IJCAI 2025 Papers — Page 6
International Joint Conference on Artificial Intelligence · 1014 papers
Improvements to the Generate-and-Complete Approach to Conformant Planning
Liangda Fang (Jinan University), Quanlong Guan (Jinan University)
Benchmark
🎯 What it does: This paper improves the generative-completion (GC) method by incorporating SAT-based verification and counterexample-guided completion processes into constrained planning, and enhances solution quality by eliminating redundant actions.
Improving Consistency Identification in Task-oriented Dialogue Through Multi-Agent Collaboration
Peng Wang (Central South University), Libo Qin (Central South University)
RecognitionLarge Language ModelAgentic AIPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a multi-agent collaborative zero-shot task-oriented dialogue consistency identification framework called MAC-CIToD.
Improving Efficiency of Answer Set Planning with Rough Solutions from Large Language Models for Robotic Task Planning
Xinrui Lin (University of Science and Technology of China), Yanyong Zhang (University of Science and Technology of China)
Computational EfficiencyRobotic IntelligenceTransformerLarge Language ModelText
🎯 What it does: Proposes the CLMASP framework, integrating rough plans generated by large language models with answer set programming (ASP) to achieve efficient execution of robot task planning.
Improving Generalization in Meta-Learning via Meta-Gradient Augmentation
Ren Wang (Shandong University), Yilong Yin (Shandong University)
Meta LearningImage
🎯 What it does: By introducing data-agnostic gradient augmentation (MGAug) in a two-round framework of meta-learning, leveraging network pruning to break memorization and generate diverse meta-gradients, thus enhancing the generalization ability of meta-learning.
Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection
Jianing He (Tongji University), Zhihua Wei (Tongji University)
ClassificationComputational EfficiencyTransformerTextBenchmark
🎯 What it does: Proposes an early exit method based on CAP (Certainty-Aware Probability) scores, integrating class-related logits with class-agnostic information (NSP scores) to improve confidence estimation in predictions.
Imputation-free Incomplete Multi-view Clustering via Knowledge Distillation
Benyu Wu (Shandong University), Guoxian Yu (Shandong University)
Knowledge DistillationRepresentation LearningAuto EncoderContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose an interpolation-free missing multi-view clustering framework called I2MVC, which divides incomplete multi-view data into complete data and fully missing views. The complete data is clustered using a teacher model trained with self-supervised contrastive learning, while the fully missing views are clustered using a lightweight student model obtained through knowledge distillation. The final clustering result is obtained by merging the two parts.
In-context Learning Demonstration Generation with Text Distillation
Wuyuqing Wang (Xidian University), Cheng Deng (Xidian University)
Knowledge DistillationData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: Proposed a demonstration generation framework DDG based on data distillation, which automatically synthesizes representative demonstration examples that capture information from the original dataset by matching gradients between the generator and computational model, as well as employing a teacher-student EMA mechanism, to enhance the In-Context Learning (ICL) performance of large language models.
In-Context Meta LoRA Generation
Yihua Shao (Hong Kong Polytechnic University), Jingcai Guo (Hong Kong Polytechnic University)
Object DetectionMeta LearningConvolutional Neural NetworkSupervised Fine-TuningAuto EncoderImageText
🎯 What it does: In a multi-task environment, the In-Context Meta LoRA (ICM-LoRA) framework is proposed, which utilizes a conditional variational autoencoder (CVAE) to generate LoRA parameters based on task vectors, enabling task-specific customization of large language/visual models without requiring additional fine-tuning.
Incentivizing Safer Actions in Policy Optimization for Constrained Reinforcement Learning
Somnath Hazra (Indian Institute of Technology Kharagpur), Soumyajit Dey (Indian Institute of Technology Kharagpur)
Reinforcement LearningBenchmark
🎯 What it does: Propose a safe reinforcement learning algorithm named Incrementally Penalized Proximal Policy Optimization (IP3O), which balances return maximization with constraint satisfaction by incorporating an adaptive incentive-penalty mechanism into the reward function.
Inconsistency Handling in DatalogMTL
Meghyn Bienvenu (University of Bordeaux), Atefe Khodadaditaghanaki (Paderborn University)
🎯 What it does: This paper proposes and studies three conflict and repair concepts (s, p, i) for handling inconsistencies in DatalogMTL, and systematically analyzes their properties and data complexity.
Inconsistency-Based Federated Active Learning
Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
Federated LearningKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposed a federated active learning framework named IFAL, addressing data heterogeneity and noise issues in federated learning by designing two inconsistency sampling strategies (inter-model inconsistency and intra-model inconsistency) and combining clustering to achieve diversity sampling.
Incorporating Legal Logic into Deep Learning: An Intelligent Approach to Probation Prediction
Qinghua Wang (Shandong University), Cunquan Qu (Shandong University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposed a multi-task dual-track theory bail prediction model (MT-DT) that integrates legal logic into deep learning for determining whether a defendant meets bail conditions.
Incorporating Visual Experts to Resolve the Information Loss in Multimodal Large Language Models
Xin He (Huawei Inc), Qi Tian (Huawei Inc)
Representation LearningTransformerLarge Language ModelMixture of ExpertsVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose the IVE framework to enhance the visual perception capability of multimodal large language models by integrating multiple visual experts (semantic, low-level, document-related encoders) and structural knowledge extraction.
Indirect Alignment and Relationship Preservation for Domain Generalization
Wei Wei (Shanxi University), Lin Li (Shanxi University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: A new domain generalization framework is studied, leveraging sample difference preservation and sample consistency alignment to enhance the model's generalization ability to unknown target domains.
Indirect Online Preference Optimization via Reinforcement Learning
En Wang (Jilin University), Wenbin Liu (Jilin University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningText
🎯 What it does: Propose an indirect online preference optimization algorithm (IOPO) based on adversarial reinforcement learning to improve preference alignment in large language models.
Inference of Human-derived Specifications of Object Placement via Demonstration
Alex Cuellar (Massachusetts Institute of Technology), Julie A Shah (Massachusetts Institute of Technology)
🎯 What it does: Proposed a position-enhanced RCC (PARCC) spatial specification language and presented an inference framework based on demonstration learning; validated through human-machine experiments that the framework can better capture human spatial arrangement preferences compared to manually provided specifications.
Inferring Causal Protein Signaling Networks with Reinforcement Learning via Artificial Bee Colony Neural Architecture Search
Jihao Zhai (Beijing University of Technology), Jinduo Liu (Beijing University of Technology)
Neural Architecture SearchRecurrent Neural NetworkTransformerReinforcement LearningBiomedical Data
🎯 What it does: Automatically optimize the hyperparameters of an Actor-Critic reinforcement learning model using artificial bee colony search, and use this model to infer causal protein signaling networks from single-cell protein expression data
INFP: INdustrial Video Anomaly Detection via Frequency Prioritization
Qianzi Yu (University of Science and Technology of China), Yu Kang (University of Science and Technology of China)
Anomaly DetectionConvolutional Neural NetworkVideo
🎯 What it does: Proposed an industrial video anomaly detection framework INFP based on frequency domain features, utilizing an encoder-decoder structure to predict future frames for anomaly detection.
InfVC: An Inference-Enhanced Local Search Algorithm for the Minimum Vertex Cover Problem in Massive Graphs
Rui Sun (Northeast Normal University), Jian Gao (Northeast Normal University)
OptimizationGraph
🎯 What it does: Propose a reasoning-enhanced local search algorithm called InfVC for solving the minimum vertex cover problem on large-scale graphs.
Initial Models and Serialisability in Abstract Dialectical Frameworks
Lars Bengel (University of Hagen), Matthias Thimm (University of Hagen)
Explainability and InterpretabilityComputational Efficiency
🎯 What it does: Proposed the initial model and serialization concepts for fine-grained explanation of conflicts and construction of acceptable models within the abstract dialectical framework (ADF).
Injecting Imbalance Sensitivity for Multi-Task Learning
Zhipeng Zhou (University of Science and Technology of China), Wei Gong (University of Science and Technology of China)
OptimizationImageTextBenchmark
🎯 What it does: Propose IMGrad, a gradient descent method that injects imbalance sensitivity in multi-task learning, capable of simultaneously avoiding Pareto failure and individual progress imbalance.
InnateCoder: Learning Programmatic Options with Foundation Models
Rubens O. Moraes (Universidade Federal de Viçosa), Levi H. S. Lelis (University of Alberta)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Propose the INNATECODER system, which generates procedural options from scratch before training using foundation models, providing reinforcement learning agents with 'innate skills'.
Instance Relation Learning Network with Label Knowledge Propagation for Few-shot Multi-label Intent Detection
Shiman Zhao (Peking University), Kam-Fai Wong (Peking University)
ClassificationGraph Neural NetworkTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper proposes an end-to-end few-shot multi-label intent detection method that directly infers multi-labels through instance relationships using an instance relationship learning network and label knowledge propagation, avoiding error propagation issues in traditional two-stage methods.
Instantiation-based Formalization of Logical Reasoning Tasks Using Language Models and Logical Solvers
Mohammad Raza (Qatar Computing Research Institute), Natasa Milic-Frayling (Qatar Computing Research Institute)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the Semantic Self-Verification (SSV) method, which combines large language models with logic solvers, using concrete instantiations to verify and correct abstract formalized logical reasoning tasks.
InstGAN: Instant Actor-Critic-Driven GAN for De Novo Molecule Generation and Property Optimization
Huidong Tang (Shandong Institute of Commerce and Technology), Yasuhiko Morimoto (Hiroshima University)
OptimizationDrug DiscoveryRecurrent Neural NetworkReinforcement LearningGenerative Adversarial NetworkBiomedical Data
🎯 What it does: This paper proposes InstGAN, a GAN based on actor-critic reinforcement learning, which generates multi-attribute optimized molecules at the SMILES level by utilizing immediate rewards and global rewards.
Instructing Text-to-Image Diffusion Models via Classifier-Guided Semantic Optimization
Yuanyuan Chang (Xi'an Jiaotong University), Guang Dai (State Grid Corporation of China)
GenerationData SynthesisOptimizationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Semantic embeddings obtained by optimizing the gradient of the attribute classifier guide text-to-image diffusion models to achieve high-quality, decoupled image editing.
INT: Instance-Specific Negative Mining for Task-Generic Promptable Segmentation
Jian Hu (Queen Mary University of London), Shaogang Gong (Queen Mary University of London)
SegmentationTransformerPrompt EngineeringVision Language ModelDiffusion modelImageBiomedical Data
🎯 What it does: Propose an unsupervised, training-free task-general prompt-able segmentation method INT, which generates instance-specific prompts through progressive negative sampling based on VLM output differences, achieving multi-sample segmentation under a single task-general prompt.
Integrating Answer Set Programming and Large Language Models for Enhanced Structured Representation of Complex Knowledge in Natural Language
Mario Alviano (University of Calabria), Luis Angel Rodriguez Reiners
Representation LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes integrating Answer Set Programming (ASP) with Large Language Models (LLM) by automatically generating prompts via YAML configuration files, enabling the extraction of structured facts from natural language. It leverages ASP for logical reasoning, ultimately achieving a structured representation of complex knowledge.
Integrating Independent Layer-Wise Rank Selection with Low-Rank SVD Training for Model Compression: A Theory-Driven Approach
Yifan Guo (Towson University), Alyssa Yu (Poolesville High School)
CompressionComputational EfficiencyRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This study integrates hierarchical independent rank selection into the low-rank SVD training process and provides adaptive rank search based on theoretical loss error bounds.
Integration of Old and New Knowledge for Generalized Intent Discovery: A Consistency-driven Prototype-Prompting Framework
Xiao Wei (Tianjin University), Jianwu Dang (Shenzhen Institute of Advanced Technology)
ClassificationDomain AdaptationTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: Propose a consistency-driven prototype-prompt framework (CPP) that integrates old and new knowledge in the general intent discovery task, enhancing the recognition of unknown intents;
Inter3D: A Benchmark and Strong Baseline for Human-Interactive 3D Object Reconstruction
Gan Chen (Shenzhen University), Guang Zhou (Guangdong Laboratory of Artificial Intelligence and Digital Economy)
GenerationData SynthesisNeural Radiance FieldGaussian SplattingMeshBenchmark
🎯 What it does: Propose the Inter3D benchmark and baseline methods to address 3D reconstruction and synthesis of unseen states for interactive objects in different states;
Interaction-Data-guided Conditional Instrumental Variables for Debiasing Recommender Systems
Zhirong Huang (Guangxi Normal University), Shichao Zhang (Guangxi Normal University)
Recommendation SystemGraph Neural NetworkAuto EncoderTabular
🎯 What it does: Proposes a debiasing method for recommendation systems, IDCIV-RS, which automatically generates conditional instrumental variables (CIVs) based on interaction data.
Interactive Multimodal Learning via Flat Gradient Modification
Qing-Yuan Jiang (Nanjing University of Science and Technology), Yang Yang (Nanjing University of Science and Technology)
OptimizationRepresentation LearningConvolutional Neural NetworkTransformerImageVideoTextMultimodalityAudio
🎯 What it does: Propose an interactive multimodal learning framework named IGM, which utilizes flat gradient modification strategies to transfer gradients between different modalities and combines Sharpness-Aware Minimization (SAM) to smooth the loss landscape, addressing modality imbalance and plasticity issues.
Interpretable DNFs
Martin C. Cooper (University of Toulouse), Clément Carbonnel (University of Montpellier)
ClassificationExplainability and InterpretabilityTabular
🎯 What it does: Propose an interpretability definition based on k-DNF, and study two types of interpretable classifiers: decision trees with depth k and the newly proposed nested k-DNF; provide theoretical properties, construction methods, and learning heuristic algorithms for nested k-DNF; conduct comparative experiments with decision trees of depth k on multiple UCI and Kaggle datasets.
Interval Selection with Binary Predictions
Christodoulos Karavasilis (University of Toronto)
OptimizationTabular
🎯 What it does: This paper studies the online interval selection problem, designing and analyzing algorithms under both irreversible and reversible decision scenarios, with experimental validation on real-world data.
Inverse Game Theory: An Incenter-Based Approach
Lvye Cui (Beijing Institute of Technology), Dario Paccagnan (Imperial College London)
Optimization
🎯 What it does: Propose a general inverse game framework that estimates players' utility parameters from observed Nash equilibria through an incenter selection method.
Iterated Belief Change as Learning
Nicolas Schwind (National Institute of Advanced Industrial Science and Technology), Pierre Marquis (Univ. Artois)
ClassificationTabular
🎯 What it does: This paper proposes migrating improvement operators from belief change theory to an online binary classification learning framework, and provides learning and inference algorithms that are polynomial-time solvable.
IterMeme: Expert-Guided Multimodal LLM for Interactive Meme Creation with Layout-Aware Generation
Yaqi Cai (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
GenerationLarge Language ModelSupervised Fine-TuningMixture of ExpertsDiffusion modelImageTextMultimodality
🎯 What it does: Designed the IterMeme framework, which utilizes a unified multimodal large language model (MLLM) to achieve end-to-end interactive meme generation, transforming user intent into text, layout, and images, while supporting personalization.
Joint-Perturbation Simultaneous Pseudo-Gradient
Carlos Martin (Carnegie Mellon University), Tuomas Sandholm (Carnegie Mellon University)
Optimization
🎯 What it does: The study investigates finding an approximate Nash equilibrium in black-box continuous action games where gradients are inaccessible, and proposes a novel joint perturbation pseudo-gradient (JPSPG) method.
K-Buffers: A Plug-in Method for Enhancing Neural Fields with Multiple Buffers
Haofan Ren (Hangzhou Dianzi University), Chenggang Yan (Hangzhou Dianzi University)
GenerationComputational EfficiencyConvolutional Neural NetworkNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: Proposed the K-Buffers scheme, combining multi-layer z-buffer with K-Feature Fusion Network (KFN), along with redundant query point pruning and feature correction, to improve the rendering quality of neural point fields and 3D Gaussian fields.
Keypoints as Dynamic Centroids for Unified Human Pose and Segmentation
Niaz Ahmad (Toronto Metropolitan University), Guanghui Wang (Toronto Metropolitan University)
SegmentationPose EstimationImage
🎯 What it does: Proposes a unified human pose estimation and instance segmentation framework called KDC based on dynamic centroids (KeyCentroid and MaskCentroid).
kgMBQA: Quality Knowledge Graph-driven Multimodal Blind Image Assessment
Wuyuan Xie (Shenzhen University), Miaohui Wang (Shenzhen University)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelImageMultimodalityGraph
🎯 What it does: Proposed a knowledge graph-based multimodal blind image quality assessment framework called kgMBQA, which can generate interpretable quality description texts and provide scores.
KIPPO: Koopman-Inspired Proximal Policy Optimization
Andrei Cozma (University of Tennessee), Hairong Qi (University of Tennessee)
OptimizationRepresentation LearningReinforcement LearningAuto EncoderSequential
🎯 What it does: Proposes KIPPO, which integrates a Koopman-inspired auxiliary linear latent space learning module into PPO to reduce gradient variance and enhance convergence stability in continuous control tasks.
Knowledge Editing for Multi-Hop Question Answering Using Semantic Analysis
Dominic Simon (University of Florida), Rickard Ewetz (University of Florida)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the CHECK framework, which performs semantic analysis and type checking on knowledge editing in multi-hop question answering to achieve correction and updating of the reasoning chain.
KnowMDD: Knowledge-guided Cross Contrastive Learning for Major Depressive Disorder Diagnosis
Anchen Lin (Hangzhou City University), Binbin Zhou (Hangzhou City University)
Representation LearningGraph Neural NetworkContrastive LearningGraphBiomedical Data
🎯 What it does: Proposes a knowledge-guided cross-contrastive learning framework called KnowMDD, which constructs multi-view brain networks using multi-template brain atlases, enhances feature learning by integrating the default mode network (DMN) subgraph with attention mechanisms, and obtains robust MDD diagnostic representations through cross-contrastive learning.
KnowRA: Knowledge Retrieval Augmented Method for Document-level Relation Extraction with Comprehensive Reasoning Abilities
Chengcheng Mai (Nanjing Normal University), Yihua Huang (Nanjing University)
RetrievalRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextGraphRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Designed and implemented the KnowRA model for document-level relation extraction, integrating multi-layer heterogeneous graph encoding, coreference resolution, external knowledge retrieval and filtering, and axial attention mechanisms to achieve comprehensive reasoning capabilities.
KP-PINNs: Kernel Packet Accelerated Physics Informed Neural Networks
Siyuan Yang (Hong Kong University of Science and Technology (Guangzhou)), Wenjia Wang (Hong Kong University of Science and Technology (Guangzhou))
Computational EfficiencyPhysics Related
🎯 What it does: This paper proposes a novel physics-informed neural network framework called KP-PINNs, which solves various partial differential equations in both forward and inverse problems by using the RKHS norm instead of the traditional L2 error and accelerating the kernel matrix inversion using the Kernel Packet (KP) method.
L2M2: A Hierarchical Framework Integrating Large Language Model and Multi-agent Reinforcement Learning
Minghong Geng (Singapore Management University), Ah-Hwee Tan (Singapore Management University)
TransformerLarge Language ModelReinforcement LearningAgentic AIBenchmark
🎯 What it does: Proposed the L2M2 framework, which uses large language models for high-level planning while reinforcement learning agents handle low-level execution, supporting zero-shot planning and integration with pre-trained strategies;
Label Distribution Learning with Biased Annotations Assisted by Multi-Label Learning
Zhiqiang Kou (Southeast University), Xin Geng (Southeast University)
ClassificationImageTextBiomedical Data
🎯 What it does: This paper proposes a novel Biased Label Distribution Learning (BLDL) framework, which first converts biased soft label distributions into multi-label representations, and then recovers the true label distribution and trains the model by leveraging the low-rank structure in the multi-label space.
Language-Based Bayesian Optimization Research Assistant (BORA)
Abdoulatif Cissé (University of Liverpool), Andrew I. Cooper (University of Liverpool)
OptimizationTransformerLarge Language ModelPrompt EngineeringAgriculture RelatedPhysics Related
🎯 What it does: Developed a hybrid Bayesian optimization framework named BORA, integrating large language models (LLMs) with traditional Bayesian optimization (BO) to dynamically adjust LLM interventions and enhance the efficiency of experimental design search.
Language-Conditioned Open-Vocabulary Mobile Manipulation with Pretrained Models
Shen Tan (Harbin Institute of Technology), Guanghui Sun (Harbin Institute of Technology)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkLarge Language ModelVision Language ModelVision-Language-Action ModelImageTextMultimodality
🎯 What it does: This paper proposes the LOVMM framework, which utilizes LLM and VLM to achieve open-vocabulary mobile manipulation driven by free-text instructions, covering navigation and 6-DoF grasping and placement.
Language-Guided Hybrid Representation Learning for Visual Grounding on Remote Sensing Images
Biao Liu (Xidian University), Youlin Huang (East China Jiaotong University)
Representation LearningConvolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Proposes a remote sensing image visual localization framework named LGFormer, which efficiently fuses multimodal features in the decoding stage using language-guided hybrid queries to achieve precise localization.
Large-Scale Trade-Off Curve Computation for Incentive Allocation with Cardinality and Matroid Constraints
Yu Cong (University of Electronic Science and Technology of China), Yi Zhou (University of Electronic Science and Technology of China)
Optimization
🎯 What it does: Studies the large-scale incentive allocation problem, derives the overall trade-off curve between budget and profit, and proposes a dynamic algorithm that maintains this curve under cardinality constraints and matrix constraints.
Latte: Transfering LLMs' Latent-level Knowledge for Few-shot Tabular Learning
Ruxue Shi (Jilin University), Xin Wang (Jilin University)
ClassificationKnowledge DistillationRepresentation LearningData-Centric LearningMeta LearningTransformerLarge Language ModelTabular
🎯 What it does: Propose the Latte framework, which extracts latent knowledge through LLM during the training phase to guide few-shot table learning.
Lazy Testing of Machine-Learning Models
Anastasia Isychev (TU Wien), Maria Christakis (TU Wien)
Computational EfficiencyImageTextTabularAudio
🎯 What it does: Implemented a lazy testing framework called LAZ, which dynamically skips redundant model calls and reorders the calling sequence when performing hyperproperty testing on machine learning models, thereby improving test throughput.
Learn from Global Rather Than Local: Consistent Context-Aware Representation Learning for Multi-View Graph Clustering
Lele Fu (Sun Yat-sen University), Chuan Chen (Sun Yat-sen University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a global consistent context-aware representation learning framework (CCARL) based on global anchors and fused Gromov-Wasserstein optimal transport for multi-view graph clustering;
Learn Multi-task Anchor: Joint View Imputation and Label Generation for Incomplete Multi-view Clustering
Xinxin Wang (University of Macau), Yicong Zhou (University of Macau)
OptimizationRepresentation LearningData-Centric LearningMultimodality
🎯 What it does: Propose a Joint View Imputation and Label Generation (JVILG) method, which utilizes fine-grained anchors to simultaneously recover missing views and directly generate soft labels, thereby achieving incomplete multi-view clustering.
Learn to Think: Bootstrapping LLM Logic Through Graph Representation Learning
Hang Gao (Institute Of Software Chinese Academy Of Sciences), Huaping Liu (Tsinghua University)
OptimizationRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextGraphChain-of-Thought
🎯 What it does: Propose the L2T framework, modeling the LLM reasoning process as a graph, and jointly learning reasoning strategies through LLM and GNN to accomplish multi-task reasoning without task-specific prompts.
Learnable Frequency Decomposition for Image Forgery Detection and Localization
Dong Li (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an image forgery detection and localization framework based on the frequency domain, F2D-Net, which utilizes depth Fourier transform to decompose images into phase and amplitude components, and further explores phase information through phase emphasis interaction blocks to achieve precise identification and localization of forgery traces.
Learning Advanced Self-Attention for Linear Transformers in the Singular Value Domain
Hyowon Wi (Korea Advanced Institute of Science and Technology), Noseong Park (Korea Advanced Institute of Science and Technology)
ClassificationComputational EfficiencyTransformerImageTime SeriesSequentialBenchmark
🎯 What it does: This paper proposes a new linear Transformer self-attention mechanism called Attention Graph Filter (AGF), which enhances the expressiveness of self-attention by directly learning graph filters in the singular value domain.
Learning Causally Disentangled Representations for Fair Personality Detection
Yangfu Zhu (Capital Normal University), Bin Wu (Beijing University of Posts and Telecommunications)
ClassificationRepresentation LearningTransformerText
🎯 What it does: This paper proposes an IPDN model, a personalized personality detection model based on causal intervention, which can automatically learn and eliminate individual biases in user-generated text without using prior attributes.
Learning from Logical Constraints with Lower- and Upper-Bound Arithmetic Circuits
Lucile Dierckx (UCLouvain), Siegfried Nijssen (UCLouvain)
OptimizationTabular
🎯 What it does: Propose a new neuro-symbolic learning framework called LUBAC, which approximates gradients by compiling lower and upper bound arithmetic circuits under CNF constraints that cannot be fully compiled.
Learning Heterogeneous Performance-Fairness Trade-offs in Federated Learning
Rongguang Ye (Southern University of Science and Technology), Ming Tang (Southern University of Science and Technology)
OptimizationFederated LearningTabularBenchmark
🎯 What it does: Proposes the HetPFL framework, which simultaneously learns the local Pareto frontier and global Pareto frontier for each client in federated learning.
Learning Neural Jump Stochastic Differential Equations with Latent Graph for Multivariate Temporal Point Processes
Yuchen Wang (Northwestern Polytechnical University), Xianghua Li (Northwestern Polytechnical University)
Graph Neural NetworkTime SeriesStochastic Differential Equation
🎯 What it does: Propose a model based on multi-dimensional neural jump stochastic differential equations (SDE) and an encoder-free latent graph structure for modeling multivariate point processes;
Learning Neural Vocoder from Range-Null Space Decomposition
Andong Li (Institute of Acoustics, Chinese Academy of Sciences), Chengshi Zheng (Institute of Acoustics, Chinese Academy of Sciences)
GenerationData SynthesisComputational EfficiencyConvolutional Neural NetworkGenerative Adversarial NetworkAudio
🎯 What it does: Propose a frequency-domain based neural vocoder RNDVoC, achieving linear inversion of Mel spectra and residual detail reconstruction through range-nullspace decomposition;
Learning Optimal Oblique Decision Trees with (Max)SAT
Florent Avellaneda (Universit' e du Qu'ec ' Montr' eal)
ClassificationTabular
🎯 What it does: Learning the Optimal Oblique Decision Tree
Learning Probabilistic Temporal Logic Specifications for Stochastic Systems
Rajarshi Roy (University of Oxford), Marta Kwiatkowska (University of Oxford)
OptimizationExplainability and InterpretabilityReinforcement LearningSequential
🎯 What it does: This paper proposes a passive learning algorithm for learning concise probabilistic linear temporal logic (PLTL) specifications from positive and negative sample Markov chains, capable of distinguishing systems with random behavior.
Learning Real Facial Concepts for Independent Deepfake Detection
Ming-Hui Liu (Shandong University), Xin-Shun Xu (Shandong University)
Anomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed and implemented the RealID framework, integrating the Real Concept Capture module and the Independent Dual-Decision classifier, achieving deepfake detection based on dual decisions from real human face concepts and forged traces.
Learning Robust Multi-view Representation Using Dual-masked VAEs
Jiedong Wang (Sichuan University), Hao Wang (Sichuan University)
Representation LearningData-Centric LearningMixture of ExpertsAuto EncoderImage
🎯 What it does: This paper proposes a multi-view autoencoder called DualVAE, which can learn robust shared and view-specific representations in a general environment where view missingness and sample noise coexist.
Learning to Explain: Towards Human-Aligned Explainability in Deep Reinforcement Learning via Attention Guidance
Bokai Ji (Xidian University), Gang Xiao (National Key Laboratory for Complex Systems Simulation)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerReinforcement LearningImageBenchmark
🎯 What it does: Developed an attention-guided interpretable deep reinforcement learning framework, Concept-PPO, which can generate concept-level explanations highly consistent with human cognition during the decision-making process.
Learning to Extrapolate and Adjust: Two-Stage Meta-Learning for Concept Drift in Online Time Series Forecasting
Weiqi Chen (Damo Academy, Alibaba Group), Liang Sun (Damo Academy, Alibaba Group)
OptimizationMeta LearningRecurrent Neural NetworkTransformerTime Series
🎯 What it does: Proposes the LEAF (Learning to Extrapolate and Adjust) framework, which employs a two-stage meta-learning approach to address concept drift in online time series prediction.
LEKA: LLM-Enhanced Knowledge Augmentation
Xinhao Zhang (Portland State University), Kunpeng Liu (Portland State University)
Domain AdaptationData-Centric LearningTransformerLarge Language ModelTabularRetrieval-Augmented Generation
🎯 What it does: Designed and implemented the LEKA framework, which leverages LLMs to automatically retrieve and fuse external datasets for knowledge enhancement and transfer learning.
LensNet: An End-to-End Learning Framework for Empirical Point Spread Function Modeling and Lensless Imaging Reconstruction
Jiesong Bai (University of Macau), Xuhang Chen (University of Macau)
RestorationSuper ResolutionConvolutional Neural NetworkImagePhysics Related
🎯 What it does: This paper proposes LensNet, an end-to-end deep learning framework for point spread function (PSF) modeling and image reconstruction in optical encoded lensless imaging systems.
Let’s Group: A Plug-and-Play SubGraph Learning Method for Memory-Efficient Spatio-Temporal Graph Modeling
Wenchao Weng (Zhejiang University of Technology), Xiangjie Kong (Zhejiang University of Technology)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTime Series
🎯 What it does: Perform subgraph learning on large-scale spatiotemporal graph models, providing pluggable subgraph partitioning and aggregation modules to significantly reduce GPU memory consumption.
Leveraging MLLM Embeddings and Attribute Smoothing for Compositional Zero-Shot Learning
Xudong Yan (Beijing Jiaotong University), Haojun Fei (Qifu Technology)
ClassificationTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: This paper proposes the TRIDENT framework, which enhances combination zero-shot learning (CZSL) performance using multimodal LLM embeddings, attribute smoothing, and visual feature decoupling.
Leveraging Peer-Informed Label Consistency for Robust Graph Neural Networks with Noisy Labels
Kailai Li (Shanghai Jiao Tong University), Jie Li (Shanghai Jiao Tong University)
Anomaly DetectionRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose the ProCon framework, which identifies mislabeled nodes by leveraging label consistency among similar nodes and enhances the robustness of representations through self-supervised learning;
Leveraging Personalized PageRank and Higher-Order Topological Structures for Heterophily Mitigation in Graph Neural Networks
Yumeng Wang (Tianjin University), Minglai Shao (Tianjin University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: In this paper, the authors propose HPGNN, a graph neural network that integrates high-order personalized PageRank (HiPPR) with Simplicial Complex (SC) structure, for solving node classification problems in heterogeneous graphs;
Leveraging Pretrained Diffusion Models for Zero-Shot Part Assembly
Ruiyuan Zhang (Zhejiang University), Chao Wu (Zhejiang University)
Pose EstimationDiffusion modelScore-based ModelPoint Cloud
🎯 What it does: Proposed a zero-shot 3D part assembly method that automatically predicts part poses and completes assembly by leveraging a pre-trained point cloud diffusion model and the ICP iterative process.
LiBOG: Lifelong Learning for Black-Box Optimizer Generation
Jiyuan Pei (Victoria University of Wellington), Mengjie Zhang (Victoria University of Wellington)
OptimizationMeta LearningRecurrent Neural NetworkReinforcement LearningBenchmark
🎯 What it does: Developed LiBOG, a framework applying lifelong learning in Meta-Black-Box Optimization (MetaBBO), which can continuously learn and generate high-performance black-box optimizers across sequentially emerging distributions of optimization problems.
Linear Trading Position with Sparse Spectrum
Zhao-Rong Lai (Jinan University), Haisheng Yang (Sun Yat-Sen University)
OptimizationTabularFinance Related
🎯 What it does: Proposes the Linear Trading Position Sparse Spectrum (LTPSS) method, which constructs trading strategies using all principal components with adjustable spectral energy, and solves the problem via the Krasnosel'skiǐ-Mann fixed-point algorithm.
ListenNet: A Lightweight Spatio-Temporal Enhancement Nested Network for Auditory Attention Detection
Cunhang Fan (Anhui University), Zhao Lv (Anhui University)
ClassificationComputational EfficiencyConvolutional Neural NetworkTime SeriesBiomedical Data
🎯 What it does: Developed a lightweight spatiotemporal enhanced nested network called ListenNet for auditory attention detection using electroencephalogram (EEG) signals in multi-speaker environments.
LLM-enhanced Score Function Evolution for Causal Structure Learning
Zidong Wang (City University of Hong Kong), Xiaoguang Gao (Northwestern Polytechnical University)
OptimizationTransformerLarge Language ModelPrompt EngineeringGraphTabular
🎯 What it does: This study proposes the L-SFE framework, which leverages large language models and evolutionary algorithms to automatically explore and generate improved score functions, thereby enhancing the performance of causal structure learning.
LLM-TPF: Multiscale Temporal Periodicity-Semantic Fusion LLMs for Time Series Forecasting
Qihong Pan (Zhejiang University of Technology), Chenyang Xu (Zhejiang University of Technology)
Convolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTime Series
🎯 What it does: Proposed a time series forecasting framework based on large language models, named LLM-TPF, which integrates three modules: frequency domain periodic features, time domain prompt information, and cross-modal fusion.
LLM4VKG: Leveraging Large Language Models for Virtual Knowledge Graph Construction
Guohui Xiao (Southeast University), Davide Lanti (Free University of Bozen-Bolzano)
TransformerLarge Language ModelTextGraphRetrieval-Augmented Generation
🎯 What it does: This paper proposes the LLM4VKG framework, which utilizes large language models (LLMs) to automatically complete the construction process of virtual knowledge graphs (VKGs), including ontology completion and mapping generation.
Localizing Before Answering: A Benchmark for Grounded Medical Visual Question Answering
Dung Nguyen (Hanoi University of Science and Technology), Vu Minh Hieu Phan (Australian Institute for Machine Learning, University of Adelaide)
SegmentationLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningMultimodalityBiomedical DataBenchmark
🎯 What it does: This paper designs and constructs the HEAL-MedVQA benchmark to evaluate the localization and hallucination capabilities of medical multimodal LLMs, and proposes the LobA framework, which performs image localization before answering, then enhances answer reliability through self-prompting and attention reweighting.
LoD: Loss-difference OOD Detection by Intentionally Label-Noisifying Unlabeled Wild Data
Chuanxing Geng (Nanjing University of Aeronautics and Astronautics), Pong C. Yuen (Hong Kong Baptist University)
Anomaly DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Leverages the uniform annotation of unlabeled wild data into K+1 classes, and observes loss differences in the K+1 classification task to achieve automatic filtering and detection of OOD samples.
Logarithmic Approximations for Fair k-Set Selection
Shi Li (Nanjing University), Ruilong Zhang (Technical University of Munich)
Optimization
🎯 What it does: This paper studies the fair k-set selection problem, which involves selecting k sets from a given set system such that the (weighted) number of occurrences of each element in the selected sets is as balanced as possible; by converting the set system into a bipartite graph, the problem is equivalent to selecting k right vertices such that the maximum number of adjacent left vertices is minimized.
Logic Distillation: Learning from Code Function by Function for Decision-making Tasks
Dong Chen (Zhengzhou University), Mingliang Xu (Zhengzhou University)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a Logic Distillation (Logic Distillation, LD) framework that decomposes the decision-making logic of large language models (L-LLM) into callable functions. It fine-tunes small language models (S-LLM) using a function library, enabling them to invoke relevant functions in stages to complete sequential decision-making tasks.
LogiCase: Effective Test Case Generation from Logical Description in Competitive Programming
Sicheol Sung (Yonsei University), Sang-Ki Ko (University of Seoul)
GenerationAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes Context-Free Grammars with Counters (CCFG) method for automatically generating high-quality test cases that comply with semantics and syntax from the natural language descriptions of competition problems.
Long-Term Individual Causal Effect Estimation via Identifiable Latent Representation Learning
Ruichu Cai (Guangdong University of Technology), Jiecheng Guo (DiDi China Ride Hailing Business Group)
Representation LearningAuto EncoderTabular
🎯 What it does: Proposes a framework for estimating long-term individual causal effects by leveraging multi-source data to identify potential confounding variables, combining short-term potential outcome learning with an identifiable variational autoencoder to achieve long-term effect prediction.
Low-Light Video Enhancement via Spatial-Temporal Consistent Decomposition
Xiaogang Xu (Chinese University of Hong Kong), Bei Yu (Chinese University of Hong Kong)
RestorationConvolutional Neural NetworkAuto EncoderVideo
🎯 What it does: Proposed a low-light video enhancement framework based on perspective-independent/perspective-dependent decomposition, leveraging cross-frame correspondence and spatial-temporal continuity constraints to achieve unified image decomposition and enhancement.
LP-Based Weighted Model Integration over Non-Linear Real Arithmetic
S. Akshay (Indian Institute of Technology Bombay), Đorđe Žikelić (Singapore Management University)
OptimizationComputational EfficiencyBenchmark
🎯 What it does: This paper proposes a weighted model integration (WMI) approximation method based on linear programming, capable of handling rational and square root functions in nonlinear real arithmetic.
LPDetective: Dusting the LLM Chats for Prompt Template Abusers
Yang Luo (Peking University), Zhonghai Wu (Peking University)
Anomaly DetectionLarge Language ModelText
🎯 What it does: This paper proposes an unsupervised method called LPDetective, which detects robot abuse in LLM chatbots by mining robot prompt templates from chat logs.
LRGR: Self-Supervised Incomplete Multi-View Clustering via Local Refinement and Global Realignment
Yanwanyu Xi (China University of Geosciences), Xinwang Liu (National University of Defense Technology)
Representation LearningGraph Neural NetworkContrastive LearningImageTabular
🎯 What it does: Propose a self-supervised local refinement and global realignment (LRGR) model to address the distribution shift and homogeneous representation issues caused by missing samples in incomplete multi-view clustering.
LTLf+ and PPLTL+: Extending LTLf and PPLTL to Infinite Traces
Benjamin Aminof (University of Rome La Sapienza), Moshe Y. Vardi (Rice University)
🎯 What it does: This paper proposes two novel temporal logics, LTLf+ and PPLTL+, for describing properties of infinite traces, and provides complete synthesis, satisfiability, and model checking algorithms based on DFA.
M^2LLM: Multi-view Molecular Representation Learning with Large Language Models
Jiaxin Ju (Griffith University), Shirui Pan (Griffith University)
Representation LearningDrug DiscoveryTransformerLarge Language ModelMixture of ExpertsGraphTabularBiomedical Data
🎯 What it does: This study proposes the M2LLM multi-view framework, which utilizes large language models to generate three views (molecular structure, tasks, and rules), and fuses them for molecular property prediction.
M3ANet: Multi-scale and Multi-Modal Alignment Network for Brain-Assisted Target Speaker Extraction
Cunhang Fan (Anhui University), Zhao Lv (Anhui University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningMultimodalityBiomedical DataAudio
🎯 What it does: This paper proposes a multi-scale multi-modal alignment network, M3ANet, for brain-assisted target speaker extraction, addressing the time deviation between EEG and speech and the insufficiency in speech feature extraction.
M4Bench: A Benchmark of Multi-domain Multi-granularity Multi-image Understanding for Multi-modal Large Language Models
Xiaojun Ye (Zhejiang University), Sheng Zhou (Zhejiang University)
TransformerVision Language ModelImageVideoMultimodalityBenchmark
🎯 What it does: Proposed and implemented the M4 Bench benchmark to evaluate the performance of multimodal large language models in multi-domain, multi-granularity multi-image comparison tasks.
MA-RAG: Automating Role Engineering for RESTful APIs with Multi-Head Attention and Retrieval-Augmented Generation
Yang Luo (Peking University), Zhonghai Wu (Peking University)
TransformerContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: This study proposes MA-RAG, a fully automated role engineering method based on multi-head attention and retrieval-augmented generation, specifically designed for RESTful systems, which achieves permission role allocation through fine-grained control flow analysis and code semantic representation.
MAGE: Multimodal Alignment and Generation Enhancement via Bridging Visual and Semantic Spaces
Shaojun E (Global Tone Communication Technology Co Ltd), Ziyan Chen (Global Tone Communication Technology Co Ltd)
GenerationTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose the MAGE model, which achieves dual alignment of visual and textual features in both dimensions and semantics through the Intelligent Alignment Network (IAN), and enhances cross-modal consistency via dual loss functions (image generation loss + image-text distance minimization loss); simultaneously construct the HMDSet multimodal tool calling dataset to expand the model's 'Any-to-Any' output capability.