AAAI 2025 Papers — Page 15
AAAI Conference on Artificial Intelligence · 3028 papers
Human and AI Perceptual Differences in Image Classification Errors
Minghao Liu (University of California), James Davis (University of California)
ClassificationRecognitionImage
🎯 What it does: This paper studies the perceptual differences in error distribution between humans and machines in image classification tasks, and utilizes this difference to design and evaluate human-machine collaboration strategies.
HVAdam: A Full-Dimension Adaptive Optimizer
Yiheng Zhang (Wuhan University), Xiaoguang Niu (Wuhan University)
GenerationOptimizationConvolutional Neural NetworkRecurrent Neural NetworkTransformerGenerative Adversarial NetworkImageText
🎯 What it does: A new adaptive optimizer HVAdam is proposed, which constructs hidden vectors using full-dimensional parameter information to address the slow convergence and instability issues of traditional optimizers in the 'valley' problem.
HVDualformer: Histogram-Vision Dual Transformer for White Balance
Yan-Tsung Peng (National Chengchi University), Guan-Rong Chen (National Chengchi University)
RestorationTransformerImage
🎯 What it does: This paper proposes a dual transformer model called HVDualformer, based on histograms and Vision Transformers, for end-to-end image white balance correction.
HVIS: A Human-like Vision and Inference System for Human Motion Prediction
Kedi Lyu (Jilin University), Yingying Jiao (Institute for Infocomm Research)
GenerationPose EstimationRecurrent Neural NetworkGraph Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a Human Visual and Inference System (HVIS) that achieves high-quality human motion prediction by simulating human retinal, visual cortex, and brain learning patterns.
HYBOOD: A Hybrid Generative Model for Out-of-Distribution Detection with Corruption Estimation
Giwoong Lee (IOPS), Jeongyeol Choe (IOPS)
GenerationAnomaly DetectionFlow-based ModelImage
🎯 What it does: Designed the HYBOOD hybrid generative model, which combines regularized flow and global average pooling classifiers to estimate covariate shift difficulty and achieve OOD detection within a single model.
Hybrid Data-Free Knowledge Distillation
Jialiang Tang (Nanjing University of Science and Technology), Chen Gong (Shanghai Jiao Tong University)
ClassificationGenerationData SynthesisKnowledge DistillationGenerative Adversarial NetworkImageBiomedical Data
🎯 What it does: A hybrid data-free knowledge distillation framework HiDFD is proposed, which trains a GAN to generate high-quality synthetic data using a very small number of real samples in conjunction with a teacher network, and then trains the student network using a mix of synthetic and real samples, referred to as 'mixed data'.
Hybrid Decentralized Optimization: Leveraging Both First- and Zeroth-Order Optimizers for Faster Convergence
Shayan Talaei (Stanford University), Dan Alistarh (Institute of Science and Technology Austria)
OptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A hybrid decentralized optimization framework is proposed, allowing for collaborative learning on distributed nodes using both gradient (first-order) and non-gradient (zero-order) estimators simultaneously.
Hybrid Reasoning About Relative Position and Orientation of Objects and Navigating Agents Using Answer Set Programming
Yusuf Izmirlioglu (University of Roehampton)
🎯 What it does: A new hybrid spatial reasoning operator HOPA is proposed, which unifies qualitative direction, distance, and quantitative constraints into OPRA, and implements consistency checking, reasoning, conflict explanation, and unknown information inference using Answer Set Programming (ASP), supporting uncertain and hypothetical information.
Hybrid-Driving: An Autonomous Driving Decision Framework Integrating Large Language Models, Knowledge Graphs and Driving Rules
Jiabao Wang (Institute of Software Chinese Academy of Sciences), Yukuan Yang (Institute of Software Chinese Academy of Sciences)
Autonomous DrivingTransformerLarge Language ModelPrompt EngineeringGraph
🎯 What it does: A Hybrid-Driving framework is proposed, combining LLM, Scenario Evolution Knowledge Graph (SEKG), and driving rules to achieve safe and reliable autonomous driving decisions.
HybridReg: Robust 3D Point Cloud Registration with Hybrid Motions
Keyu Du (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
RecognitionPose EstimationPoint Cloud
🎯 What it does: The HybridReg method is proposed, which combines uncertainty mask learning to achieve robust 3D point cloud registration for mixed motion (rigid background and non-rigid foreground).
HYGENE: A Diffusion-Based Hypergraph Generation Method
Dorian Gailhard (Telecom Paris), Jhony H. Giraldo (Telecom Paris)
GenerationData SynthesisGraph Neural NetworkDiffusion modelMeshGraph
🎯 What it does: We propose and implement HYGENE, a diffusion-based hypergraph generation model that can sample realistic hypergraphs from a given data distribution.
Hyperbolic-Constraint Point Cloud Reconstruction from Single RGB-D Images
Wenrui Li (Harbin Institute of Technology), Xiaopeng Fan (Harbin Institute of Technology)
RestorationDepth EstimationTransformerPoint Cloud
🎯 What it does: This paper introduces hyperspherical geometry into single-view RGB-D image point cloud reconstruction, utilizing Poincaré sphere mapping, hyperspherical Chamfer distance, and regularized triplet loss to achieve hierarchical embedding of local and global point clouds.
HyperDefender: A Robust Framework for Hyperbolic GNNs
Nikita Malik (Indian Institute of Technology), Sandeep Kumar (Indian Institute of Technology)
ClassificationAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper proposes and implements HyperDefender, a robust defense framework for hyperbolic graph neural networks (Hy-GNN) to resist adversarial attacks and noise on node features and graph structures, and validates its effectiveness in node classification tasks.
Hypergraph Attacks via Injecting Homogeneous Nodes into Elite Hyperedges
Meixia He (Northwestern Polytechnical University), Yangming Guo (Northwestern Polytechnical University)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: A node injection attack framework for hypergraph neural networks, called IE-Attack, is proposed, which utilizes the node crossing phenomenon and hyperedge group identity to generate homogeneous nodes and inject them into elite hyperedges to disrupt model predictions.
Hypergraph Learning for Unsupervised Graph Alignment via Optimal Transport
Yuguang Yan (Guangdong University of Technology), Michael Ng (Hong Kong Baptist University)
OptimizationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes an unsupervised graph alignment method that utilizes optimal transport to learn hypergraph structures to capture higher-order node relationships, and extracts structural and feature information on both the graph layer and hypergraph layer.
HyperMixer: Specializable Hypergraph Channel Mixing for Long-term Multivariate Time Series Forecasting
Changyuan Tian (Aerospace Information Research Institute Chinese Academy of Sciences), Li Jin (Aerospace Information Research Institute Chinese Academy of Sciences)
Graph Neural NetworkMixture of ExpertsTime Series
🎯 What it does: We propose HyperMixer, a pluggable hypergraph channel mixing plugin for Long Multi-Variable Time Series (LMTS) forecasting, capable of capturing high-order interactions among channels and time-varying correlation patterns.
Hyperparametric Robust and Dynamic Influence Maximization
Arkaprava Saha (DesCartes Program), Laks V. S. Lakshmanan (National University of Singapore)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes the RIME algorithm, which addresses the robust influence maximization problem based on hyperparameters in dynamic networks with node/edge additions and deletions, and can return a near-optimal seed set at each time step.
IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities
Bin Wang (360 AI Research), Yuhui Yin (360 AI Research)
OptimizationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes an Inner-Adaptor Architecture (IAA) that extends multimodal capabilities by inserting trainable adapter layers within a frozen large language model, while preserving the original NLP performance of the language model.
ICE-T: Interactions-aware Cross-column Contrastive Embedding for Heterogeneous Tabular Datasets
Tomas Tokar (Vector Institute for AI), Scott Sanner (University of Toronto)
Representation LearningData-Centric LearningContrastive LearningMultimodalityTabular
🎯 What it does: A contrastive learning framework named ICE-T is proposed, treating each column as a separate modality, using inter-column average aggregation as anchors to obtain column-specific and global embeddings through linear time contrastive learning.
ICM-Assistant: Instruction-tuning Multimodal Large Language Models for Rule-based Explainable Image Content Moderation
Mengyang Wu (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringImageTextMultimodalityChain-of-Thought
🎯 What it does: Designed and implemented a rule-based multimodal large language model ICM-Assistant for image content moderation, and constructed the ICM-Instruct dataset that can be generated according to rules;
ID-GMLM: Intelligent Decision-Making with Integrated Graph Models and Large Language Models
Zhenhua Meng (Beijing University of Posts and Telecommunications), Budan Wu (Beijing University of Posts and Telecommunications)
Recommendation SystemOptimizationGraph Neural NetworkLarge Language ModelTabular
🎯 What it does: A multi-task learning framework ID-GMLM that combines graph models with large language models is proposed for intelligent decision support in multi-criteria decision-making and preference learning.
ID-Sculpt: ID-aware 3D Head Generation from Single In-the-wild Portrait Image
Jinkun Hao (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: The ID-Sculpt method is proposed, which can generate high-quality and identity-consistent 3D head models from a single outdoor portrait image.
Identifying Macro Conditional Independencies and Macro Total Effects in Summary Causal Graphs with Latent Confounding
Simon Ferreira (Sorbonne Université), Charles K. Assaad (Sorbonne Université)
Time Series
🎯 What it does: This study investigates the identification issues of macro conditional independence and macro total effects in summary causal graphs (SCG) that include potential confounding.
Identifying Predictions That Influence the Future: Detecting Performative Concept Drift in Data Streams
Brandon Gower-Winter (Utrecht University), Arno Siebes (Utrecht University)
Anomaly DetectionTabularFinance Related
🎯 What it does: This paper studies and proposes an algorithm called CB-PDD for detecting performative drift caused by model predictions in data streams, and validates its effectiveness through synthetic and semi-synthetic data.
Identifying Query-Relevant Neurons in Large Language Models for Long-Form Texts
Lihu Chen (Imperial College), Francesca Toni (Imperial College)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes an architecture-agnostic Query-Relevant Neuron Clustering Attribution (QRNCA) framework for locating query-relevant neurons in large decoder language models, supporting long text answers.
Identity-Text Video Corpus Grounding
Bin Huang (Tsinghua University), Wenwu Zhu (Tsinghua University)
RetrievalTransformerLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes the Identity-Text Video Semantic Retrieval task (ITVCG), constructs the TVR-IT dataset, and designs the Video-Locator model to achieve joint understanding and retrieval of videos, text, and identity images.
IDseq: Decoupled and Sequentially Detecting and Grounding Multi-Modal Media Manipulation
Runxin Liu (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
Object DetectionAnomaly DetectionTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an image-driven decoupled sequence framework (IDseq) for detecting and locating multimodal manipulations in image-text pairs;
IMAGDressing-v1: Customizable Virtual Dressing
Fei Shen (Nanjing University of Science and Technology), Jinhui Tang (Nanjing University)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: The Virtual Dressing (VD) task is proposed, and the IMAGDressing-v1 model is developed to generate customizable portraits with editable clothing images; simultaneously, the IGPair large-scale clothing-outfit pair dataset is constructed, and a comprehensive affinity metric CAMI is designed.
Image Conductor: Precision Control for Interactive Video Synthesis
Yaowei Li (Peking University), Yuexian Zou (Tencent)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Proposes the Image Conductor method, which generates interactive controllable camera motion and object motion videos from a single image;
Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models
Chutian Meng (Zhejiang University), Yueting Zhuang (Zhejiang University)
GenerationData SynthesisLarge Language ModelPrompt EngineeringDiffusion modelImageMultimodalityBenchmark
🎯 What it does: The ImageRepainter framework is proposed, which evaluates the generation quality of text-to-image models through the image regeneration task (generating an image with the same content given a reference image) and utilizes a multimodal LLM for image understanding and iterative prompt generation and revision, ultimately assessing similarity through image-to-image comparison.
Image-to-video Adaptation with Outlier Modeling and Robust Self-learning
Junbao Zhuo (University of Science and Technology Beijing), Huimin Ma (Microsoft)
Domain AdaptationConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningImageVideo
🎯 What it does: In unsupervised domain adaptation from images to videos, the authors propose a two-stage method: the first stage models outlier frames by introducing an 'outlier' category in the classifier and employing batch nuclear norm maximization along with pseudo outlier class loss; the second stage selects high-quality pseudo labels using a consistency measure based on label propagation and employs FixMatch for semi-supervised video-level learning to alleviate modality gaps.
ImagePiece: Content-aware Re-tokenization for Efficient Image Recognition
Seungdong Yoa (LG AI Research), Woohyung Lim (LG AI Research)
RecognitionComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes ImagePiece, a content-aware re-tokenization strategy for Vision Transformers that merges low-semantic tokens into semantic units through local consistency bias, combined with token pruning/merging to enhance inference speed and accuracy.
Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph
Weihao Wang (Tongji University), Bin He (Tongji University)
Object DetectionGenerationPose EstimationGraph Neural NetworkImagePoint Cloud
🎯 What it does: The paper proposes a framework named Imagine for 3D part assembly based on a single image, utilizing a structural knowledge graph to guide the assembly of parts from coarse to fine.
Imitate Before Detect: Aligning Machine Stylistic Preference for Machine-Revised Text Detection
Jiaqi Chen (Fudan University), Lefei Zhang (Wuhan University)
ClassificationOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a 'Imitate-Then-Detect' (ImBD) framework, which first simulates the writing style of LLM through style preference optimization, and then uses Style Conditional Probability Curvature (Style-CPC) to distinguish between machine-rewritten text and purely human text.
iMoT: Inertial Motion Transformer for Inertial Navigation
Son Minh Nguyen (University of Twente), Paul Havinga (University of Twente)
TransformerSimultaneous Localization and MappingTime Series
🎯 What it does: A Transformer-based iMoT model is proposed to achieve more accurate inertial positioning through cross-modal information of acceleration and angular velocity.
Imperceptible 3D Point Cloud Attacks on Lattice-based Barycentric Coordinates
Keke Tang (Guangzhou University), Zhihong Tian (University of Science and Technology of China)
Adversarial AttackPoint Cloud
🎯 What it does: By constructing a separable permutohedral lattice on the three-dimensional point cloud, extracting the barycentric coordinates (LBC) within each unit, and adding constraints in this local parameter space, we achieve perturbation attacks on the point cloud while maintaining the manifold properties, resulting in more covert adversarial samples.
Implicit Location-Caption Alignment via Complementary Masking for Weakly-Supervised Dense Video Captioning
Shiping Ge (Nanjing University), Qing Gu (Nanjing University)
RecognitionGenerationTransformerLarge Language ModelVideoText
🎯 What it does: A weakly supervised dense video captioning method is proposed, utilizing complementary masking to achieve implicit alignment between event locations and captions, avoiding the traditional complex event proposal process.
Implicit Relative Labeling-Importance Aware Multi-Label Metric Learning
Jun-Xiang Mao (Southeast University), Min-Ling Zhang (Lenovo Group)
ClassificationOptimizationText
🎯 What it does: This paper proposes a multi-label metric learning framework ILIA based on implicit relative label importance (RLI), which can automatically recover and utilize RLI to learn more accurate similarity metrics without the need for label importance annotations.
Implicit Word Reordering with Knowledge Distillation for Cross-Lingual Dependency Parsing
Zhuoran Li (Beihang University), Richong Zhang (Beihang University)
Knowledge DistillationRecurrent Neural NetworkTransformerText
🎯 What it does: This paper proposes an implicit word order rearrangement and knowledge distillation framework (IWR-KD), which trains a teacher model based on target language POS to guide the student model in learning dependency syntax parsing in the source language while implicitly transferring the word order relationships of the target language.
Importance Weighting Can Help Large Language Models Self-Improve
Chunyang Jiang (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
GenerationOptimizationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes a DS weight to measure the degree of distribution shift in self-generated data, and combines it with self-consistency to construct the IWSI self-improvement framework;
Improved Approximation Algorithms for Clustered TSP and Subgroup Planning
Jingyang Zhao (University of Electronic Science and Technology of China), Ziliang Xiong (University of Electronic Science and Technology of China)
Optimization
🎯 What it does: This paper proposes new approximation algorithms for the Group Traveling Salesman Problem (SGPP) and the Clustering Traveling Salesman Problem (CTSP), providing various polynomial and FPT-level approximation ratios.
Improved Bounds for Online Facility Location with Predictions
Dimitris Fotakis (National Technical University of Athens), Thanos Tolias (National Technical University of Athens)
Optimization
🎯 What it does: This paper proposes a learning-enhanced algorithm for the Online Facility Location (OFL) problem and provides a competitive ratio analysis of the algorithm under different prediction error scenarios.
Improved Fixed-Parameter Bounds for Min-Sum-Radii and Diameters k-Clustering and Their Fair Variants
Sandip Banerjee (IDSIA USI-SUPSI), Alon Hovav (Hebrew University of Jerusalem)
Anomaly DetectionOptimization
🎯 What it does: This paper provides improved upper and lower bounds for the Minimum Sum Radius (MSR) and Minimum Sum Diameter (MSD) clustering problems with a fixed number of clusters k. An exact MSD algorithm is proposed with a running time of n^O(k). Additionally, (1 + ε)-approximation algorithms are provided for both MSR and MSD, with a running time of O(kn) + (1/ε)O(d^k), applicable to metric spaces of dual dimensions d.
Improved Maximin Share Approximations for Chores by Bin Packing
Jugal Garg (University of Illinois), Erel Segal-Halevi (Ariel University)
Optimization
🎯 What it does: This paper addresses the maximum minimum share (MMS) fair allocation problem for indivisible chores. It improves the heterogeneous version of the First-Fit Decreasing (FFD) algorithm, known as the HFFD algorithm, and proposes three new MMS approximation results: 1) an MMS approximation of 1-out-of-⌊9/11n⌋ for any additive instance; 2) an exact MMS allocation for factored (reducible) instances; 3) a 15/13-MMS approximation for personalized bivalued instances.
Improved Rates of Differentially Private Nonconvex-Strongly-Concave Minimax Optimization
Ruijia Zhang (Chinese University of Hong Kong), Di Wang (King Abdullah University of Science and Technology)
OptimizationSafty and PrivacyTabular
🎯 What it does: Study the non-convex-concave minimax problem under the differential privacy framework, analyze DP-SGDA, and propose an improved PrivateDiff Minimax algorithm;
Improved Regret Bounds for Online Fair Division with Bandit Learning
Benjamin Schiffer (Harvard University), Shirley Zhang (Harvard University)
OptimizationReinforcement Learning
🎯 What it does: This paper studies the problem of online fair allocation, particularly in cases where the types of items are limited and the players' values for the items come from an unknown distribution. An algorithm is proposed to maximize expected social welfare while satisfying expected proportionality constraints.
Improving Cancer Gene Prediction by Enhancing Common Information Between the PPI Network and Gene Functional Association
Chao Deng (Central South University), Jianxin Wang (Central South University)
ClassificationDrug DiscoveryGraph Neural NetworkGraphBiomedical Data
🎯 What it does: The DISFusion method is proposed, which integrates multi-omics features, protein-protein interaction networks (PPI), and gene functional associations (hypergraphs) to identify cancer genes, and enhances the common information between PPI and functional associations through cross-view decorrelation loss.
Improving Complex Reasoning over Knowledge Graph with Logic-Aware Curriculum Tuning
Tianle Xia (Wuhan University), Dacheng Tao (Nanyang Technological University)
TransformerLarge Language ModelSupervised Fine-TuningGraph
🎯 What it does: Using large language models (LLM) through instruction tuning and knowledge graph (KG) context, a logic-aware curriculum learning framework (LACT) is proposed to achieve reasoning for complex logical queries (EFO 1) on incomplete KGs.
Improving Cooperation in Language Games with Bayesian Inference and the Cognitive Hierarchy
Joseph Bills (Brigham Young University), Diego Blaylock (Brigham Young University)
Reinforcement LearningText
🎯 What it does: A Bayesian cooperative agent that can adapt to the semantic and pragmatic uncertainties of partners is designed in the Codenames language game.
Improving Deep Learning Speed and Performance Through Synaptic Neural Balance
Antonios Alexos (University of California), Pierre Baldi (University of California)
OptimizationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkImageText
🎯 What it does: This paper proposes and validates the Synaptic Neural Balance theory, which maintains a balance between input and output weight costs during the training process by proportionally balancing the input and output synaptic weights of individual neurons, thereby accelerating convergence and improving accuracy.
Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators
Dingkang Yang (Fudan University), Lihua Zhang (Fudan University)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: By introducing a hallucination and truth comparator during the inference phase, comparative decoding is performed on the next token prediction of large language models, thereby reducing the hallucinated content generated by the model.
Improving Federated Domain Generalization Through Dynamical Weights Calculated from Data Influences on Global Model Update
Zikun Zhou (Sichuan University), Feihu Huang (University of Electronic Science and Technology of China)
Domain AdaptationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: A data influence-based federated domain generalization method (DI) is proposed, which dynamically allocates local model weights by calculating the influence of each source domain sample on the global model update, and further adjusts the weights on the server side to enhance the global model's generalization performance on unseen domains.
Improving Generalization for AI-Synthesized Voice Detection
Hainan Ren (Purdue University), Shu Hu (Purdue University)
ClassificationDomain AdaptationContrastive LearningAudio
🎯 What it does: This paper proposes a decoupling framework that enhances the generalization ability of AI-generated speech detection by extracting domain-independent synthetic speech forgery features.
Improving Generalization in Offline Reinforcement Learning via Latent Distribution Representation Learning
Da Wang (Shanxi University), Jiye Liang (Shanxi University)
Representation LearningReinforcement LearningTabular
🎯 What it does: Proposes the Latent Distribution Representation Learning (LAD) framework, which utilizes multiple latent distributions in offline data to perform min-max adversarial recognition of the maximum distribution gap and subsequently reduce the gap, thereby learning representations with stronger generalization capabilities.
Improving Generalization of Deep Neural Networks by Optimum Shifting
Yuyan Zhou (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Singapore University of Technology and Design)
ClassificationObject DetectionOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A technique named Optimum Shifting (OS) is proposed and validated, which treats network matrix multiplication as an underdetermined linear equation system. It shifts parameters from sharp minima to flatter minima while keeping the training loss unchanged, thereby improving the model's generalization performance.
Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization
Yechao Zhang (Huazhong University of Science and Technology), Yanjun Zhang (Huazhong University of Science and Technology)
OptimizationAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: A dynamic max-min optimization framework DM-UAP is proposed to generate universal adversarial perturbations with stronger generalization capabilities across samples and models.
Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path
Yuchen Ren (Xi'an Jiaotong University), Chao Shen (Information Engineering University)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: Improved transferable adversarial attacks based on integrated gradients by refining the integral path (multi-path, diversification, monotonicity) to enhance the attack success rate.
Improving Model Probability Calibration by Integration of Large Data Sources with Biased Labels
Renat Sergazinov (Texas A M University), Hakan Brunzell (Amazon)
Tabular
🎯 What it does: A CalEM method based on the approximate EM algorithm is designed and implemented to improve probability calibration in the presence of a small number of high-quality labels mixed with a large number of biased labels.
Improving Multimodal Social Media Popularity Prediction via Selective Retrieval Knowledge Augmentation
Xovee Xu (University of Electronic Science and Technology of China), Jingkuan Song (University of Electronic Science and Technology of China)
RetrievalRecommendation SystemGraph Neural NetworkImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A retrieval-based multimodal social media popularity prediction framework SKAPP is proposed, which enhances prediction accuracy by retrieving relevant UGC and filtering effective information.
Improving Natural Language Understanding for LLMs via Large-Scale Instruction Synthesis
Lin Yuan (Ant Group), Jun Zhou (Ant Group)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A large-scale high-quality natural language understanding instruction synthesis framework, Hum, is proposed to enhance the performance of LLMs in NLU through multi-task instruction generation.
Improving Pareto Set Learning for Expensive Multi-objective Optimization via Stein Variational Hypernetworks
Minh-Duc Nguyen (VinUniversity), Dung D. Le (Singapore University of Technology and Design)
OptimizationTabular
🎯 What it does: A method named SVH-PSL is proposed, which combines Stein Variational Gradient Descent (SVGD) with Hypernetwork to achieve Pareto set learning for expensive multi-objective optimization problems through multi-sample sampling and local kernel functions.
Improving Private Random Forest Prediction Using Matrix Representation
Arisa Tajima (University of Massachusetts Amherst), Amir Houmansadr (University of Massachusetts Amherst)
ClassificationSafty and PrivacyTabular
🎯 What it does: This paper proposes a matrix representation-based random forest method that utilizes a matrix mechanism to achieve differential privacy in training and prediction, significantly improving the accuracy of private random forests.
Improving Retrieval Augmented Language Model with Self-Reasoning
Yuan Xia (Baidu Inc), Haifeng Huang (Baidu Inc)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A self-reasoning framework called SELF-REASONING is proposed to improve the reliability and traceability of retrieval-augmented language models.
Improving the Lower Bound in Branch-and-Bound Algorithms for MaxSAT
Shuolin Li (Aix Marseille Université), Felip Manyà (Universitat de Girona)
OptimizationBenchmark
🎯 What it does: This paper proposes an 'unlocking mechanism' that allows the reuse of soft clauses in discovered cores within the Branch-and-Bound (BnB) MaxSAT solver to enhance lower bound estimation, thereby improving solving efficiency; this mechanism has been integrated into WMaxCDCL (weighted) and MaxCDCL (unweighted) and experimentally validated on large-scale MaxSAT evaluation benchmarks.
Improving Transformer Based Line Segment Detection with Matched Predicting and Re-ranking
Xin Tong (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
Object DetectionTransformerImage
🎯 What it does: A segment detection method based on Transformer called RANK-LETR is proposed, which employs matching prediction, reordering, and ranking loss, enabling rapid convergence and improved detection quality without the need for bipartite matching.
In-Context Policy Adaptation via Cross-Domain Skill Diffusion
Minjong Yoo (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Domain AdaptationAutonomous DrivingReinforcement LearningDiffusion modelContrastive LearningSequential
🎯 What it does: The ICPAD framework is proposed to achieve instant policy adaptation for cross-domain long-term tasks without model updates.
In-context Prompt-augmented Micro-video Popularity Prediction
Zhangtao Cheng (University of Electronic Science and Technology of China), Fan Zhou (University of Electronic Science and Technology of China)
RetrievalRecommendation SystemTransformerPrompt EngineeringVision Language ModelVideoMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper studies the prediction of micro-video popularity and proposes a retrieval-enhanced contextual prompt framework (ICPF) that improves prediction performance without fine-tuning the parameters of the pre-trained model.
In-Dataset Trajectory Return Regularization for Offline Preference-based Reinforcement Learning
Songjun Tu (State Key Laboratory of Multimodal Artificial Intelligence Systems, Chinese Academy of Sciences), Dongbin Zhao
Reinforcement Learning from Human FeedbackTransformerReinforcement LearningTabular
🎯 What it does: This paper proposes the DTR method, which combines Decision Transformer and TD learning to address the reward bias problem in offline preference reinforcement learning.
In-depth Analysis of Low-rank Matrix Factorisation in a Federated Setting
Constantin Philippenko (Inria Paris), Laurent Massoulié (Inria Paris)
OptimizationFederated LearningTabular
🎯 What it does: In a federated learning environment, this paper proposes a low-rank matrix factorization algorithm based on random power iteration initialization and local gradient descent.
In2NeCT: Inter-class and Intra-class Neural Collapse Tuning for Semantic Segmentation of Imbalanced Remote Sensing Images
Junao Shen (Zhejiang University), Wei Zhang (Zhejiang University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: A framework called In2NeCT based on neural folding optimization is proposed, which uses a fixed ETF classifier and cross-entropy, contrastive regularization to enhance the performance of minority classes in remote sensing image semantic segmentation.
Inapproximability of Optimal Multi-Agent Pathfinding Problems
Xing Tan (Lakehead University), Alban Grastien (Paris-Saclay University)
OptimizationGraph
🎯 What it does: The paper proves that solving optimal multi-agent pathfinding (MAPF/diMAPF) on directed and undirected graphs, as well as the optimization versions with a given maximum makespan or maximum number of target reaches, is NP-hard and inapproximable through various polynomial-time reductions.
Incomplete and Unpaired Multi-View Graph Clustering with Cross-View Feature Fusion
Liang Zhao (Dalian University of Technology), Bo Xu (Dalian University of Technology)
Representation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: This paper proposes a clustering framework MGCCFF that simultaneously handles incomplete and unpaired multi-view data, utilizing pseudo-label learning, cross-view feature fusion, and self-expressive autoencoders to achieve graph structure learning for more accurate spectral clustering.
Incomplete Modality Disentangled Representation for Ophthalmic Disease Grading and Diagnosis
Chengzhi Liu (University of Exeter), Yanda Meng (University of Exeter)
ClassificationKnowledge DistillationRepresentation LearningTransformerMultimodalityBiomedical Data
🎯 What it does: This paper proposes the Incomplete Modality Disentangled Representation (IMDR) strategy, combined with the Joint Proxy Learning (JPL) module, to achieve disease grading and diagnosis in the presence of missing multimodal data in ophthalmology.
Incomplete Multi-view Clustering via Diffusion Contrastive Generation
Yuanyang Zhang (Southeast University), Jie Xu
Diffusion modelContrastive LearningMultimodality
🎯 What it does: This paper proposes a missing multi-view clustering method (DCG) that integrates diffusion models and contrastive learning, capable of recovering missing views and achieving end-to-end clustering with only a small number of paired samples.
Incomplete Multi-View Multi-Label Classification via Diffusion-Guided Redundancy Removal
Shilong Ou (Beijing University of Posts and Telecommunications), Yuankai Qi (Macquarie University)
ClassificationDiffusion modelAuto EncoderImage
🎯 What it does: Proposes the DiffSumm framework, which utilizes diffusion models to recover missing view information and designs a redundant view identification strategy to achieve incomplete multi-view multi-label classification.
Increasing Revenue in Efficient Combinatorial Auctions by Learning to Generate Artificial Competition
Maria-Florina Balcan (Carnegie Mellon University), Tuomas Sandholm (Carnegie Mellon University)
Optimization
🎯 What it does: A new combinatorial auction mechanism f-VCG is proposed, which enhances revenue by adding 'artificial competition' to the VCG pricing while striving to maintain efficiency and incentive compatibility. This mechanism is based on the (π, κ)-IR model, which constrains the number and amount of excessive charges acceptable to participants, and subsequently solves for the optimal bidding competitors under three information models (complete distribution, quantile, historical data) and provides a learning algorithm.
Incremental Nyström-based Multiple Kernel Clustering
Yu Feng (National University of Defense Technology), Xinwang Liu (National University of Defense Technology)
Image
🎯 What it does: This paper proposes an Incremental Nyström Multi-kernel Clustering method (INMKC), which achieves efficient clustering of large-scale dynamic data through hierarchical sampling and incremental learning.
Individually Stable Dynamics in Coalition Formation over Graphs
Angelo Fanelli (Paris-Dauphine University), Luca Moscardelli (University of Chieti-Pescara)
Graph
🎯 What it does: This paper studies the convergence of Individual Stability (IS) dynamics in the Thaddeus game, providing convergence and cycle criteria and upper bounds under different preference types and graph structures.
Inductive Learning of Logical Theories with LLMs: A Expressivity-graded Analysis
João Pedro Gandarela de Souza (Idiap Research Institute), André Freitas (University of Manchester)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A logical theory induction method that iteratively combines large language models (LLMs) with formal reasoning engines is proposed, and a graded evaluation of LLM performance in logical theory induction is conducted based on rule expressiveness.
Infer Human’s Intentions Before Following Natural Language Instructions
Yanming Wan (University of Washington), Natasha Jaques (University of Washington)
Robotic IntelligenceTransformerLarge Language ModelMultimodalityChain-of-Thought
🎯 What it does: The FISER framework is proposed, modeling human intentions as intermediate reasoning steps, separating social reasoning from embodied reasoning to better execute ambiguous natural language instructions.
Infer the Whole from a Glimpse of a Part: Keypoint-Based Knowledge Graph for Vehicle Re-Identification
Kai Lv (Beijing Jiaotong University), Youfang Lin (Beijing Jiaotong University)
RecognitionAutonomous DrivingTransformerImage
🎯 What it does: A keypoint-based vehicle re-identification framework is designed, which first detects vehicle keypoints, then infers the features of invisible parts using knowledge graphs and Transformers, and finally fuses global and local information to generate a complete vehicle representation.
Infinite-Canvas: Higher-Resolution Video Outpainting with Extensive Content Generation
Qihua Chen (University of Science and Technology of China), Wei Liu (Tencent)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A high-resolution video outpainting method based on diffusion models is proposed, capable of achieving 9× content generation while maintaining spatial and temporal consistency.
Influence-Based Fair Selection for Sample-Discriminative Backdoor Attack
Qi Wei (Nanyang Technological University), Bo An (Skywork AI)
Adversarial AttackImage
🎯 What it does: A fair sample selection method based on influence functions (IFS) is proposed to enhance the attack success rate (ASR) of backdoor attacks under low visibility triggers (with ε being very small) and to address the ASR fluctuation issues caused by class imbalance.
Information-Theoretic Generative Clustering of Documents
Xin Du (Waseda University), Kumiko Tanaka-Ishii (Waseda University)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A generative document clustering method is proposed that uses large language models to generate text and implements KL divergence from information theory.
Inheriting Generalized Learngene for Efficient Knowledge Transfer across Multiple Tasks
Yuankun Zu (Southeast University), Xin Geng (Southeast University)
Object DetectionSegmentationKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: CLKG enhances multi-task visual performance by first using a small convolution kernel model to perform dense feature distillation from a large convolution kernel CNN, selecting and inheriting learning genes to initialize sub-models of different architectures.
IniRetinex: Rethinking Retinex-type Low-Light Image Enhancer via Initialization Perspective
Guodong Fan (Shandong Technology and Business University), Min Gan (Fuzhou University)
RestorationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A low-light image enhancement method based on initial illumination adaptive compensation and optimization, called IniRetinex, is proposed.
InpDiffusion: Image Inpainting Localization via Conditional Diffusion Models
Kai Wang (Beijing University of Posts and Telecommunications), Jiwei Zhang (Beijing University of Posts and Telecommunications)
RestorationTransformerDiffusion modelImage
🎯 What it does: This paper proposes an image inpainting localization method based on a conditional diffusion model (InpDiffusion), which transforms the localization task into mask generation and achieves progressively refined predictions through edge supervision and a dual-stream multi-scale feature extractor.
InstantSticker: Realistic Decal Blending via Disentangled Object Reconstruction
Yi Zhang (Shanghai Jiao Tong University), Wenjun Zhang (Shanghai Jiao Tong University)
GenerationOptimizationNeural Radiance FieldMesh
🎯 What it does: We propose InstantSticker, a decoupled reconstruction pipeline based on IBL that enables instant and realistic decal integration on reconstructed 3D surfaces, supporting real-time rendering.
Instruct Where the Model Fails: Generative Data Augmentation via Guided Self-contrastive Fine-tuning
Weijian Ma (Fudan University), Shouhong Ding (Tencent)
GenerationData SynthesisVision Language ModelDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes a dynamic data augmentation strategy based on VLM coordinators and self-contrast fine-tuning to enhance the generalization ability of few-shot class incremental learning models.
InstructAvatar: Text-Guided Emotion and Motion Control for Avatar Generation
Yuchi Wang (Peking University), Jiang Bian (Peking University)
GenerationLarge Language ModelDiffusion modelVideoText
🎯 What it does: Achieving emotional expression and facial action control of 2D talking avatars through natural language instructions.
Instruction-Augmented Long-Horizon Planning: Embedding Grounding Mechanisms in Embodied Mobile Manipulation
Fangyuan Wang (Hong Kong Polytechnic University), David Navarro-Alarcon (Hong Kong Polytechnic University)
Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes the IALP system, which utilizes large language models and perception-based grounded predicates to construct PDDL problems in real-time, and achieves long-term mobile manipulation tasks for humanoid robots through planning with LLM in a closed loop.
Instruction-guided Multi-Granularity Segmentation and Captioning with Large Multimodal Model
Xu Yuan (Hong Kong Polytechnic University), Jinsong Lan (TAO Technology)
SegmentationGenerationTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: A multi-granularity segmentation and captioning model (MGLMM) is proposed, which can adaptively switch between fine-grained and overall segmentation and description based on user instructions. The MGSCData dataset and a unified SegCap data format are also created.
InstructOCR: Instruction Boosting Scene Text Spotting
Chen Duan (Meituan), Junfeng Luo (Meituan)
RecognitionTransformerVision Language ModelImageText
🎯 What it does: A scene text recognition framework called InstructOCR is proposed, which guides the model to better understand the text in images through human language instructions.
Int*-Match: Balancing Intra-Class Compactness and Inter-Class Discrepancy for Semi-Supervised Speaker Recognition
Xingmei Wang (Harbin Engineering University), Zijian Liu (Harbin Engineering University)
RecognitionConvolutional Neural NetworkContrastive LearningAudio
🎯 What it does: Proposes Int*-Match, a pseudo-label selection method that simultaneously considers intra-class compactness and inter-class differences in semi-supervised speaker recognition.
Int2Planner: An Intention-based Multi-modal Motion Planner for Integrated Prediction and Planning
Xiaolei Chen (Shanghai Jiao Tong University), Pai Peng (COWAROBOT Co. Ltd.)
Autonomous DrivingTransformerMultimodalityPoint CloudBenchmark
🎯 What it does: Designed and implemented Int2Planner, an intent-driven multimodal motion planner that integrates prediction and planning, capable of providing multiple predicted and planned trajectories in a single forward inference.
InteDisUX: Intepretation-Guided Discriminative User-Centric Explanation for Time Series
Viet-Hung Tran (Queen's University Belfast), Son T. Mai (Queen's University Belfast)
Explainability and InterpretabilityRecurrent Neural NetworkTime SeriesElectrocardiogram
🎯 What it does: A framework for interpretability based on segmented integral gradients, InteDisUX, is proposed for deep time series classification models.
Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily
Siqi Liu (Tianjin University), Weixiong Zhang (Hong Kong Polytechnic University)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A collaborative training framework EG-GCN based on edge discrimination and grouped convolution is proposed to aggregate similar and dissimilar neighbors in heterogeneous graphs, thereby improving node classification performance.
Integrating Inference and Experimental Design for Contextual Behavioral Model Learning
Gongtao Zhou (Beijing Institute of Technology), Haoran Yu (Beijing Institute of Technology)
Recommendation SystemAnomaly DetectionOptimizationTabularFinance Related
🎯 What it does: The study investigates how to learn behavioral models in user contexts through sequential experimental design combined with Bayesian inference.
Integrating Large Language Models and Möbius Group Transformations for Temporal Knowledge Graph Embedding on the Riemann Sphere
Sensen Zhang (Renmin University of China), Yuefeng Ma
TransformerLarge Language ModelSupervised Fine-TuningGraphTime Series
🎯 What it does: This paper proposes embedding temporal knowledge graphs into projective geometric space and combines Möbius group transformations with large language models to construct the 5EL model, which better captures chain, hierarchical, and cyclic temporal relationships.
Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation
Yuhao Qing (Shanghai University), Yueying Wang (Shanghai University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: In unsupervised semantic segmentation, the IL2Vseg method is proposed, which enhances segmentation continuity and accuracy by supplementing self-supervised pre-trained features with low-level visual cues.