AAAI 2026 Papers — Page 21
AAAI Conference on Artificial Intelligence · 4149 papers
LAMDA: Two-Phase HPO via Learning Prior from Low-Fidelity Data
Fan Li, Ke Li (Central South University)
OptimizationHyperparameter SearchTabularBenchmark
🎯 What it does: Propose the Lambda two-phase HPO framework, which first learns a reliable prior in low-fidelity (LF) tasks and then uses this prior to guide high-fidelity (HF) search.
LAMDAS: LLM as an Implicit Classifier for Domain-specific Data Selection
Jian Wu (Ant Group), Yue Zhang (Westlake University)
ClassificationComputational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Proposes the LAMDAS method, which utilizes a pre-trained LLM as an implicit classifier to select domain-specific data, addressing the challenges of scarce high-quality data and noise in large volumes of unverified data.
LAMIC: Layout-Aware Multi-Image Composition via Scalability of Multimodal Diffusion Transformer
Yuzhuo Chen, Weiming Zhang (University Of Science And Technology Of China)
GenerationTransformerVision Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Propose a zero-training multi-graph layout-aware synthesis framework named LAMIC, achieving joint generation of multiple reference images and spatial layouts.
LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models
Alvi Md Ishmam (Virginia Tech), Chris Thomas (Virginia Tech)
Adversarial AttackLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Proposed a black-box method called LAMP to learn universal adversarial perturbations (UAP), enabling attacks on multi-image input multimodal large language models (MLLMs).
LampQ: Towards Accurate Layer-wise Mixed Precision Quantization for Vision Transformers
Minjun Kim (Seoul National University), U Kang (Seoul National University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: Propose a visual Transformer PTQ method based on hierarchical mixed-precision quantization (LAMPQ), which can assign different bit-widths to each layer, significantly improving the accuracy of low-precision models
LandCraft: Designing the Structured 3D Landscapes via Text Guidance
Zhihao Liu (University of Tokyo), Naoto Yokoya (University of Tokyo)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageTextMesh
🎯 What it does: Propose the LandCraft system, which generates high-quality, editable large-scale 3D landscapes based on text.
Landsat30-AU: A Vision-Language Dataset for Australian Landsat Imagery
Sai Ma (Australian National University), John A. Taylor (Australian National University)
Large Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the Landsat30-AU audio-visual language dataset (with two subsets: CAP and VQA), covering four satellites (Landsat 5/7/8/9), spanning 36 years, and with 30-meter resolution, and conducted benchmark evaluations on existing Vision-Language Models (VLMs).
LangGPS: Language Separability Guided Data Pre-Selection for Joint Multilingual Instruction Tuning
Yangfan Ye (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
Data-Centric LearningText
🎯 What it does: Propose a two-stage lightweight data pre-selection framework called LangGPS, which uses language separability as a guidance signal to filter multilingual instruction-tuning data;
Language Drift in Multilingual Retrieval-Augmented Generation: Characterization and Decoding-Time Mitigation
Bo Li (Hebei University of Technology), Rui Xie (Hebei University of Technology)
GenerationRetrievalTransformerTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Investigated the language drift phenomenon in multilingual retrieval-augmented generation (RAG) and proposed a lightweight decoding control method called Soft Constrained Decoding (SCD) to alleviate drift.
Language Model Distillation: A Temporal Difference Imitation Learning Perspective
Zishun Yu (University of Illinois Chicago), Xinhua Zhang (University of Illinois Chicago)
Knowledge DistillationTransformerText
🎯 What it does: This paper proposes a language model distillation framework based on temporal difference (TD) learning, named Bellman Distill, which achieves efficient distillation by screening the action space based on the sparsity of the teacher model distribution.
Language-Guided and Motion-Aware Gait Representation for Generalizable Recognition
ZhengXian Wu (Tsinghua University), Haoqian Wang (Tsinghua University)
RecognitionTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Propose the LMGait framework, combining language-guided motion perception with motion temporal capture to enhance gait recognition performance.
Large Connectome Model: An fMRI Foundation Model of Brain Connectomes Empowered by Brain-Environment Interaction in Multitask Learning Landscape
Ziquan Wei (University of North Carolina at Chapel Hill), Guorong Wu (University of North Carolina at Chapel Hill)
ClassificationRecognitionTransformerSupervised Fine-TuningBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Proposed a Large-Scale Brain Connectivity Model (LCM) using a decoder-only Transformer architecture, pre-trained on a large number of functional connectivity matrices (fMRI) through multi-task learning, and fine-tuned semi-supervisedly to achieve various clinical and behavioral prediction tasks.
Large Language Model Unlearning for Source Code
Xue Jiang (Peking University), Ge Li (Alibaba Group)
AI Code AssistantLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposes a precise forgetting method called PROD for LLM source code, which suppresses undesirable code at the token level and redistributes probabilities to preserve the language model's knowledge of programming languages;
Large Language Models Meet Extreme Multi-label Classification: Scaling and Multi-modal Framework
Diego Ortego (NielsenIQ), Juan C. SanMiguel (NielsenIQ)
ClassificationTransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: In the extreme multi-label classification (XMC) task, this paper investigates how to effectively utilize large decoder language models and visual metadata, and proposes a multimodal framework called ViXML;
Large Language Models Struggle with Unreasonability in Math Problems
Jingyuan Ma (Peking University), Zhifang Sui (Huawei)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Constructed the UMP benchmark to evaluate LLMs' ability to detect and respond to unreasonable math problems.
LAS: Loss-less ANN-SNN Conversion for Fully Spike-Driven Large Language Models
Long Chen (Sichuan University), Yanan Sun (Sichuan University)
Computational EfficiencySpiking Neural NetworkTransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper proposes the LAS framework, which losslessly converts pre-trained large language models into fully spiking neural networks, achieving inference with lower energy consumption.
Latent Knowledge-Guided Video Diffusion for Scientific Phenomena Generation from a Single Initial Frame
Qinglong Cao (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
GenerationData SynthesisSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderOptical FlowImageVideoPhysics Related
🎯 What it does: Proposed a parameterizable video diffusion framework based on latent knowledge of potential scientific phenomena, generating physically plausible fluid and meteorological videos from single-frame initial images.
Latent Self-Consistency for Reliable Majority-Set Selection in Short- and Long-Answer Reasoning
Jungsuk Oh (Seoul National University), Jay-Yoon Lee (Seoul National University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the Latent Self-Consistency (LSC) method, which appends learnable summary tokens after generation by large language models, utilizing contrastive learning to obtain semantic embeddings, thereby achieving unified consistency selection for both short and long answers.
Latent State-Predictive Exploration for Deep Reinforcement Learning
Yiming Wang (University of Macau), Leong Hou U (University of Macau)
Robotic IntelligenceReinforcement LearningDiffusion modelImage
🎯 What it does: Propose an exploration framework LSPE that leverages latent state prediction and directional variation rewards to enhance sample efficiency of deep RL in high-dimensional, sparse reward long-horizon tasks.
LatentLLM: Activation-Aware Transform to Multi-Head Latent Attention
Toshiaki Koike-Akino (Mitsubishi Electric Research Laboratories), Matthew Brand
CompressionTransformerTextMultimodality
🎯 What it does: Perform training-free compression on pre-trained large language models and multi-modal models, converting them into a low-dimensional multi-head latent attention (MLA) structure.
LatentVLA: Taming Latent Space for Generalizable and Long-Horizon Bimanual Manipulation
Junming Wang (University of Hong Kong)
Robotic IntelligenceTransformerSupervised Fine-TuningVision-Language-Action ModelDiffusion modelVideoTextSequential
🎯 What it does: Proposed a vision-language-action framework called LatentVLA for long-term dual-arm operations, which includes a three-stage pipeline: Temporal-Attentive Latent Action Model (TA-LAM) for continuous latent action learning, Latent Action Diffusion Transformer (LADT) for long-term planning, and finally an expert policy head that maps latent plans to robot control commands.
LaTeX2Layout: High-Fidelity, Scalable Document Layout Annotation Pipeline for Layout Detection
Feijiang Han (University of Pennsylvania), Lyle Ungar (University of Pennsylvania)
Object DetectionData SynthesisSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Built a pipeline that directly generates pixel-level layout annotations using a LaTeX compiler, and fine-tuned a general-purpose vision-language model by programatically generating synthetic data to accomplish document layout parsing.
LayerEdit: Disentangled Multi-Object Editing via Conflict-Aware Multi-Layer Learning
Fengyi Fu (University Of Science And Technology Of China), Zhendong Mao (University Of Science And Technology Of China)
Image HarmonizationTransformerDiffusion modelImageBenchmark
🎯 What it does: LayerEdit achieves untrained multi-object text-driven image editing through multi-layer decomposition, conflict-aware editing, and transparency-guided fusion.
Laytrol: Preserving Pretrained Knowledge in Layout Control for Multimodal Diffusion Transformers
Sida Huang (Northwestern Polytechnical University), Hongyuan Zhang (University Of Hong Kong)
GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposed the Laytrol network for layout-to-image generation, leveraging parameter copying to retain pre-trained knowledge, constructed the LaySyn dataset, and designed specialized initialization, object-level RoPE, and random prompt dropping mechanisms.
LC3: Long Cross-Language Code Clone Detection Enhanced by Opcode Sequences and Affinity Aggregation
Xilin Lan (Central South University), Li Kuang (Central South University)
RetrievalRecurrent Neural NetworkTransformerLarge Language ModelMultimodality
🎯 What it does: Proposes the LC3 framework, which addresses the issues of insufficient language-agnostic representations and loss of long-code information in cross-language code clone detection by fusing source code and opcode sequences through bimodal encoding.
LeanRAG: Knowledge-Graph-Based Generation with Semantic Aggregation and Hierarchical Retrieval
Yaoze Zhang (Shanghai Artificial Intelligence Laboratory), Botian Shi (Shanghai Artificial Intelligence Laboratory)
RetrievalComputational EfficiencyRepresentation LearningTransformerLarge Language ModelTextGraphAgriculture RelatedRetrieval-Augmented Generation
🎯 What it does: LeanRAG proposes a framework for semantic aggregation and hierarchical retrieval on knowledge graphs, combining structured retrieval with LLM generation;
Leap of FAITH from GNN-to-MLP: Fairness Aware Inference via DisTillation of GrapH Knowledge
Vipul Kumar Singh (Indian Institute of Technology Delhi), Jayadeva (Indian Institute of Technology Delhi)
Knowledge DistillationGraph Neural NetworkGraphTabular
🎯 What it does: Transfer the topological-aware representations of graph neural networks (GNN) to multi-layer perceptrons (MLP) via knowledge distillation, achieving graph-free inference while maintaining fairness.
Learn from Global Correlations: Enhancing Evolutionary Algorithm via Spectral GNN
Kaichen Ouyang, Dayu Hu (Sun Yat-sen University)
OptimizationGraph Neural NetworkBenchmark
🎯 What it does: Proposed a evolutionary algorithm framework GNE based on spectral graph neural networks, modeling the population as a graph and utilizing spectral filtering to update individuals, thereby achieving global information learning.
Learnable Matrix Profile for Motif Discovery on Multivariate Time Series
Mingkai Lin (Nanjing University), Wenzhong Li (Nanjing University)
ClassificationAnomaly DetectionGraph Neural NetworkTime SeriesElectrocardiogramBenchmark
🎯 What it does: Propose a learnable multi-dimensional matrix spectrum (L-MAP) framework for efficiently discovering frequent subsequences (motif) and detecting anomalies (discord) in multivariate time series.
Learnable Permutation for Structured Sparsity on Transformer Models
Zekai Li (Advanced Micro Devices Inc), Emad Barsoum (Advanced Micro Devices Inc)
Computational EfficiencyKnowledge DistillationTransformerImageTextMultimodality
🎯 What it does: Proposes an end-to-end learnable channel permutation framework to achieve structured sparsity (N:M sparsity) in Transformer models by rearranging the weight matrix to enhance performance after pruning.
Learner-Tailored Program Repair: A Solution Generator with Iterative Edit-Driven Retrieval Enhancement
Zhenlong Dai (Zhejiang University), Jingyuan Chen (Zhejiang University)
AI Code AssistantTransformerLarge Language ModelTextSequentialBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed the Learner-Customized Program Repair Task (LPR) and designed the LSGEN framework, which generates repair code and corresponding bug descriptions by leveraging retrieval databases, edit-based retrieval, differential analysis, and iterative retrieval enhancement;
Learning 3D Occupancy from Beam Overlap in 2D Rotating mmWave Radar
Yu Du (Dalian University of Technology), Weimin Wang (Dalian University of Technology)
SegmentationAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: Propose a method that learns 3D occupancy prediction from single-frame radar data by leveraging beam overlap and intensity differences in rotating 2D mmWave radar scans.
Learning 3D Texture-Aware Representations for Parsing Diverse Human Clothing and Body Parts
Kiran Chhatre (Kth Royal Institute Of Technology), Srikrishna Karanam (Kth Royal Institute Of Technology)
SegmentationConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: Propose Spectrum, a unified network that achieves pixel-level parsing of human body parts and clothing, as well as instance-level grouping, in a single image.
Learning a Fix and Explore Framework for Continuous Generalized Category Discovery
Chunming Li (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)
ClassificationRecognitionKnowledge DistillationRepresentation LearningContrastive LearningImage
🎯 What it does: Proposes the Fix and Explore (FaE) framework designed to simultaneously retain memory of old classes and discover new classes in continuous general category discovery tasks.
Learning Adaptive and Expandable Mixture Model for Continual Learning
Fei Ye (University of Electronic Science and Technology of China), ShiJie Zhou (University of Electronic Science and Technology of China)
ClassificationDomain AdaptationTransformerMixture of ExpertsImageBenchmark
🎯 What it does: In the multi-domain task incremental learning scenario, we propose a continuous learning framework based on pre-trained models, which includes dual representation backbone networks, an expandable mixture of experts module, and an adaptive fusion with dynamic knowledge calibration mechanism.
Learning Better UAV-Based Cross-View Object Geo-Localization from Multi-Modal Prompts: MoP-UAV Benchmark and MoPT Framework
Xiaohan Zhang (Zhejiang University), Hui-Liang Shen (Zhejiang University)
RetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Proposes the MoP-UAV multimodal prompting benchmark and the MoPT multimodal prompting-guided cross-view object geolocation framework, supporting the joint use of three types of prompts: language, bounding box, and point.
Learning Beyond Domains: Misleading Prompts and Pseudo-Label Contrast for Text Domain Generalization
Qizhi Li (Sichuan University), Xu Wang (Sichuan University)
ClassificationDomain AdaptationTransformerPrompt EngineeringContrastive LearningText
🎯 What it does: Achieve text domain generalization through prompting and pseudo-label contrastive learning, proposing the GenPromptCL framework;
Learning Beyond Vision: Vision-Language Distillation and Edge-Aware Mix Diffusion in Semi-Supervised Semantic Segmentation
Rui Yang (Shanghai University), Shaorong Xie (Shanghai University)
SegmentationKnowledge DistillationTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: In semi-supervised semantic segmentation, a pseudo-label refinement framework called ViLaDiff is constructed by generating image descriptions, fusing them with visual features, and introducing mixed noise diffusion in the label space.
Learning Branching Policies for MILPs with Proximal Policy Optimization
Abdelouahed Ben Mhamed (University Mohammed VI Polytechnic), Amal Seghrouchni (Sorbonne University)
OptimizationTransformerReinforcement LearningTabular
🎯 What it does: Proposes the Tree-Gate branching strategy TGPPO based on PPO, which directly learns variable selection during the Branch-and-Bound process.
Learning Cell-Aware Hierarchical Multi-Modal Representations for Robust Molecular Modeling
Mengran Li (Sun Yat-sen University), Stan Z. Li (Sun Yat-sen University)
Representation LearningDrug DiscoveryGraph Neural NetworkContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose a multi-modal representation framework (CHMR) that jointly models molecular structure, cell phenotypes, and gene expression to address missing cell modalities and hierarchical dependency issues.
Learning Compact Latent Space for Representing Neural Signed Distance Functions with High-fidelity Geometry Details
Qiang Bai, Zhizhong Han (Wayne State University)
GenerationRepresentation LearningPoint CloudMesh
🎯 What it does: Investigated a dual-branch neural SDF network that utilizes a shared compact latent space and spatial voxel grid features, enabling the representation of multiple 3D shapes on a single latent code, while enhancing high-frequency detail reconstruction through a balanced sampling strategy.
Learning Conjugate Direction Fields for Planar Quadrilateral Mesh Generation
Jiong Tao (University of Bath), Bailin Deng (Cardiff University)
GenerationData SynthesisGraph Neural NetworkMesh
🎯 What it does: Utilizes deep learning to predict conjugate direction fields (CDF) on freeform surfaces and directly generates initial layouts suitable for constructing planar quadrilateral meshes, avoiding the traditional costly nonlinear optimization process.
Learning Depth from Past Selves: Self-Evolution Contrast for Robust Depth Estimation
Jing Cao (Harbin Institute Of Technology), Yong Huang (Harbin Institute Of Technology)
Depth EstimationTransformerContrastive LearningImage
🎯 What it does: Proposes the SEC-Depth framework, which constructs a self-evolving contrastive loss using historical models (Latency Models) during training to enhance the robustness of self-supervised depth estimation in adverse weather conditions.
Learning DFAs from Positive Examples Only via Word Counting
Benjamin Bordais (TU Dortmund University), Daniel Neider (TU Dortmund University)
OptimizationComputational EfficiencyTextSequential
🎯 What it does: This study proposes a new perspective based on word counting for learning deterministic finite automata (DFA) using only positive examples, and provides the corresponding NP-completeness proof; meanwhile, it designs an integer linear programming (ILP) solver and a heuristic preprocessing algorithm;
Learning Diffusion Policy from Primitive Skills for Robot Manipulation
Zhihao Gu (University of Hong Kong), Dong Xu (University of Hong Kong)
Robotic IntelligenceTransformerMixture of ExpertsVision Language ModelDiffusion modelMultimodality
🎯 What it does: Proposed a Skill-Diffusion Policy (SDP) based on fine motor skills, which decomposes complex tasks into eight reusable primitive skills and dynamically assigns skills through a visual-language model and a router network, achieving more precise low-level action generation.
Learning Diverse Bimanual Dexterous Manipulation Skills from Human Demonstrations
Bohan Zhou (Peking University), Zongqing Lu (Peking University)
Knowledge DistillationData-Centric LearningRobotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: Automatically construct a large number of dexterous manipulation tasks using human hand demonstrations and train a general visual policy through a teacher-student learning framework.
Learning Dynamics as Feedback: An Adaptive Entropy Flow Dynamics Framework for Long-tailed Human Action Recognition
Yuan Dong (University of Science and Technology of China), Pengkun Wang (University of Science and Technology of China)
ClassificationRecognitionGraph Neural NetworkVideo
🎯 What it does: Proposed a closed-loop self-regulating framework AEED, which uses entropy flow to monitor learning progress and dynamically adjust class weights, thereby improving the performance of long-tailed action recognition.
Learning Fair Graph Representations via Probability of Necessity and Sufficiency
Chuxun Liu, Lin Liu (University Of South Australia)
Safty and PrivacyRepresentation LearningGraph Neural NetworkGenerative Adversarial NetworkGraph
🎯 What it does: Propose FairSNR, a framework that learns fair graph representations through probabilistic necessity and sufficiency (PNS) constraints, balancing predictive performance and fairness.
Learning Fair Representations with Kolmogorov-Arnold Networks
Amisha Priyadarshini (University of California, Irvine), Sergio Gago-Masague (University of California, Irvine)
Explainability and InterpretabilityRepresentation LearningTabular
🎯 What it does: Propose an adversarial debiasing model based on the Kolmogorov-Arnold network (KAN) and introduce an adaptive fairness penalty update mechanism;
Learning from Answer Sets via Single-Shot Disjunctive ASP Encoding
Roberto Borelli (University of Udine), Agostino Dovier (University of Udine)
Computational EfficiencyBenchmark
🎯 What it does: This paper proposes a discrete ASP encoding with a single projection to efficiently solve the inductive learning from answer sets (LAS) task, replacing traditional multi-round iterative ASP calls;
Learning from Guidelines: Structured Prompt Optimization for Expert Annotation Tasks
Wenliang Zhong (University of Texas at Arlington), Junzhou Huang (University of Texas at Arlington)
OptimizationData-Centric LearningTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes a guidance document-based prompting optimization framework (GDP), enabling large language models to follow complex domain guidelines and complete annotation tasks with only a few labeled samples.
Learning from Human Gaze: Human-like Robot Social Navigation in Dense Crowds
Zhecheng Yu (Southeast University), Brian Y. Lim (Southeast University)
Robotic IntelligenceConvolutional Neural NetworkTransformerVideoMultimodality
🎯 What it does: Developed the GazeNav dataset and the Gaze2Nav framework, enabling robot social navigation in crowded environments by utilizing human gaze information
Learning from Imperfect Data: Robust Inference of Dynamic Systems Using Simulation-Based Generative Model
Hyunwoo Cho (Pohang University of Science and Technology), Hyung Ju Hwang (Pohang University of Science and Technology)
OptimizationData-Centric LearningGenerative Adversarial NetworkTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: Propose a SiGMoID framework based on HyperPINN and Wasserstein GAN for simultaneously estimating system parameters, quantifying noise, and reconstructing unobserved system states from noise-sparse, partially observable data.
Learning from Long-Term Engagement: Adaptive Tutoring Dialogue Planning for Personalized Education
Zhiang Dong (Zhejiang University), Jingyuan Chen (Zhejiang University)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Designed and implemented the LEAP system, which leverages students' long-term learning processes for adaptive tutoring planning, and constructed the LEAD dataset based on real student multi-round submissions.
Learning from Reasoning Failures via Synthetic Data Generation
Gabriela Ben Melech Stan (Intel Labs), Phillip Howard (Thoughtworks)
Data SynthesisTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a synthetic data generation method based on the failure analysis of existing multimodal models, utilizing state-of-the-art models to automatically diagnose errors and generate targeted new question-answer pairs and image descriptions, ultimately constructing 553k high-quality multimodal instruction tuning data.
Learning from Scoring Disagreements: Contrastive Error Mining for Efficient and Robust LLM-based Assessment
Lei Chen (Jinan University), Weiqi Luo (Jinan University)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Propose the CEM-FT framework, which automatically identifies high-value hard samples where the score differences between fully fine-tuned models and few-shot models are significant, and applies lightweight LoRA fine-tuning on these samples to improve the accuracy and consistency of LLM automatic scoring.
Learning from the Undesirable: Robust Adaptation of Language Models Without Forgetting
Yunhun Nam (Korea University), Jongheon Jeong (Korea University)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a regularization method called Learning-from-the-Undesirable (LfU) for supervised fine-tuning of large language models under limited data conditions, significantly reducing overfitting while retaining pre-trained knowledge.
Learning Heuristic Functions for HTN Planning
Daniel Höller
OptimizationGraph Neural NetworkGraphBenchmark
🎯 What it does: This paper first proposes a method to learn heuristic functions in hierarchical task network (HTN) planning.
Learning Heuristic Functions with Graph Neural Networks for Numeric Planning
Valerio Borelli (University of Brescia), Ivan Serina (University of Brescia)
OptimizationGraph Neural NetworkGraphBenchmark
🎯 What it does: This paper proposes a heuristic function based on graph neural networks (GNN) for solving numerical planning problems, evaluated on multiple IPC 2023 numerical planning benchmarks.
Learning in Zero-Sum Markov Games: Relaxing Strong Reachability and Mixing Time Assumptions
Reda Ouhamma (Ecole Polytechnique Federale De Lausanne), Maryam Kamgarpour (Ecole Polytechnique Federale De Lausanne)
Reinforcement Learning
🎯 What it does: Proposed a reward-based decentralized learning algorithm for infinite-horizon zero-sum Markov games, achieving polynomial-time convergence to ε-Nash equilibrium without relying on strong reachability or uniform mixing time assumptions.
Learning Intrinsic Hierarchy for Generalized Category Discovery
Yu Duan (Xidian University), Quanxue Gao (Xidian University)
RecognitionContrastive LearningImage
🎯 What it does: Propose a lightweight module LEAH, which utilizes learnable queries and a hierarchical constructor to automatically extract key objects and model hierarchical relationships within images in the Generalized Category Discovery task
Learning Knowledge from Textual Descriptions for 3D Human Pose Estimation
Yi Wu (University of Science and Technology of China), Linxiang Tan (China Merchants Bank)
Pose EstimationTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper proposes a method that utilizes automatically generated text descriptions to assist in 3D human pose estimation by aligning text features with pose features to reduce depth ambiguity.
Learning Label Distribution with Dirichlet Process Mixture Model
Minglong Wang (Nanjing University of Aeronautics and Astronautics), Xiuyi Jia (Nanjing University of Science and Technology)
ClassificationRepresentation LearningData-Centric LearningImage
🎯 What it does: This paper proposes a label distribution learning framework (LDL-DPM) based on the Dirichlet process mixture model (DPMM), aiming to capture annotator heterogeneity and adaptively determine the number of mixture components.
Learning Latent Imaging Biomarkers for Interpretable Microvascular Invasion Prediction in Hepatocellular Carcinoma
Ji Rao (Tongji University), Ye Luo (Tongji University)
ClassificationExplainability and InterpretabilityTransformerContrastive LearningMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a two-stage interpretable framework IRCL, which first learns potential image features through dual-layer contrastive learning and clusters them to generate imaging biomarkers, then predicts microvascular invasion (MVI) by aligning these markers with patient features in the original images, and achieves spatial interpretation of image markers through a learnable mask.
Learning Network Dismantling Without Handcrafted Inputs
Haozhe Tian (Imperial College London), Homayoun Hamedmoghadam (RMIT University)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: Propose a network decomposition method called MIND based on graph neural networks and reinforcement learning, completely independent of manual features;
Learning Neural Operators from Partial Observations via Latent Autoregressive Modeling
Jingren Hou (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)
RestorationBenchmarkPhysics Related
🎯 What it does: Proposes a method for training neural operators under partially observed data, incorporating mask prediction training strategies and a physics-aware latent autoregressive propagation module.
Learning Object-Centric Motion Priors from Human for Robotic Dexterous Manipulation
Zhengdong Hong (Zhejiang University), Guofeng Zhang (Zhejiang University)
Robotic IntelligenceConvolutional Neural NetworkTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes a motion prior learned from human hand-object interaction data, and utilizes this prior to guide reinforcement learning to achieve dexterous manipulation of different objects by multi-fingered robots, and realize zero-shot simulation-to-reality transfer.
Learning Optimal Prompt Ensemble for Multi-source Visual Prompt Transfer
Enming Zhang (Tsinghua University), Yang Li (Tsinghua University)
ClassificationTransformerPrompt EngineeringImage
🎯 What it does: Propose the HGPrompt framework for adaptive weighted fusion of multi-source visual prompts, significantly enhancing the transferability of frozen Vision Transformers on downstream tasks.
Learning Personalised Human Internal Cognition from External Expressive Behaviours for Real Personality Recognition
Xiangyu Kong (University of Exeter), Siyang Song (University of Oxford)
RecognitionGraph Neural NetworkTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: Propose a real-time personality recognition framework based on audio-visual behavior simulation of individual internal cognition, achieving regression of real personality traits by analyzing cognitive graphs using a personalized network weight generator and 2D graph neural network.
Learning Procedural-Aware Video Representations Through State-Grounded Hierarchy Unfolding
Jinghan Zhao (Beihang University), Feng Lu (Beihang University)
Representation LearningConvolutional Neural NetworkLarge Language ModelVideoTextMultimodality
🎯 What it does: Propose a Task-Step-State (TSS) three-layer framework, integrating a visual state layer into the task-step hierarchy to address the semantic gap between abstract descriptions and visual data, and design a progressive pre-training strategy to gradually unfold this hierarchical structure;
Learning Protein–Ligand Binding in Hyperbolic Space
Jianhui Wang (Tsinghua University), Yanyan Lan (Tsinghua University)
Drug DiscoveryGraph Neural NetworkTransformerContrastive LearningBiomedical Data
🎯 What it does: Propose HypSeek, a three-tower architecture that embeds ligands, protein pockets, and protein sequences into the hyperbolic space of the Lorentz model, to unify virtual screening and affinity ranking.
Learning Spatial Decay for Vision Transformers
Yuxin Mao (Northwestern Polytechnical University), Yuchao Dai (Northwestern Polytechnical University)
ClassificationGenerationTransformerLarge Language ModelImage
🎯 What it does: Propose a Vision Transformer model called SDT based on data-dependent spatial decay, achieving adaptive modeling of image spatial relationships.
Learning Structurally Stabilized Representations for Lossless DNA Storage
Ben Cao (Dalian University of Technology), Qiang Zhang (Dalian University)
CompressionRepresentation LearningGraph Neural NetworkImageTextMultimodalityBenchmark
🎯 What it does: Propose an end-to-end DNA storage method RSRL that combines Reed-Solomon codes with single-stranded structured representation learning.
Learning Subgroups with Maximum Treatment Effects Without Causal Heuristics
Lincen Yang (Leiden University), Saber Salehkaleybar (Leiden University)
Explainability and InterpretabilitySupervised Fine-TuningTabular
🎯 What it does: Under the structural causal model framework, the problem of identifying subgroups with maximum average treatment effect is transformed into a standard supervised learning problem, and subgroup discovery is achieved through CART trees;
Learning Systems Expansion with Efficient Heterogeneity-aware Knowledge Transfer
Gaole Dai (Nanyang Technological University), Mo Li (Hong Kong University Of Science And Technology)
OptimizationKnowledge DistillationTransformerMultimodalityBiomedical DataAlzheimer's DiseaseAudio
🎯 What it does: Investigate the problem of expanding learning systems, proposing the HaT framework to achieve efficient heterogeneous perception knowledge transfer.
Learning the Latent Structure: A Feature-Centric Approach to Graph Data Augmentation
Yu Song (Michigan State University), Hui Liu (Michigan State University)
ClassificationRepresentation LearningData-Centric LearningGraph Neural NetworkGraph
🎯 What it does: Proposed a feature-based graph data augmentation framework called SelfAug, which compensates for missing graph structures by learning residuals in the embedding space;
Learning Time in Static Classifiers
Xi Ding (Griffith University), Yongsheng Gao (Griffith University)
ClassificationAnomaly DetectionTransformerImageVideoTime Series
🎯 What it does: Propose a SEQ learning framework that aligns predicted sequences with class-specific time prototypes using soft DTW, enabling a standard feedforward classifier to possess temporal reasoning capabilities.
Learning to Cluster Rare Cell Types: Implicit Semantic Data Augmentation for Spatial Multi-modal Omics Analysis
Daixian Liu (Tsinghua University), Jingcai Guo (Hong Kong Polytechnic University)
Data SynthesisRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose the CRCT framework, achieving clustering of rare cell types through implicit semantic data augmentation and adaptive graph learning
Learning to Collaborate: An Orchestrated-Decentralized Framework for Peer-to-Peer LLM Federation
Inderjeet Singh (Fujitsu Research of Europe), Motoyoshi Sekiya (Fujitsu Research of Europe)
Federated LearningSafty and PrivacyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a decentralized LLM federated learning framework KNEXA-FL, which assigns optimal peer-to-peer knowledge distillation tasks to heterogeneous LLM agents through a centralized matcher, enabling multi-institutional data collaboration without sharing raw data.
Learning to Compress Graphs via Dual Agents for Consistent Topological Robustness Evaluation
Qisen Chai (Southwest University), Tao Jia (Southwest University)
CompressionRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningAgentic AIGraph
🎯 What it does: Proposes a dual-agent reinforcement learning framework called Cutter for compressing large-scale graph structures while preserving graph robustness features.
Learning to Compress: Unlocking the Potential of Large Language Models for Text Representation
Yeqin Zhang (Nanjing University), Cam-Tu Nguyen (Nanjing University)
CompressionKnowledge DistillationRepresentation LearningTransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: Explored using context compression as an unsupervised pre-training task to improve text representations in large language models.
Learning to Cooperate with Minimal Observability
Chin-wing Leung (University of Warwick), Mirco Musolesi (University College London)
Reinforcement Learning
🎯 What it does: In this paper, the authors design an 'Observer Model' that enables reinforcement learning agents to achieve large-scale group cooperation through collaboration and partner selection mechanisms, even with limited observational capabilities.
Learning to Curate Context: Jointly Optimizing Retrieval and Prediction for Multimodal Social Media Popularity
Xovee Xu (University of Electronic Science and Technology of China), Jingkuan Song (University of Electronic Science and Technology of China)
Recommendation SystemOptimizationTransformerMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposes a framework named JRPP, achieving joint optimization of retrieval and multimodal social media popularity prediction.
Learning to Deliberate: Meta-policy Collaboration for Agentic LLMs with Multi-agent Reinforcement Learning
Wei Yang (University of Southern California), Jesse Thomason (University of Southern California)
TransformerReinforcement LearningAgentic AITextBenchmark
🎯 What it does: Propose the Meta-Policy Deliberation Framework (MPDF), enabling multi-agent LLMs to dynamically adjust strategies through internal metacognitive decisions (Persist, Refine, Concede) during collaboration;
Learning to Disentangle Latent Reasoning Rules with Language VAEs: A Systematic Study
Yingji Zhang (University of Manchester), André Freitas
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelAuto EncoderTextRetrieval-Augmented Generation
🎯 What it does: The study explicitly encodes interpretable reasoning rules into the latent space of a language variational autoencoder (VAE) to achieve controllable reasoning in natural language inference (NLI).
Learning to Explore: Policy-Guided Outlier Synthesis for Graph Out-of-Distribution Detection
Li Sun (North China Electric Power University), Philip S. Yu (Beijing University of Posts and Telecommunications)
Data SynthesisAnomaly DetectionGraph Neural NetworkReinforcement LearningAuto EncoderContrastive LearningGraphBenchmark
🎯 What it does: Propose the PGOS framework, which explores low-density regions in the structured latent space obtained from prototype contrastive learning by learning strategies, automatically generating pseudo OOD graphs to enhance unsupervised graph OOD detection.
Learning to Generate and Extract: A Multi-Agent Collaboration Framework for Zero-Shot Document-Level Event Arguments Extraction
Guangjun Zhang, Ru Li (Shanxi University)
Data SynthesisTransformerLarge Language ModelReinforcement LearningAgentic AIText
🎯 What it does: Propose a multi-agent collaborative framework that iteratively generates high-quality synthetic document-level event argumentation data through the interaction between generation agents and evaluation agents, thereby improving the performance of zero-shot event argument extraction.
Learning to Generate Structured Meshes with In-Context: Toward Generalization in Mesh Generation
Jing Xiao (National University of Defense Technology), Jie Liu (National University of Defense Technology)
GenerationData SynthesisMeta LearningTransformerPrompt EngineeringMesh
🎯 What it does: Proposes a meta-learning framework called ICL-Mesh based on in-context learning, enabling rapid mesh generation for unknown geometries using a single network without parameter updates.
Learning to LEAP: Efficient Dense Point Tracking by Focusing Where It Matters
Chenzhi Zhao (Beijing University of Posts and Telecommunications), Wendong Wang (Beijing University of Posts and Telecommunications)
Object TrackingTransformerVideo
🎯 What it does: Design and train a self-supervised Tracking Any Point (TAP) model named LEAP-Track, achieving efficient point tracking through adaptive sparse attention and sparse k-NN random walks.
Learning to Optimize Job Shop Scheduling Under Structural Uncertainty
Rui Zhang (Beihang University), Jing Yuan (University at Buffalo)
OptimizationGraph Neural NetworkReinforcement LearningGraphTabular
🎯 What it does: For the job shop scheduling problem (JSSP) with structural uncertainty, an asynchronous actor-critic (UP-AAC) framework is proposed. It trains the critic using deterministic states reconstructed through hindsight, and provides the actor with global risk information via an uncertainty-aware model (UPM), enabling stable learning of scheduling strategies in uncertain environments.
Learning to Parse and Reconstruct: Bidirectional Modeling of Question-to-Program Mapping
Zeying Duan (Xi'an Jiaotong University), Weijia Wu (Xi'an Jiaotong University)
GenerationAI Code AssistantTransformerAuto EncoderText
🎯 What it does: Propose the BiPaR framework to achieve bidirectional modeling between questions and programs, performing both program parsing and reverse question generation from programs to form complementary learning signals;
Learning to Rank: How GNNs Solve Max-Clique and Sparse PCA
Elad Shoham (Ben-Gurion University of the Negev), Dan Vilenchik (Ben-Gurion University of the Negev)
OptimizationExplainability and InterpretabilityGraph Neural NetworkSpiking Neural NetworkGraphTabularBenchmark
🎯 What it does: The study uses graph neural networks (GNNs) to solve the maximum clique problem and proposes a concept explanation framework based on degree sorting. It employs a Least-Probable Removal (LPR) decoder to enhance performance and demonstrates that the framework is also effective in sparse PCA tasks.
Learning to Tell Apart: Weakly Supervised Video Anomaly Detection via Disentangled Semantic Alignment
Wenti Yin (Huazhong University of Science and Technology), Nong Sang (Huazhong University of Science and Technology)
Anomaly DetectionGraph Neural NetworkTransformerVision Language ModelContrastive LearningVideo
🎯 What it does: This paper proposes the DSANet framework, which performs coarse-to-fine granularity detection and classification of video anomalies in a weakly supervised manner;
Learning Topology-Aware Dynamic Associations for Robust Multi-Person Pose Estimation
Shengnan Hu (Central China Normal University), Yahong Chen (Central China Normal University)
Pose EstimationTransformerImage
🎯 What it does: To address occlusion, scale variation, and complex interactions in multi-person human pose estimation, the authors propose the TopoDA framework.
Learning Topology-Driven Multi-Subspace Fusion for Grassmannian Deep Networks
Xuan Yu (Jiangnan University), Tianyang Xu (Jiangnan University)
ClassificationRecognitionVideoGraphBiomedical Data
🎯 What it does: Proposed a topology-driven multi-subspace fusion framework aimed at capturing complex geometric structures by dynamically selecting and weighting task-related subspaces.
Learning Underwater Image Enhancement Iteratively Without Reference Images
Yi Tang (Kitami Institute of Technology), Hiroshi Masui (Kitami Institute of Technology)
RestorationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes an unsupervised iterative diffusion model framework for underwater image enhancement, decomposing the task into colorization and color compensation, and enhancing warm color information through a quantization mechanism.
Learning Vision-Based Neural Network Controllers with Semi-Probabilistic Safety Guarantees
Xinhang Ma (Washington University in St. Louis), Yevgeniy Vorobeychik (Washington University in St. Louis)
Autonomous DrivingSafty and PrivacyGenerative Adversarial NetworkImage
🎯 What it does: Proposes a semi-probabilistic safety verification framework (SPV) that combines reachability analysis with conditional generative networks and distribution-agnostic tail bounds to achieve scalable verification and training for visual controllers;
Learning Whom to Align With: Progressive Anomaly Combination Detection for Partially View-Aligned Clustering
Hang Gao (Jilin University), You Zhou (Jilin University)
Anomaly DetectionRepresentation LearningAuto EncoderContrastive LearningMultimodality
🎯 What it does: To address the partial view alignment problem in multi-view data, a progressive method is proposed that reinterprets view alignment as anomaly combination detection, combined with triple self-supervised consistency clustering to achieve clustering without noise or paired data.
Learning with Preserving for Continual Multitask Learning
Hanchen David Wang (Vanderbilt University), Meiyi Ma (Vanderbilt University)
Knowledge DistillationRepresentation LearningSupervised Fine-TuningImageTime SeriesBenchmark
🎯 What it does: Proposes a buffer-free continual multi-task learning framework LwP, which avoids catastrophic forgetting by preserving the geometric structure of the shared feature space.
Learning with Structure: Computing Consistent Subsets on Structurally-Regular Graphs
Aritra Banik (National Institute of Science, Education and Research), Abhishek Sahu (National Institute of Science, Education and Research)
OptimizationGraph
🎯 What it does: Studies the Minimum Consistent Subset (MCS) problem in graph metric spaces, proposing fixed-parameter tractable (FPT) algorithms based on two structural parameters: vertex cover number and neighborhood diversity. The algorithm complexities are O∗(k^k) (k is the vertex cover number) and O∗(r^r) (r is the neighborhood diversity), and do not grow exponentially with the number of colors c.