systematically evaluate the capabilities of vision-language models in interactive physical reasoning tasks.
DeepProofLog: Efficient Proving in Deep Stochastic Logic Programs
Ying Jiao (KU Leuven), Giuseppe Marra (University of Siena)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningReinforcement LearningImageTextGraph
🎯 What it does: Propose DeepProofLog (DPrL), a neural network-based deep stochastic logic program that can guide proof steps in real-time during reasoning and model reasoning as a Markov Decision Process (MDP).
DeepRAHT: Learning Predictive RAHT for Point Cloud Attribute Compression
Chunyang Fu (City University of Hong Kong), Zhu Li (University of Missouri-Kansas City)
CompressionConvolutional Neural NetworkPoint Cloud
🎯 What it does: Proposed an end-to-end differentiable DeepRAHT framework, achieving complete training and compression workflow of RAHT in deep learning.
DeepRWCap: Neural-Guided Random-Walk Capacitance Solver for IC Design
Hector Rodriguez Rodriguez (Tsinghua University), Wenjian Yu (Tsinghua University)
Computational EfficiencyConvolutional Neural NetworkImagePhysics Related
🎯 What it does: Developed a neural network-guided random walk solver called DeepRWCap for accelerating multi-medium IC capacitance extraction.
DeepSenseMoE: Harnessing Power of Time Series Foundation Models for Few-Shot Human Activity Recognition
Zenan Fu (Nanjing Normal University), Hao Wu (Yunnan University)
RecognitionTransformerSupervised Fine-TuningMixture of ExpertsContrastive LearningMultimodalityTime Series
🎯 What it does: This paper proposes the DeepSenseMoE module, which performs parameter-efficient fine-tuning of pre-trained time series foundation models through multi-scale convolution Mixture-of-Experts to address the scarcity and heterogeneity of wearable sensor data.
DeepTracer: Tracing Stolen Model via Deep Coupled Watermarks
Yunfei Yang (Chinese Academy of Sciences), He Li (Chinese Academy of Sciences)
Safty and PrivacyAdversarial AttackImage
🎯 What it does: DeepTracer proposes a deeply coupled watermarking framework that enhances watermark robustness by strengthening the coupling between the watermark task and the main task to defend against model stealing attacks.
DeepWriter: A Multi-Agent Collaboration Framework for Information-rich Ultra-long Book Writing
Ming Wang (North China University of Technology), Guotong Geng (North China University of Technology)
GenerationRetrievalTransformerLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose DeepWriter — a multi-agent collaborative framework for generating ultra-long information-rich books;
DEFANet: Dual-Path Edge-Target Collaboration with Frequency-Aware Enhancement for Infrared Small Target Detection
Shuaiyuan Du (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose a dual-path edge-target collaborative frequency-aware network called DEFANet for infrared small target detection.
DeFB: Decomposed Feature Learning for Real-Time Multi-Person Eyeblink Detection in Untrimmed In-the-Wild Videos
Jinfang Gan (Huazhong University of Science and Technology), Zhiguo Cao (ByteDance)
Object DetectionObject TrackingTransformerVideo
🎯 What it does: Propose the DeFB framework to achieve an end-to-end real-time system for multi-person face detection, tracking, and eye movement detection;
Deferred Poisoning: Making the Model More Vulnerable via Hessian Singularization
Yuhao He, Jiantao Zhou (Foshan University)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Proposes a stealthy data poisoning method called Deferred Poisoning Attack (DPA), which utilizes Hessian matrix singularization to enhance model local curvature, making the model perform normally during training and validation stages but become highly vulnerable to adversarial attacks and natural noise after deployment.
Deformable Polygonal Flow Matching with Informed Priors and Hierarchical Graph Constraints
Arnaud Gueze (Ecole Polytechnique), Marie-Paule Cani (Homiwoo)
GenerationData SynthesisTransformerFlow-based ModelGraph
🎯 What it does: Propose a Deformable Polygon Flow Matching (DPFM) framework based on flow matching for generating and reconstructing polygon layouts, such as puzzles and floor plans, supporting independent translation, rotation, and local deformation.
DeformTrace: A Deformable State Space Model with Relay Tokens for Temporal Forgery Localization
Xiaodong Zhu (Wuhan University), Zhongyuan Wang (Wuhan University)
Anomaly DetectionTransformerVideoMultimodalityBenchmarkAudio
🎯 What it does: This paper proposes the DeformTrace framework, which achieves high-precision localization of video and audio temporal forgery segments by leveraging deformable state space models and relay token mechanisms;
DeFT-LoRA: Decoupled and Fused Tuning with LoRA Experts for Universal Cross-Domain Retrieval
Ke Xu (Anhui University), Xun Yang (Hebei University of Technology)
RetrievalDomain AdaptationPrompt EngineeringMixture of ExpertsVision Language ModelContrastive LearningImage
🎯 What it does: Propose the DeFT-LoRA framework, combining LoRA and MoE to achieve parameter-efficient UCDR;
DeFuzzRAG: Handling Fuzzy Time Expressions for Temporal Robustness in Retrieval-Augmented Generation
Ling-Chun Chen (National Taiwan University), Ming-Syan Chen (National Taiwan University)
RetrievalTransformerTextRetrieval-Augmented Generation
🎯 What it does: Proposes the DeFuzzRAG framework, addressing the issue of temporal inconsistency caused by fuzzy time expressions in Retrieval-Augmented Generation (RAG), through a three-stage process: time intent extraction, document time metadata generation, and time consistency filtering.
DEGRE: Dynamic Gating Ensembles for Trust-Aware Rejection in Medical Image Diagnostics
Hai Nguyen Hong (FPT University), Cuong Do (FPT University)
ClassificationAnomaly DetectionMixture of ExpertsImageBiomedical DataMagnetic Resonance ImagingComputed TomographyBenchmark
🎯 What it does: Proposes the DEGRE framework, which achieves adaptive rejection prediction by learning consensus confidence and inconsistency through a dynamic gating network based on deep ensemble.
DegVoC: Revisiting Neural Vocoder from a Degradation Perspective
Andong Li (Chinese Academy of Sciences), Chengshi Zheng (Tencent)
RestorationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkAudio
🎯 What it does: Redefine the speech waveform generation task from a degradation perspective and propose the DegVoC neural vocoder.
DehazeGS: Seeing Through Fog with 3D Gaussian Splatting
Jinze Yu, Xiaopeng Zhang (Chongqing University)
RestorationDepth EstimationAutonomous DrivingConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: Propose a physics-driven 3D Gaussian Splatting (3DGS) framework called DehazeGS, which can restore fog-free scenes and synthesize new views using only multi-view foggy images.
DEIG: Detail-Enhanced Instance Generation with Fine-Grained Semantic Control
Shiyan Du, Dongyu Zhang (Sun Yat-sen University)
GenerationData SynthesisConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Propose the DEIG framework for achieving fine-grained semantic control in multi-instance image generation, capable of precisely matching instance attributes and positions in complex text descriptions;
Delayed Feedback Modeling with Influence Functions
Chenlu Ding (University of Science and Technology of China), Andrew Rabinovich (Upwork)
Recommendation SystemTabularSequential
🎯 What it does: This paper proposes an influence function-based delayed feedback modeling framework, IF-DFM, which can directly estimate the impact of label flipping and new incoming data on model parameters, enabling efficient updates to the CVR prediction model without requiring full retraining.
DeLightMono: Enhancing Self-Supervised Monocular Depth Estimation in Endoscopy by Decoupling Uneven Illumination
Mingyang Ou (Southern University of Science and Technology), Jiang Liu (Southern University of Science and Technology)
Depth EstimationConvolutional Neural NetworkBiomedical Data
🎯 What it does: Proposed a self-supervised monocular depth estimation framework called DeLightMono, which separates endoscopic images using an illumination-reflection-depth (IRD) model to mitigate the impact of uneven illumination on depth estimation.
DeLo: Dual Decomposed Low-Rank Experts Collaboration for Continual Missing Modality Learning
Xiwei Liu (Mohamed bin Zayed University of Artificial Intelligence), Imran Razzak (Mohamed bin Zayed University of Artificial Intelligence)
Computational EfficiencyMixture of ExpertsMultimodalityBenchmark
🎯 What it does: Proposes the DeLo framework, which utilizes Dual Decomposed Low-Rank Experts to address the Continual Missing Modality Learning (CMML) problem, balancing parameter efficiency and robustness to modality missing scenarios.
DeloopSGNN: Revisiting Spectral GNNs Through the Lens of Spatial Aggregation
Duanyu Li (National University of Defense Technology), Ruibo Wang (National University of Defense Technology)
Representation LearningAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: By constructing a spectral-to-spatial mapping theorem (S2SMT) and proposing a DeloopSGNN, the spectral GNN is transformed into spatial aggregation, eliminating cycles and over-smoothing caused by multiple aggregations, thereby enhancing expressiveness and adversarial robustness.
Delphi: A Neuro-Symbolic Framework for Individualized, Safe and Interpretable Treatment Recommendation
Muchan Tao (Nanjing University), Tieniu Tan (Nanjing University)
Recommendation SystemExplainability and InterpretabilityGraph Neural NetworkLarge Language ModelReinforcement LearningTabularBiomedical DataElectronic Health Records
🎯 What it does: Developed a neuro-symbolic causal reinforcement learning framework named Delphi for personalized, interpretable, and safe sepsis treatment recommendations.
Delta Matters: An Analytically Tractable Model for beta–delta Discounting Agents
Yasunori Akagi (NTT Human Informatics Laboratories), Takeshi Kurashima (NTT Human Informatics Laboratories)
OptimizationExplainability and Interpretability
🎯 What it does: This paper generalizes the β-δ discount model from the special case of δ=1 to the general case of 0<δ≤1, and provides a closed-form expression for agent behavior; based on this, it further analyzes task abandonment conditions, designs algorithms for optimal goal setting and reward scheduling, and explores the impact of δ on behavior and intervention effectiveness.
Delving into Dynamic Scene Cue-Consistency for Robust 3D Multi-Object Tracking
Haonan Zhang (Zhejiang University), Zheng Yang (Shandong Land-Sea-Nexus Digital Technology Co., Ltd.)
Object TrackingAutonomous DrivingTransformerVideoPoint Cloud
🎯 What it does: A 3D multi-object tracking method (DSC-Track) based on dynamic scene clue consistency is proposed for autonomous driving environments, achieving more robust data association by fusing geometric consistency and spatiotemporal context information.
Demystifying Foreground-Background Memorization in Diffusion Models
Jimmy Z. Di (University of Wisconsin-Madison), Franziska Boenisch (CISPA Helmholtz Center for Information Security)
GenerationExplainability and InterpretabilityVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed the FB-Mem metric based on foreground-background segmentation to detect local memory in fine-grained diffusion model-generated images, and achieved stronger memory elimination through clustering-based Neuron-level intervention (NeMo-C).
Demystifying GNN-to-MLP Knowledge Transfer: Theoretical Grounding and Dual-Stream Distillation Method
Zhiyuan Yu (Nanjing University), Sanglu Lu (Nanjing University)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Propose a theoretical framework based on the Neural Tangent Kernel (NTK) to explain how GNN-to-MLP knowledge distillation enables MLPs to implicitly acquire graph propagation capabilities, and design a dual-stream aligned MLP (DA-MLP) to improve the distillation process.
DeNAS-ViT: Data Efficient NAS-Optimized Vision Transformer for Ultrasound Image Segmentation
Renqi Chen (Fudan University), Kehan Wu (Southern University of Science and Technology)
SegmentationNeural Architecture SearchTransformerContrastive LearningImageBiomedical DataUltrasound
🎯 What it does: Propose a data-efficient, NAS-optimized Vision Transformer (DeNAS-ViT) framework for ultrasound image segmentation, addressing dual challenges of multi-scale feature extraction and robustness with limited labeled data.
DeNC++: Efficient Diffusion-Enhanced Neural Codec for End-to-end Semantic Streaming at the Edge
Qihua Zhou (Shenzhen University), Laizhong Cui (Shenzhen University)
CompressionComputational EfficiencyConvolutional Neural NetworkDiffusion modelAuto EncoderVideo
🎯 What it does: Developed DeNC++, a lightweight neural-enhanced video streaming (NeVS) solution capable of compressing and restoring videos on edge devices without relying on media servers;
DenoDet V2: Phase-Amplitude Cross Denoising for SAR Object Detection
Kang Ni (Nanjing University of Posts and Telecommunications), Yimian Dai (Nankai University)
Object DetectionImagePhysics Related
🎯 What it does: This paper proposes DenoDet V2, which leverages the complementary characteristics of SAR image amplitude and phase in the frequency domain, achieving more refined denoising and detection through phase-guided soft threshold denoising and phase-amplitude token exchange mechanisms.
DenoiseGS: Delta-Based 3D Gaussian Splatting with B-spline Trajectory Optimization for Dynamic Driving Scene Reconstruction
Junjie Linghu, Qiang Ling (University of Science and Technology of China)
Autonomous DrivingOptimizationGaussian SplattingMultimodality
🎯 What it does: This paper proposes the DenoiseGS method for high-quality reconstruction of dynamic driving scenes through explicit 3D Gaussian Splatting, focusing on solving problems of camera parameter noise and dynamic object annotation errors.
Denoising Mixup for Regression
Zhengzhang Hou (Jilin University), Ximing Li (University of Pretoria)
ImageTabularTime Series
🎯 What it does: Propose a Mixup denoising method called DE-MIXUP for regression tasks, which uses a noise estimation layer to correct label noise in mixed samples within the deep feature space;
Dense Cross-Scale Image Alignment with Fully Spatial Correlation and Just Noticeable Difference Guidance
Jinkun You (University of Macau), Yicong Zhou (University of Macau)
Convolutional Neural NetworkOptical FlowImage
🎯 What it does: This paper proposes an unsupervised cross-scale image alignment method, improving alignment accuracy through dense cross-scale regression, full-space correlation modules, and JND guidance.
DensiCrafter: Physically-Constrained Generation and Fabrication of Self-Supporting Hollow Structures
Shengqi Dang (Shanghai Research Institute for Intelligent Autonomous Systems), Nan Cao (Shanghai Research Institute for Intelligent Autonomous Systems)
GenerationOptimizationImageTextPhysics Related
🎯 What it does: Propose DensiCrafter, achieving lightweight and self-supporting hollow structures by optimizing the continuous density field of the voxel grid output from 3D generative models.
DentalGS: Pose-Free 3D Gaussian Splatting from Five Intraoral Images for Novel View Synthesis
Honghao Dai, Wenping Wang (Shandong University)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkGaussian SplattingImageMeshBiomedical Data
🎯 What it does: Proposed a new perspective synthesis framework for dentistry based on 3D Gaussian scattering, DentalGS, which constructs a 3D dental model using five intraoral photos without pose information and pre-correction scan data, achieving high-quality novel view rendering through iterative camera pose fitting, RepairNet repair, and lighting-aware Gaussian modeling.
DeOcc-1-to-3: 3D De-Occlusion from a Single Image via Self-Supervised Multi-View Diffusion
Yansong Qu (Xiamen University), Liujuan Cao (Xiamen University)
RestorationGenerationData SynthesisTransformerDiffusion modelImageMeshBenchmark
🎯 What it does: Proposed an end-to-end self-supervised multi-view generative framework, DeOcc-1-to-3, which can directly generate six structurally consistent new views from a single occluded image, achieving 3D de-occlusion.
Dep-MAP: A Multi-level Alignment Framework with Semantic Prototypes for Video-based Automatic Depression Assessment
Hao Wang (Qilu University of Technology), Qingxiang Wang (Qilu University of Technology)
ClassificationConvolutional Neural NetworkVision Language ModelContrastive LearningVideo
🎯 What it does: Proposed the Dep-MAP video-based automatic depression assessment framework, which utilizes a dual-branch structure to extract visual and emotional semantic features, and achieves key frame selection and depression severity judgment through semantic prototype clustering, cross-layer contrastive learning, and multi-scale fusion.
Departures: Distributional Transport for Single-Cell Perturbation Prediction with Neural Schrödinger Bridges
Changxi Chi (Zhejiang University), Stan Z. Li (Westlake University)
Drug DiscoveryBiomedical DataStochastic Differential Equation
🎯 What it does: This work proposes the Departures framework, which utilizes the Schrödinger Bridge (SB) approximation to achieve single-cell perturbation prediction. By combining discrete (gene activation states) and continuous (gene expression levels) bridge models, and adopting MiniBatch OT to directly obtain control-perturbed sample pairings, it avoids the challenges of bidirectional models and reverse processes in traditional SB. Ultimately, it achieves high-precision distribution-level prediction of single-cell gene expression and activation states under different gene or compound perturbation conditions.
DEPO: Dual-Efficiency Preference Optimization for LLM Agents
Sirui Chen (Tongji University), Chaochao Lu (Shanghai Artificial Intelligence Laboratory)
OptimizationReinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIPrompt EngineeringText
🎯 What it does: Designed and implemented the DEPO method, which performs preference optimization for dual efficiency (step-level and trajectory-level) in LLM agents.
Depth-Synergized Mamba Meets Memory Experts for All-Day Image Reflection Separation
Siyan Fang (Huazhong University of Science and Technology), Yuehuan Wang (Huazhong University of Science and Technology)
RestorationDepth EstimationTransformerMixture of ExpertsImage
🎯 What it does: This paper proposes the Depth-Memory Decoupling Network (DMDNet), achieving all-weather (especially nighttime) image reflection separation through depth-guided scanning and memory experts.
Dereflection Any Image with Diffusion Priors and Diversified Data
Jichen Hu (Shanghai Jiao Tong University), Wei Shen (Huawei Inc)
RestorationDiffusion modelImage
🎯 What it does: Proposes a complete solution for single-image reflection removal, including a high-quality and diverse reflection dataset DRR, a one-step diffusion-based removal framework, and a progressive training strategy.
Description Logics with Two Types of Definite Descriptions: Complexity, Expressiveness, and Automated Deduction
Michał Sochański (University of Łódź), Michał Zawidzki (University of Łódź)
Computational Efficiency
🎯 What it does: Studied the extension of description logic ALC with two types of deterministic descriptions (local {ιC} and global ιC.D), analyzing their complexity, expressiveness, and automatic reasoning capabilities.
Designed to Spread: A Generative Approach to Enhance Information Diffusion
Ziqing Qian (Tongji University), Nan Cao (Tongji University)
GenerationData SynthesisTransformerReinforcement LearningDiffusion modelImageTextMultimodality
🎯 What it does: Designed a generative framework to modify textual or image content into more shareable versions;
Designing Optimal Mechanisms to Locate Facilities with Insufficient Capacity for Bayesian Agents
Gennaro Auricchio (University of Padova), Jie Zhang (University of Bath)
Optimization
🎯 What it does: This paper studies the facility location problem under scarce resources, seeking the optimal mechanism within the Bayesian mechanism design framework to maximize social welfare.
Designing Truthful Mechanisms for Asymptotic Fair Division
Jugal Garg (University of Illinois Urbana Champaign), Yuang Eric Shen (University of Illinois Urbana Champaign)
Optimization
🎯 What it does: This paper proposes a new random mechanism PRD, which can achieve approximate fair (envy-free) allocation in polynomial time under asymptotic settings of equitable distribution;
DesireKV: Decoupling Sensitivity and Importance for Reasoning-Aware KV Cache Compression
Pengyu Cheng (Xi'an Jiao Tong University), Jiacheng Liu (Hong Kong University of Science and Technology)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Designed and implemented a KV cache compression framework called DesireKV, specifically for efficiently compressing long sequences generated during the chain-of-thought reasoning process of large language models.
Detect All-Type Deepfake Audio: Wavelet Prompt Tuning for Enhanced Auditory Perception
Yuankun Xie (State Key Laboratory of Media Convergence and Communication, Communication University of China), Long Ye (University of Chinese Academy of Sciences)
ClassificationAnomaly DetectionTransformerPrompt EngineeringContrastive LearningBenchmarkAudio
🎯 What it does: Construct a cross-type deepfake audio detection (ADD) benchmark and propose a SSL training paradigm combining Prompt Tuning with Wavelet Prompt Tuning, achieving unified detection of four types of deepfake audio: speech, sound, singing, and music.
Detecting Emotional Dynamic Trajectories: An Evaluation Framework for Emotional Support in Language Models
Zhouxing Tan (Peking University), Junfei Liu (Peking University)
Large Language ModelTextBenchmark
🎯 What it does: This paper constructs an evaluation framework based on emotional trajectories, using large-scale simulated dialogues to assess the long-term dynamic performance of LLMs in emotional support.
Detecting Fake News in Short Videos Through Multi-View Aggregation
Nuo Li (Nanjing University of Information Science and Technology), Chao Huang (Sun Yat-sen University)
ClassificationTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: Propose a multi-perspective aggregation model called MVA that jointly detects fake news in short videos using three perspectives: content, sentiment, and publisher profile.
Detecting the Future: All-at-Once Event Sequence Forecasting with Horizon Matching
Ivan Karpukhin (Sber AI Lab), Andrey Savchenko (Sber AI Lab)
Recurrent Neural NetworkTime SeriesSequentialBiomedical DataBenchmark
🎯 What it does: Propose DEF (Detection-based Event Forecasting), which utilizes matching loss to simultaneously and in parallel predict multiple future event sequences over a long period.
Detecting Unobserved Confounders: A Kernelized Regression Approach
Yikai Chen (National University of Defense Technology), Haotian Wang (National University of Defense Technology)
Anomaly DetectionBiomedical DataBenchmark
🎯 What it does: Propose the Kernel Regression Confounder Detection (KRCD) method, which detects unobserved confounding variables in nonlinear observational data under a single environment by comparing differences between standard and higher-order kernel regression coefficients in the Reproducing Kernel Hilbert Space (RKHS).
DeToNATION: Decoupled Torch Network-Aware Training on Interlinked Online Nodes
Mogens Henrik From (University of Southern Denmark), Peter Schneider-Kamp (University of Southern Denmark)
OptimizationComputational EfficiencyHyperparameter SearchImageText
🎯 What it does: Propose FlexDeMo, which employs FSDP sharding within nodes and only synchronizes rapidly changing gradients between nodes, achieving low-bandwidth distributed training for large models.
Deviation Dynamics in Cardinal Hedonic Games
Valentin Zech (University of Oxford), Martin Bullinger (University of Oxford)
🎯 What it does: Studies the possible and necessary convergence of bias dynamics based on a single agent's preferences, and proposes a general metatheorem that treats instances where no stable partition exists as a black-box tool to prove the difficulty of dynamics.
DexGraspVLA: A Vision-Language-Action Framework Towards General Dexterous Grasping
Yifan Zhong (Institute for Artificial Intelligence, Peking University), Yuanpei Chen (Institute for Artificial Intelligence, Peking University)
Robotic IntelligenceTransformerVision-Language-Action ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposed DexGraspVLA, a hierarchical vision-language-action framework for achieving general multi-finger grasping.
Dexterous Manipulation Transfer via Progressive Kinematic-Dynamic Alignment
Wenbin Bai (Dalian University of Technology), Yi Sun (Dalian University of Technology)
Data SynthesisPose EstimationRobotic IntelligenceReinforcement LearningVideo
🎯 What it does: Construct an arm-agnostic transfer system (PKDA) using human hand manipulation videos, which automatically generates high-quality trajectories capable of performing various manipulation tasks on multi-fingered robotic hands through advanced kinematic alignment, reinforcement learning residual strategies, and wrist trajectory planning.
DFDT: Dynamic Fast Decision Tree for IoT Data Stream Mining on Edge Devices
Afonso Lourenço (Polytechnic of Porto), Goreti Marreiros (University of Porto)
ClassificationComputational EfficiencyTime SeriesSequential
🎯 What it does: This paper proposes DFDT (Dynamic Fast Decision Tree), a data stream learning algorithm for IoT edge devices, which achieves efficient online decision tree construction through activity-aware pre-pruning and adaptive parameter control.
DFMN: A Dual-feet Matching Network with Hybrid Transformer-based Feature Extractor for Unsupervised Deformable Medical Image Registration
Liwen Li (Huazhong University of Science and Technology), Fumin Guo (Huazhong University of Science and Technology)
Convolutional Neural NetworkTransformerBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose a dual-foot matching network (DFMN) to achieve unsupervised deformation medical image registration;
DFRec: Dual Fluctuation Modeling of Multi-level Intent Evolution for Next-Item Recommendation
Nengjun Zhu (Shanghai University), Hang Yu (Shanghai University)
Recommendation SystemRecurrent Neural NetworkTransformerSequential
🎯 What it does: Proposed the DFRec framework for next-item recommendation by explicitly modeling the fluctuation amplitude of user intent.
DGKAN: Dual-branch Graph Kolmogorov-Arnold Network for Unsupervised Multimodal Change Detection
Tongfei Liu (Shaanxi University of Science and Technology), Zhiyong Lv (Shaanxi University of Science and Technology)
SegmentationGraph Neural NetworkAuto EncoderMultimodality
🎯 What it does: Propose an unsupervised multimodal change detection method that uses a dual-branch GKAN (Graph Kolmogorov–Arnold Network) to construct an autoencoder, extracting common features of spatial-spectral structures and directly comparing them to obtain a change map.
DGP: A Dual-Granularity Prompting Framework for Fraud Detection with Graph-Enhanced LLMs
Yuan Li (National University Of Singapore), Cheng Chen (National University Of Singapore)
Anomaly DetectionLarge Language ModelPrompt EngineeringTextGraphTabularFinance Related
🎯 What it does: Proposed and implemented the Dual-Granularity Prompting (DGP) framework, which classifies heterogeneous fraud detection graphs using text prompts and graph-enhanced LLMs, retaining fine-grained text of target nodes and performing coarse-grained compression of neighbor information.
DGSAN: Dual-Graph Spatiotemporal Attention Network for Pulmonary Nodule Malignancy Prediction
Xiao Yu, Changmiao Wang (Shenzhen Research Institute Of Big Data)
ClassificationGraph Neural NetworkTransformerBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a Dual-Graph Spatiotemporal Attention Network (DGSAN) for predicting the malignancy of lung nodules.
DGTF: Cross-Domain Decentralized Graph Learning with Topology-Aware Knowledge Fusion
Ruisheng Zheng, Dongxiao Yu (Shandong University)
Federated LearningKnowledge DistillationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposed a cross-domain decentralized graph learning framework DGTF to address negative transfer caused by inconsistent domain labels and model heterogeneity.
DHCM-CACL: Dynamic Hierarchical Cross-modal Mamba with Confidence-Adaptive Contrastive Learning for Multimodal Emotion Recognition
Baiqiang Wu (Beihang University), Yang Li (Beihang University)
ClassificationRecognitionContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose a self-supervised multimodal emotion recognition framework named DHCM-CACL, which integrates EEG and facial expression modalities for emotion classification.
DHMRec: Collaboration-Guided Multimodal Disentanglement and Hierarchical Fusion for Recommendation
Xiaohan Zhan, Zhiyong Chen (Shandong University)
Recommendation SystemGraph Neural NetworkMultimodality
🎯 What it does: Proposed the DHMRec framework in multi-modal recommendation, first separating public and exclusive features through collaborative modal decoupling, then enhancing recommendation performance via graph neural network multi-view learning and hierarchical fusion.
DiA-gnostic VLVAE: Disentangled Alignment-Constrained Vision Language Variational AutoEncoder for Robust Radiology Reporting with Missing Modalities
Nagur Shareef Shaik (Georgia State University), Dong Hye Ye (UST)
GenerationMixture of ExpertsVision Language ModelAuto EncoderContrastive LearningImageTextMultimodalityBiomedical DataComputed TomographyElectronic Health Records
🎯 What it does: Designed a system capable of generating accurate radiology reports even when clinical information is missing and multimodal features are mixed.
DIAA: A Decoding-Efficient Inference Acceleration Approach for On-Device Large Language Models
Hao Tian (Nanjing University), Wanchun Dou (Macquarie University)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Proposed the DIAA method, achieving model-agnostic and draft-model-free inference acceleration for large language models (LLMs) on edge devices;
Diagnostic-Guided Dynamic Profile Optimization for LLM-based User Simulators in Sequential Recommendation
Hongyang Liu (Macquarie University), Xinghua Qu (Macquarie University)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringSequential
🎯 What it does: Proposes the DGDPO (Diagnostic-Guided Dynamic Profile Optimization) framework, which uses a diagnosis-treatment two-phase LLM mechanism to iteratively optimize user profiles and combines with a sequence recommender to achieve multi-round interaction and bidirectional evolution.
DiagramGPT-Llama3:Enabling Editable, High-Fidelity Diagram Generation with Vision Large Language Models
Yongyuan Chen (Zhejiang University), Xicheng Han (Zhejiang University)
RestorationGenerationTransformerLarge Language ModelVision Language ModelMultimodalityGraph
🎯 What it does: Construct an editable, high-fidelity chart generation system based on large language models and visual Transformers, supporting three tasks: text-to-chart, image-to-chart (restoration), and style transfer.
DialoGen: Towards Dialog Gesture Generation via Identity-Decoupled Style Guidance in Interactive Diffusion Model
Weiyu Zhao, Shengping Zhang (Harbin Institute of Technology)
GenerationTransformerDiffusion modelContrastive LearningVideo
🎯 What it does: Developed the DialoGen framework, which can synchronize the generation of identity-specific gestures for two speakers in a dialogue scenario;
DialogXpert: Driving Intelligent and Emotion-Aware Conversations Through Online Value-Based Reinforcement Learning with LLM Priors
Tazeek Bin Abdur Rakib (Monash University), Soujanya Poria (Nanyang Technological University)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This paper proposes DIALOGXPERT, a proactive dialogue planning framework based on frozen LLM priors, lightweight Q-networks, and emotional tracking, capable of achieving high success rates in multi-task dialogues within fewer than three rounds.
DiAPR: Dimensionally-Allocated Prototype Refinement for Non-Exemplar Class Incremental Learning
Ruixuan Gao (Sichuan University), Keren Fu (Sichuan University)
ClassificationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a fine-grained prototype refinement framework named DiAPR for non-sample class incremental learning, which gradually refines historical prototypes through three modules: distribution-aware pairing, dimensional allocation refinement, and cross-dimensional transition, to enhance classification decision boundaries.
DiCaP: Distribution-Calibrated Pseudo-labeling for Semi-Supervised Multi-Label Learning
Bo Han (Southeast University), Yuheng Jia (Southeast University)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: For semi-supervised multi-label learning, a method is proposed to enhance model performance by calibrating the distribution of pseudo-label weights.
DICE: Distilling Classifier-Free Guidance into Text Embeddings
Zhenyu Zhou (Zhejiang University), Siwei Lyu (University at Buffalo)
GenerationKnowledge DistillationDiffusion modelImageText
🎯 What it does: This study proposes the DICE method, which trains a lightweight sharpener to weight text embeddings without guided sampling, thereby achieving image quality comparable to traditional Classifier-Free Guidance (CFG) without increasing additional model evaluations;
DIET: Machine Unlearning on a Data-Diet
Nilakshan Kunananthaseelan (Monash University), Mehrtash Harandi (Monash University)
Safty and PrivacyRepresentation LearningVision Language ModelImage
🎯 What it does: Proposes a framework called DIET that performs machine unlearning in vision-language models without the need for retained data
Diff-NAT: Better Naturalistic and Aggressive Adversarial Attacks via Class-Optimized Diffusion for Object Detection
Qinglong Yan (Wuhan University), Jiayi Ma (Wuhan University)
Object DetectionAdversarial AttackTransformerDiffusion modelImage
🎯 What it does: This paper proposes Diff-NAT, a naturalistic physical adversarial patch generation method based on class-optimized diffusion, which can significantly interfere with object detection models while maintaining a highly natural appearance.
Diff-V2M: A Hierarchical Conditional Diffusion Model with Explicit Rhythmic Modeling for Video-to-Music Generation
Shulei Ji (Zhejiang University), Kejun Zhang (Zhejiang University)
GenerationTransformerDiffusion modelAuto EncoderVideoMultimodalityAudio
🎯 What it does: Propose a Video-to-Music generation framework called Diff-V2M based on hierarchical conditional diffusion models, which explicitly models rhythm and integrates multi-perspective visual features such as emotion and semantics, enabling the prediction of rhythm and generation of high-quality background music without audio input.
DIFFA: Large Language Diffusion Models Can Listen and Understand
Jiaming Zhou (Nankai University), Xuelong Li (China Telecom)
RecognitionLarge Language ModelDiffusion modelAudio
🎯 What it does: Developed DIFFA, the first large-scale audio-language model based on diffusion, which utilizes a frozen diffusion LLM and a lightweight dual adapter to achieve speech understanding and instruction execution.