AAAI 2026 Papers — Page 19
AAAI Conference on Artificial Intelligence · 4149 papers
Hyperbolic Hierarchical Alignment Reasoning Network for Text-3D Retrieval
Wenrui Li (Harbin Institute of Technology), Xiaopeng Fan (Harbin Institute of Technology)
RetrievalContrastive LearningTextPoint Cloud
🎯 What it does: Proposed the H2ARN model, which maps text and 3D point clouds into the hyperbolic space of the Lorentz model, and achieves cross-modal retrieval through hierarchical ranking loss and contribution-aware aggregation.
Hyperbolic-Enhanced Mixture-of-Experts Mamba for Sequential Recommendation
Yuwen Liu (China University of Petroleum East China), Amin Beheshti (Nanjing Institute of Technology)
Recommendation SystemGraph Neural NetworkMixture of ExpertsAuto EncoderGraphSequential
🎯 What it does: Propose a Mamba sequence recommendation framework HM2Rec that integrates hyperbolic space feature learning, structural reconstruction, and dynamic expert mixture
HyperCOD: The First Challenging Benchmark and Baseline for Hyperspectral Camouflaged Object Detection
Shuyan Bai (Beijing Institute of Technology), Jianan Li (Beijing Institute of Technology)
SegmentationTransformerPrompt EngineeringImageBenchmark
🎯 What it does: Proposed the first large-scale hyperspectral concealed object detection benchmark, HyperCOD, and designed the HSC-SAM framework based on SAM, achieving spatial-spectral decomposition and adaptive prompting for hyperspectral images to complete concealed object segmentation.
HyperD: Hybrid Periodicity Decoupling Framework for Traffic Forecasting
Minlan Shao (Jilin University), Xin Wang (Jilin University)
Graph Neural NetworkTime Series
🎯 What it does: Propose the HyperD framework, which utilizes hybrid periodic decoupling to achieve traffic flow prediction;
HyperDiag: Temporal–Regional Hypergraph Learning via Topology-Enhanced State Propagation for Brain Disease Diagnosis
Yulan Ma (Beihang University), Yang Li (Beihang University)
ClassificationGraph Neural NetworkGraphTime SeriesBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Propose a diagnostic framework named HyperDiag that integrates time evolution and regional hypergraphs for dynamic functional network modeling and classification of brain diseases using rs-fMRI.
HyperGLLM: An Efficient Framework for Endpoint Threat Detection via Hypergraph-Enhanced Large Language Models
Hongyi Zhou (360 Security Technology Inc), Hanzhong Zheng (Tsinghua University)
Anomaly DetectionComputational EfficiencyGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A framework named HyperGLLM is proposed for endpoint detection and response (EDR) logs, combining hypergraph reasoning with large language models (LLMs) to efficiently identify covert threat behaviors in ultra-long, interwoven logs.
HyperGOOD: Towards Out-of-Distribution Detection in Hypergraphs
Tingyi Cai (Zhejiang Normal University), Zhonglong Zheng (Zhejiang Normal University)
Anomaly DetectionGraph
🎯 What it does: This paper proposes the hypergraph-oriented discrete distribution detection task (HOOD) and introduces the HyperGOOD framework to realize this task.
Hypergraph-Based Multi-View Multi-Label Classification via Adaptive High-Order Semantic Fusion
Yi Shan (Beijing University of Technology), Honggui Han (Beijing University of Technology)
ClassificationGraph Neural NetworkContrastive Learning
🎯 What it does: This paper proposes a hypergraph-based multi-view multi-label classification method called HyperAHSF, which achieves efficient fusion of information within and across views through adaptive high-order semantic fusion.
HyperLoad: A Cross-Modality Enhanced Large Language Model-Based Framework for Green Data Center Cooling Load Prediction
Haoyu Jiang (Zhejiang University), Wei Zhong (Zhejiang University)
OptimizationTransformerLarge Language ModelTime Series
🎯 What it does: Propose the HyperLoad framework to achieve cooling load prediction in green energy data centers
HyperNoRA: Hyperedge Prediction via Node-Level Relation-Aware Self-Supervised Hypergraph Learning
Ming Li, Ke Lv (University Of Chinese Academy Of Sciences)
Representation LearningGraph Neural NetworkContrastive LearningGraphBenchmark
🎯 What it does: Proposed the HyperNoRA framework for hyperedge prediction, combining global node relation graphs and self-supervised contrastive learning to enhance node and hyperedge representations.
HyperSHAP: Shapley Values and Interactions for Explaining Hyperparameter Optimization
Marcel Wever (Leibniz University Hannover), Marius Lindauer (Leibniz University Hannover)
Explainability and InterpretabilityHyperparameter SearchBenchmark
🎯 What it does: This paper proposes HyperSHAP, a game-theoretic explanation framework based on Shapley values and interactions, for local and global explainability analysis of the hyperparameter optimization (HPO) process.
HyperSign: Hierarchical Hypergraph-based Co-occurrence Modeling for Sign Language Recognition and Translation
Qianren Guo (Jilin University), Yu Jiang (Jilin University)
RecognitionImage TranslationGraph Neural NetworkTransformerLarge Language ModelVideo
🎯 What it does: This paper proposes HyperSign, a hierarchical hypergraph-based skeletal network for sign language recognition and translation.
HyperSign: Saliency-Aware Spatial Graphs and Temporal Hypergraphs for Continuous Sign Language Recognition
Weiyi Ye (Zhejiang University of Technology), Xiao-Xin Li (Zhejiang University of Technology)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkVideo
🎯 What it does: Propose a multi-scale spatiotemporal hypergraph network called HyperSign, which learns features for continuous sign language recognition by utilizing significance-aware spatial graphs and temporal hypergraphs.
Hypothesis-Driven Reasoning for Large Language Models
Aakash Kumar Agarwal (Indian Institute of Technology Bombay), Moyuru Yamada (Fujitsu Research of India)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the Hypothesis-Driven Reasoning (HDR) framework, which automatically extracts semantic memory from multimodal experiences through a generate-validate loop, significantly enhancing the reasoning capabilities of large language models.
HyRNN: Hybrid Recurrent Neural Networks for Approximating Hybrid Dynamical Systems
Ricardo G. Sanfelice (University of California, Santa Cruz)
Recurrent Neural NetworkTime SeriesSequentialPhysics Related
🎯 What it does: This paper proposes a hybrid dynamic recurrent neural network (HyRNN) that can approximate solutions of hybrid systems with arbitrary precision within a given time frame, and demonstrates the approximation on a bouncing ball system in simulation experiments.
I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders
Andrey V. Galichin (AIRI), Ivan Oseledets (AIRI)
Explainability and InterpretabilityLarge Language ModelAuto EncoderText
🎯 What it does: Researchers decomposed the internal activations of large language models using sparse autoencoders (SAE) and constructed an automatic metric called ReasonScore to identify interpretable features related to reasoning.
I-CAM-UV: Integrating Causal Graphs over Non-Identical Variable Sets Using Causal Additive Models with Unobserved Variables
Hirofumi Suzuki (Fujitsu Limited), Shohei Shimizu (Shiga University)
OptimizationExplainability and InterpretabilityGraph
🎯 What it does: To address observational data with multiple variables having incomplete consistency, this paper proposes the I-CAM-UV method, which integrates CAM-UV results to enumerate all causal DAGs satisfying consistency constraints, thereby recovering causal relationships between hidden variables and unobserved variables.
I-INR: Iterative Implicit Neural Representations
Ali Haider (Kyung Hee University), Sung-Ho Bae (Kyung Hee University)
RestorationSuper ResolutionImageMesh
🎯 What it does: Investigated an iterative implicit neural representation framework, I-INR, which progressively reconstructs signals in multi-step iterations, enhancing high-frequency detail recovery and noise robustness.
I2CD: An Invertible Causal Framework for Compositional Zero-Shot Learning via Disentangle-Compose-Disentangle
Zhaoquan Yuan (Southwest Jiaotong University), Changsheng Xu (Chinese Academy of Sciences)
ClassificationVision Language ModelFlow-based ModelMultimodalityBenchmark
🎯 What it does: This paper proposes the I2CD framework, which utilizes reversible networks and causal interventions to achieve disentanglement between attributes and objects and generate counterfactual images.
I2E: Real-Time Image-to-Event Conversion for High-Performance Spiking Neural Networks
Ruichen Ma (University of Electronic Science and Technology of China), Shaogang Hu (University of Electronic Science and Technology of China)
ClassificationData SynthesisConvolutional Neural NetworkSpiking Neural NetworkImage
🎯 What it does: Proposed the I2E framework, which can convert static images into high-fidelity event streams in real-time, thereby addressing the scarcity of event data in SNN training.
IAD-R1: Reinforcing Consistent Reasoning in Industrial Anomaly Detection
Yanhui Li (Sun Yat-sen University), Chao Huang (Sun Yat-sen University)
Anomaly DetectionExplainability and InterpretabilitySupervised Fine-TuningReinforcement LearningVision Language ModelImageTextBenchmarkChain-of-Thought
🎯 What it does: Developed a two-stage post-training framework IAD-R1 to enhance the reasoning and discrimination capabilities of Vision-Language Models (VLM) in industrial defect detection.
ICAD-LLM: One-for-All Anomaly Detection via In-Context Learning with Large Language Models
Zhongyuan Wu (Beihang University), Changqing Ma (Capinfo Co., Ltd.)
Anomaly DetectionTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextMultimodalityTabularTime Series
🎯 What it does: This paper proposes a unified anomaly detection framework called ICAD-LLM, which leverages large language models to achieve consistent processing and anomaly determination for multi-modal data by providing normal sample context during inference.
ICL-Router: In-Context Learned Model Representations for LLM Routing
Chenxu Wang (Fudan University), Shuyue Hu (Beijing Institute of Technology)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes a context vector-based model routing framework called ICL-Router, which can dynamically assign the most suitable LLM for each query and supports adding new models without training.
ICLR: Inter-Chrominance and Luminance Interaction for Natural Color Restoration in Low-Light Image Enhancement
Xin Xu (Wuhan University of Science and Technology), Kui Jiang (Wuhan University of Science and Technology)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed an ICLR framework based on the HVI color space, achieving color restoration and detail enhancement in low-light images through a dual-stream interaction enhancement module and covariance correction loss.
ICM-Fusion: In-Context Meta-Optimized LoRA Fusion for Multi-Task Adaptation
Yihua Shao (Hong Kong Polytechnic University), Jingcai Guo (Peking University)
Computational EfficiencyMeta LearningTransformerAuto EncoderImageTextMultimodality
🎯 What it does: This paper proposes a method called ICM-Fusion, which utilizes task vectors and contextual information to integrate multi-task LoRA adapters through the combination of Meta-Learning and Fusion VAE, constructing a unified model capable of rapid generalization in multi-task scenarios without relying on original data.
ID-Splat: Propagating Object Identities for Segmenting 3D Aerial-view Scenes
Yijing Wang (Xidian University), Jingjing Ma (Xidian University)
SegmentationGaussian SplattingPoint Cloud
🎯 What it does: Proposed a 3D aerial view segmentation framework called ID-Splat based on multi-view object identity, which assigns semantics to 3D Gaussian Splatting points through Mask-Object Tracking and Object Integration & Propagation.
IdealTSF: Can Non-Ideal Data Contribute to Enhancing the Performance of Time Series Forecasting Models?
Hua Wang (Ludong University), Fan Zhang (Shandong Technology and Business University)
Data SynthesisOptimizationAdversarial AttackData-Centric LearningTransformerTime SeriesBenchmark
🎯 What it does: Developed the IdealTSF framework, integrating three stages: negative sample pre-training, positive sample generation, and ECOS optimization, to enhance the robustness and accuracy of time series forecasting.
IdeFN: Identifying Unclicked Space False Negatives via Relaxed Partial Optimal Transport for Conversion Rate Prediction
Weiyi Zhong (Qufu Normal University), Qiang Ni (Great Bay University)
Recommendation SystemTabular
🎯 What it does: Propose the IdeFN multi-task framework, utilizing CTR as an auxiliary task to identify and re-annotate false negative samples among unclicked instances in the exposure space, thereby enhancing CVR prediction performance.
Identifying and Analyzing Performance-Critical Tokens in Large Language Models
Yu Bai (Beijing Institute Of Technology), Jackie Chi Kit Cheung (Mila Quebec Artificial Intelligence Institute)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper conducts a fine-grained classification of prompts in large language models (LLMs) during few-shot learning (ICL), identifying and analyzing which types of tokens (template tokens, stop tokens, content tokens) most directly affect model performance, proposes the concept of 'performance-critical tokens,' and investigates their characteristics (semantic meaning, repetitiveness, structural features).
Identifying Imperfect Clones in Elections
Piotr Faliszewski (AGH University of Kraków), Ildikó Schlotter (AGH University of Kraków)
OptimizationComputational EfficiencyGraph
🎯 What it does: Studied the definitions and identification of imperfect cloning (approximate cloning, independent cloning, sub-district cloning) in elections, and provided corresponding computational complexity and algorithms.
Identity-Aware Vision-Language Model for Explainable Face Forgery Detection
Junhao Xu (Fudan University), Yu-Gang Jiang (Fudan University)
Anomaly DetectionExplainability and InterpretabilityTransformerVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: Designed a personalized vision-language model that integrates low-level visual cues with high-level semantic consistency for facial forgery detection and providing interpretable explanations.
IdentityStory: Taming Your Identity-Preserving Generator for Human-Centric Story Generation
Donghao Zhou (Hong Kong University of Science and Technology), Pheng-Ann Heng (Hong Kong University of Science and Technology)
SegmentationGenerationData SynthesisTransformerVision Language ModelDiffusion modelContrastive LearningImageTextBenchmark
🎯 What it does: Propose the IdentityStory framework to achieve text-based continuous image generation, ensuring identity consistency of characters across multiple frames.
IDK-S: Incremental Distributional Kernel for Streaming Anomaly Detection
Yang Xu (Nanjing University), Kai Ming Ting (Nanjing University)
Anomaly DetectionTime Series
🎯 What it does: Proposes IDK S-, an incrementally updatable distribution kernel for real-time anomaly detection in data streams.
IE-SRGS: An Internal-External Knowledge Fusion Framework for High-Fidelity 3D Gaussian Splatting Super-Resolution
Xiang Feng (Hangzhou Dianzi University), Yanming Zhu (Griffith University)
Depth EstimationSuper ResolutionTransformerGaussian SplattingImage
🎯 What it does: Proposes the IE-SRGS framework, leveraging the complementary characteristics of external 2D super-resolution and internal 3D Gaussian rendering to achieve super-resolution from low-resolution multi-view inputs to high-resolution 3D Gaussian Splatting.
IGFuse: Interactive 3D Gaussian Scene Reconstruction via Multi-Scans Fusion
Wenhao Hu (Zhejiang University), Gaoang Wang (Zhejiang University)
GenerationGaussian SplattingImage
🎯 What it does: By fusing multi-view 3D Gaussian fields through multiple scans of natural object rearrangements, an interactive complete 3D scene is constructed, enabling object-level editing and rendering.
IGIANet: Illumination Guided Implicit Alignment Network for Infrared–Visible UAV Detection
Xiangqi Chen (Zhejiang Normal University), Hua Wang (Hanyang University)
Object DetectionConvolutional Neural NetworkImageMultimodality
🎯 What it does: IGIANet aims to address weak alignment and modality imbalance issues in UAV RGB-IR object detection by constructing a three-module unified framework comprising illumination-guided frequency domain modulation, frequency domain difference enhancement, and implicit alignment dynamic fusion.
IGT4ETH: An Isotropic Pre-trained Graph Transformer for Ethereum Account Classification
Ao Liu, Qiang Duan (Central University of Finance and Economics)
ClassificationTransformerContrastive LearningGraphFinance Related
🎯 What it does: Propose IGT4ETH, a pre-trained graph Transformer for Ethereum account classification, which explicitly models transaction network topology and eliminates embedding directional imbalance through post-processing;
iMAD: Intelligent Multi-Agent Debate for Efficient and Accurate LLM Inference
Wei Fan (Virginia Tech), Bo Ji (Virginia Tech)
Explainability and InterpretabilityComputational EfficiencyLarge Language ModelAgentic AIPrompt EngineeringTextMultimodalityChain-of-Thought
🎯 What it does: Propose an intelligent multi-agent debate framework called iMAD, which significantly reduces token consumption while maintaining or improving answer accuracy by performing structured self-critique after a single LLM output and deciding whether to trigger a multi-agent debate based on linguistic features.
Image Content Matters: An Image Content Aware State Space Model for Accelerated MRI Reconstruction
Yucong Meng (Fudan University), Yonghong Shi (Fudan University)
RestorationBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study proposes a content-aware Mamba model (CAM), which maps MRI images from a two-dimensional space to a one-dimensional sequence with semantic preservation and texture enhancement, further utilizing a state space model to achieve high-quality accelerated MRI reconstruction.
Image Restoration via Primal Dual Hybrid Gradient and Flow Generative Model
Ji Li (Capital Normal University), Chao Wang (Capital Normal University)
RestorationFlow-based ModelRectified FlowImage
🎯 What it does: Propose a plug-and-play (PnP) framework based on flow-matching generative models, which employs the Primal–Dual Hybrid Gradient (PDHG) algorithm to handle ℓ1 and ℓ2 data fidelity terms for non-Gaussian noise, enabling multi-task image restoration.
Image-Text Knowledge Modeling for Unsupervised Multi-Scenario Person Re-Identification
Zhiqi Pang (Harbin Institute of Technology), Gaurav Sharma (University of Rochester)
RecognitionRetrievalGraph Neural NetworkTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose the unsupervised multi-scenario person re-identification (UMS-ReID) task and design a three-stage image-text knowledge modeling framework called ITKM, which leverages the CLIP model to simultaneously handle multiple scenarios such as visible-infrared, clothing changes, and resolution variations within a single model;
ImageBindDC: Compressing Multi-modal Data with ImageBind-based Condensation
Yue Min (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)
Data SynthesisCompressionMultimodality
🎯 What it does: This paper proposes ImageBindDC, a framework for multi-modal data compression within the unified feature space of ImageBind, capable of synthesizing a small amount of semantically complete synthetic data from large-scale multi-modal datasets.
ImagerySearch: Adaptive Test-Time Search for Video Generation Beyond Semantic Dependency Constraints
Meiqi Wu (University of Chinese Academy of Sciences), Kaiqi Huang (Alibaba Group)
GenerationDiffusion modelImageVideoTextMultimodalityBenchmark
🎯 What it does: Proposed a text-to-video generation method called ImagerySearch based on adaptive test-time search, which enhances video quality under long-distance semantic prompts.
ImageSet2Text: Describing Sets of Images Through Text
Piera Riccio (University of Amsterdam), Nuria M Oliver (Johannes Kepler University Linz)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextGraphBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose ImageSet2Text, capable of automatically generating natural language descriptions for large-scale image collections.
IMAGGarment+: Efficient Attribute-Wise Diffusion for Garment Generation
Jian Yu (Nanjing University of Science and Technology), Xiaoyu Du (Nanjing University)
GenerationDiffusion modelImageBenchmark
🎯 What it does: This paper proposes IMAGGarment+, an efficient attribute-aware diffusion framework that can generate high-quality garment images based on multiple attributes such as outline, color, logo, and position.
Imagine with Layout and Sketch: Enhancing Vision-Language Retrieval with Dual-Stream Multi-Modal Query Refinement
GuangHao Meng (Tsinghua Shenzhen International Graduate School, Tsinghua University), Dan Zhao (Tsinghua Shenzhen International Graduate School, Tsinghua University)
RetrievalLarge Language ModelVision Language ModelDiffusion modelContrastive LearningMultimodality
🎯 What it does: Propose the LASE framework, which leverages large language models to generate layout knowledge and diffusion models to generate sketch knowledge, followed by fusing multi-instance layouts and sketches through a Gated Dual-Stream Knowledge Module (GDKM) to enhance visual-language retrieval performance.
Implicit Neural Representation with Multi-Scale Sine Activation
Jufeng Han (Chinese Academy of Sciences), Yan Pang (Chinese Academy of Sciences)
Representation LearningImageVideoMesh
🎯 What it does: Proposed a multi-scale sine activation function (MSA) to enhance the modeling capability of implicit neural representations (INR) in high-frequency details and multi-scale structures.
Importance-Aware Data Selection for Efficient LLM Instruction Tuning
Tingyu Jiang (Alibaba Cloud Computing), Hao Henry Wang (University of Tokyo)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: For high-quality data selection in LLM instruction fine-tuning, the Model Instruction Weakness Value (MIWV) metric is proposed, and a fully automated data screening process is constructed based on this.
Imprint of the Forgotten: Stealthy Membership Inference in Unlearned Graph Neural Networks
He Zhang (RMIT University), Xun Yi (RMIT University)
Safty and PrivacyAdversarial AttackGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose a prior member inference attack framework targeting unlearned GNNs, leveraging the model's retained imprints to identify deleted edges.
Improved Algorithms for Trip-Vehicle Assignment in Ride-Sharing
Jingyang Zhao (University of Electronic Science and Technology of China), Yonghang Su (University of Electronic Science and Technology of China)
OptimizationComputational Efficiency
🎯 What it does: This paper proposes three new approximation algorithms for the passenger-vehicle assignment problem (RSAP) and provides improved approximation ratios for two cases: n = mk and n < mk.
Improved Differentially Private Algorithms for Rank Aggregation
Quentin Hillebrand (University of Copenhagen), Phanu Vajanopath (University of Tokyo)
Safty and Privacy
🎯 What it does: This paper proposes an improved differential privacy ranking aggregation algorithm, covering both Kemeny and footrule ranking aggregation tasks.
Improved Fully Dynamic Submodular Maximization Under Matroid Constraints
Yiwei Gao (Chinese Academy of Sciences), Zhijie Zhang (Fuzhou University)
Optimization
🎯 What it does: Proposes a dynamic algorithm for submodular maximization under arbitrary matroid constraints with base size k in a fully dynamic environment (supporting insertions and deletions), maintaining an approximate optimal solution for the current element set after each update.
Improved Masked Image Generation with Knowledge-Augmented Token Representations
Guotao Liang (Harbin Institute of Technology), Yunming Ye (Harbin Institute of Technology)
GenerationData SynthesisGraph Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: Proposed a knowledge-enhanced masked image generation framework, KA-MIG, which improves the generation quality of Masked Image Generation by introducing three token-level knowledge graphs based on training data (co-occurrence graph, semantic similarity graph, position-token incompatibility graph).
Improved Runtime Guarantees for the SPEA2 Multi-Objective Optimizer
Benjamin Doerr (Institut Polytechnique de Paris), Milan Stanković (Institut Polytechnique de Paris)
OptimizationBenchmark
🎯 What it does: This paper conducts in-depth theoretical analysis of the selection mechanism in the SPEA2 multi-objective evolutionary algorithm, proving that its expected runtime on common benchmarks (ONEMINMAX, ONEJUMPZEROJUMP, LEADINGONESTRAILINGZEROS) is superior to that of mainstream algorithms such as NSGA-II and is more robust to population size.
Improved Streaming Algorithm for Fair k-Center Clustering
Longkun Guo (Fuzhou University), Chao Chen (Fuzhou University)
OptimizationImageTabularSequential
🎯 What it does: This paper proposes a one-pass streaming algorithm for fair k-center clustering, which supports setting an upper limit on the number of centers for each group. It provides a 3+ε approximation algorithm for semi-structured data streams (arriving in group order) and further extends its application to offline scenarios.
Improving Batch Normalization with Test-Time Adaptation for Robust Object Detection in Self-Driving
Dacheng Liao (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Object DetectionAutonomous DrivingOptimizationTransformerImagePoint Cloud
🎯 What it does: Proposes an adaptive method based on a learnable BN layer, combining geometric confidence maximization (GCM) and entropy minimization (EM) losses, and achieving robust test-time adaptation through a two-stage adaptive strategy with semantic consistency.
Improving Deepfake Detection with Reinforcement Learning-Based Adaptive Data Augmentation
Yuxuan Chou (Tsinghua University), Shu-Tao Xia (Tsinghua University)
Anomaly DetectionData-Centric LearningReinforcement LearningImageVideo
🎯 What it does: Generate multi-domain, multi-level forged samples using adaptive reinforcement learning and causal invariant learning to enhance the generalization ability of deepfake detectors.
Improving Exact Algorithm for Pseudo Boolean Optimization with Two New Phase Selection Heuristics
Yujiao Zhao (Northeast Normal University), Minghao Yin (Northeast Normal University)
OptimizationReinforcement LearningBenchmark
🎯 What it does: Propose two new phase selection strategies and integrate them into existing CDCL PBO solvers to improve solving efficiency.
Improving Generalization in LLM Structured Pruning via Function-Aware Neuron Grouping
Tao Yu (Chinese Academy of Sciences), Jinqiao Wang (Chinese Academy of Sciences)
Computational EfficiencyTransformerText
🎯 What it does: Proposed a post-training structured pruning method called FANG, which groups pruning based on the functional specialization of LLM neurons to enhance generalization in downstream tasks.
Improving Generalization in Offline Meta-Reinforcement Learning via Cross-task Contexts
Hongcai He (University of Electronic Science and Technology of China), Jie Shao (University of Electronic Science and Technology of China)
Meta LearningReinforcement LearningAuto EncoderContrastive LearningTabular
🎯 What it does: Propose the CTMRL framework, leveraging cross-task context to enhance the generalization ability of offline meta-reinforcement learning.
Improving Implicit Discourse Relation Recognition with Natural Language Explanations from LLMs
Heng Wang (East China Jiaotong University), Changxing Wu (East China Jiaotong University)
ClassificationRecognitionExplainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Leverage large language models to generate natural language explanations, and jointly train relation prediction and explanation generation in a lightweight IDRR model to enhance performance and explainability.
Improving Large Molecular Language Model via Relation-aware Multimodal Collaboration
Jinyoung Park (Korea Advanced Institute of Science and Technology), Hyunwoo J. Kim (Korea Advanced Institute of Science and Technology)
Drug DiscoveryGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityGraphBiomedical Data
🎯 What it does: Propose the CoLLaMo model, which synergistically encodes three modalities—1D SELFIES, 2D molecular graphs, and 3D spatial conformations—into a unified token space interpretable by language models through a relation-aware multi-layer projector; simultaneously introduce molecule-centered evaluation metrics (CHARM/RCHARM) and a GPT-4o evaluator to detect hallucinations and description quality in model outputs.
Improving Long-Context Summarization with Multi-Granularity Retrieval Optimization
Xueyu Chen (Tongji University), Cairong Zhao (Tongji University)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: By hierarchically structuring long texts, the method first generates summaries for chapters, blocks, and sentence blocks, constructing a multi-granularity retrieval tree. It then filters context using vector retrieval and re-ranking, ultimately performing question answering or summary generation within an LLM.
Improving Region Representation Learning from Urban Imagery with Noisy Long-Caption Supervision
Yimei Zhang (Zhejiang University of Technology), Xiangjie Kong (Zhejiang University of Technology)
Knowledge DistillationRepresentation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextTabular
🎯 What it does: This paper proposes a cross-modal pre-training framework named UrbanLN, aiming to utilize long-text descriptions generated by multi-modal large language models to achieve fine-grained semantic alignment between urban street scenes and satellite images, while suppressing noise in urban region representation learning.
Improving Sparse IMU-based Motion Capture with Motion Label Smoothing
Zhaorui Meng (Xiamen University), Yipeng Qin (Cardiff University)
Pose EstimationRecurrent Neural NetworkTransformerTime SeriesSequential
🎯 What it does: Designed a label smoothing method for sparse IMU-based human motion capture, generating soft labels using Perlin noise based on the skeleton structure;
Improving Stochastic Action-Constrained Reinforcement Learning via Truncated Distributions
Roland Stolz (Technical University of Munich), Matthias Althoff (Technical University of Munich)
Reinforcement Learning from Human FeedbackReinforcement LearningTabularTime SeriesBenchmark
🎯 What it does: This paper proposes an accurate estimation and efficient sampling method for truncated distributions in action-constrained reinforcement learning, and verifies its effectiveness in three benchmark environments.
Improving Sustainability of Adversarial Examples in Class-Incremental Learning
Taifeng Liu (Xidian University), Zhuo Ma (Xidian University)
Adversarial AttackConvolutional Neural NetworkVision Language ModelImage
🎯 What it does: Propose the SAE method to enhance the sustainability of adversarial samples in class-incremental learning.
Improving Target Presence and Plurality Recognition for Generalized Referring Image Segmentation
Namyup Kim (Pohang University of Science and Technology), Suha Kwak (Pohang University of Science and Technology)
RecognitionSegmentationTransformerVision Language ModelMultimodality
🎯 What it does: Proposes a new framework for Generalized Referring Image Segmentation (RIS), specifically addressing three scenarios: no-target, single-target, and multi-target. The framework introduces learnable target queries to determine target existence and plurality, and generates synthetic no-target and multi-target samples through data augmentation during training, thereby improving segmentation and recognition performance across all scenarios.
Improving the Accuracy of Dense Retrieval on the Quantized Indexes via Gradient Optimization of the Target Embeddings
Cong Tan (Shanghai Jiao Tong University), Tao Fang (Shanghai Jiao Tong University)
RetrievalTextBenchmark
🎯 What it does: Propose an scalable training method that significantly improves dense retrieval performance under quantized index (QHNSW) by directly updating the cached target embeddings via gradient and combining similarity-based approximate negative sampling.
Improving the Convergence Rate of Ray Search Optimization for Query-Efficient Hard-Label Attacks
Xinjie Xu (Zhejiang University of Technology), Chen Ma (Zhejiang University of Technology)
OptimizationAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed and implemented two hard-label attack algorithms based on zeroth-order optimization, ARS-OPT and PARS-OPT, leveraging the Nesterov acceleration idea and momentum mechanism. By using lookahead to pre-estimate gradients when searching for optimal ray directions, query efficiency is significantly improved.
Improving Value-based Process Verifier via Low-Cost Variance Reduction
Zetian Sun (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
Large Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Proposes Composite Monte Carlo Sampling (ComMCS) to reduce estimation errors caused by sampling variance in the value-based process verifier during training, maintaining unbiasedness without incurring additional LLM inference costs.
Impute Missing Entries with Uncertainty
Jaesung Lim (University of Seoul), Jong-June Jeon (University of Seoul)
Data-Centric LearningAuto EncoderTabular
🎯 What it does: Propose U-VAE, a missing value uncertainty modeling method based on VAE, which can provide reliable statistical inference after single or multiple imputations.
In-Token Rationality Optimization: Towards Accurate and Concise LLM Reasoning via Self-Feedback
Mingye Zhu, Yongdong Zhang (University Of Science And Technology Of China)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Propose the InTRO framework for LLM chain-of-thought reasoning, aligning the model's own answer conditional posterior with generation strategies to achieve token-level exploration and self-feedback, enhancing reasoning accuracy and conciseness.
Inapproximability of STRIPS Planning
Xing Tan (Lakehead University), Alban Grastien (Université Paris-Saclay)
Optimization
🎯 What it does: Investigated the approximation infeasibility of the optimization problem of maximizing goal satisfaction in STRIPS planning, providing a constant factor lower bound.
Incoherence as Oracle-less Measure of Error in LLM-Based Code Generation
Thomas Jean-Michel Valentin (ENS Paris-Saclay), Marcel Böhme (Max Planck Institute for Security and Privacy)
AI Code AssistantTransformerLarge Language ModelText
🎯 What it does: Proposed an oracle-free code correctness evaluation metric called incoherence to estimate the error probability of programs generated by LLMs.
Incomplete Multi-view Diabetic Retinopathy Grading via Self-Supervised Inter- and Intra-View Restoration
Zhihao Wu (Shenzhen University), Linlin Shen (Shenzhen University)
ClassificationRestorationTransformerImageBiomedical Data
🎯 What it does: Proposed the first incomplete multi-view diabetic retinopathy grading framework that maintains high accuracy even when some views are missing.
Incomplete Multi-View Unsupervised Federated Feature Selection via Cooperative Particle Swarm Optimization and Tensor-Aligned Learning
Zhiwei Ye (Hubei University of Technology), Jixin Zhang (Hubei University of Technology)
OptimizationFederated LearningMultimodality
🎯 What it does: Proposes a federated learning-based 'IMUFFS' framework for unsupervised feature selection in the presence of multi-view missing data.
Incorporating Self-Rewriting into Large Language Model Reasoning Reinforcement
Jiashu Yao (Beijing Institute of Technology), Yangyang Kang (Zhejiang University)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposed a self-rewrite framework that improves the internal reasoning quality of large reasoning models (LRM) through reinforcement learning (RL), addressing common defects in the reasoning process.
Incorporating Token Importance in Multi-Vector Retrieval
Archish S (Microsoft Research), Neeraj Kayal (Microsoft Research)
RetrievalRepresentation LearningTransformerContrastive LearningTextBenchmark
🎯 What it does: Introduce weighted Chamfer distance into the ColBERT multi-vector retrieval framework to enhance retrieval recall by leveraging token importance.
Incremental Data-Driven Policy Synthesis via Game Abstractions
Irmak Sağlam (Max Planck Institute for Software Systems), Anne-Kathrin Schmuck (Max Planck Institute for Software Systems)
OptimizationSequential
🎯 What it does: Propose an incremental data-driven abstraction and synthesis framework that can learn upper and lower bounds of reachable sets from sampled data in unknown stochastic dynamical systems, construct finite stochastic (2.5-player) games, and incrementally update winning regions and control strategies as new data arrives.
Incremental Maintenance of DatalogMTL Materialisations
Kaiyue Zhao (Shanghai Jiao Tong University), Pan Hu (University of Oxford)
Computational EfficiencyTime SeriesBenchmark
🎯 What it does: Propose the DRedMTL algorithm to achieve incremental maintenance of DatalogMTL rule sets.
IndexTTS2: A Breakthrough in Emotionally Expressive and Duration-Controlled Auto-Regressive Zero-Shot Text-to-Speech
Siyi Zhou (bilibili), Jingchen Shu (bilibili)
GenerationTransformerLarge Language ModelFlow-based ModelTextAudio
🎯 What it does: This paper proposes IndexTTS2, a zero-shot emotionally controllable, precisely duration-controllable autoregressive text-to-speech model.
Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion
Ryoma Yataka (Mitsubishi Electric Corporation), Ryuhei Takahashi (Mitsubishi Electric Research Laboratories)
Object DetectionDiffusion modelPoint Cloud
🎯 What it does: Propose a multi-perspective indoor radar object detection framework REXO, which achieves object detection through 3D bounding box diffusion in radar space.
IndoorUAV: Benchmarking Vision-Language UAV Navigation in Continuous Indoor Environments
Xu Liu (Peking University), Zhouhui Lian (Peking University)
Robotic IntelligenceLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelMultimodalitySequentialBenchmark
🎯 What it does: This paper constructs the IndoorUAV benchmark, which includes a 4-DoF UAV visual language navigation dataset under 1,075 high-fidelity indoor scenes, and proposes the IndoorUAV-Agent model to accomplish long- and short-range tasks.
Induce, Align, Predict: Zero-Shot Stance Detection via Cognitive Inductive Reasoning
Bowen Zhang (Shenzhen Technology University), Genan Dai (Shenzhen Technology University)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Propose a zero-shot stance detection framework based on cognitive inductive reasoning (CIRF), which extracts logical patterns unsupervisedly and constructs a multi-relational graph. The framework aligns input with abstract reasoning patterns through a graph kernel model to predict the stance of unseen targets.
Inductive Generative Recommendation via Retrieval-based Speculation
Yijie Ding (University of California, San Diego), Yupeng Hou (University of California, San Diego)
Recommendation SystemTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes the SpecGR framework, which combines generative recommendation with an induced draft-validation mechanism, enabling the model to recommend new items in an incremental, unlabeled setting.
Inequality in the Age of Pseudonymity
Aviv Yaish (Yale University), Lin William Cong (Ben-Gurion University)
Finance Related
🎯 What it does: This paper conducts an axiomatization study on inequality measures under a pseudo-anonymous environment, proving that traditional inequality measures (such as the Gini coefficient, GE series, etc.) cannot maintain consistency when Sybil (pseudo-name) manipulation exists, and provides a complete characterization of Sybil-proof inequality measures.
InEx: Hallucination Mitigation via Introspection and Cross-Modal Multi-Agent Collaboration
Zhongyu Yang (Xi'an Jiyun Technology Co Ltd), Wei Pang (Xi'an Jiyun Technology Co Ltd)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelAgentic AIVision Language ModelMultimodality
🎯 What it does: Designed a training-agnostic multi-agent framework InEx that automatically alleviates hallucination problems in multi-modal large language models through internal self-reflection and external cross-modal collaboration.
Inference Offloading for Cost-Sensitive Binary Classification at the Edge
Vishnu Narayanan Moothedath (KTH Royal Institute of Technology), Sharayu Moharir (KTH Royal Institute of Technology)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkImageTextBiomedical DataComputed Tomography
🎯 What it does: Propose an online two-threshold hierarchical reasoning (H2T2) strategy for cost-sensitive inference and dynamic offloading to remote models on edge devices for binary classification tasks.
Inference Scaling Law for Retrieval Augmented Generation
Shu Zhou (Nanjing University), Hao Wang (Baidu Inc)
TransformerTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Study the scaling laws of Retrieval-Augmented Generation (RAG) models during the inference phase, systematically evaluate the impact of retriever model size, generator model size, number of retrieved documents, and context window size on performance, and propose joint scaling laws and resource allocation strategies.
Inference-Aware Prompt Optimization for Aligning Black-Box Large Language Models
Saaduddin Mahmud (University of Massachusetts Amherst), Shlomo Zilberstein (University of Massachusetts Amherst)
OptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed the IAPO (Inference-Aware Prompt Optimization) framework, which jointly optimizes prompts and inference scale for black-box LLMs, and designed two training algorithms: PSST (Prompt Scaling via Sequential Trimming) and TopK Screening.
Inference-time Scaling for Diffusion-based Audio Super-resolution
Yizhu Jin (Hong Kong University of Science and Technology), Wei Xue (Hong Kong University of Science and Technology)
Super ResolutionDiffusion modelAudio
🎯 What it does: Propose exploring multiple solution paths in diffusion model audio super-resolution through scaling during inference, utilizing a validator and search algorithms to generate the optimal high-resolution audio.
Inferring Heterogeneous Private Valuations from Offline Market Data via Entropic Risk-Sensitive Utility Maximization
Xingyu Qian (Beijing Institute of Technology), Haoran Yu (Beijing Institute of Technology)
OptimizationRecurrent Neural NetworkReinforcement LearningTabularTime SeriesFinance Related
🎯 What it does: Propose a method to infer heterogeneous private valuations of participants in continuous double auctions through entropy risk-sensitive utility maximization.
Inferring Implicit Goals Across Differing Task Models
Silvia Tulli (Sorbonne University), Sarath Sreedharan (Colorado State University)
Reinforcement Learning
🎯 What it does: Proposed an algorithm to infer and align implicit subgoals across different task models, utilizing bottleneck state identification and query MDP to minimize query costs.
InfiGUI-G1: Advancing GUI Grounding with Adaptive Exploration Policy Optimization
Yuhang Liu (Zhejiang University), Fei Wu (Zhejiang University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningMultimodality
🎯 What it does: Proposed an AEPO framework based on multi-answer generation and adaptive exploration reward to improve semantic alignment of multimodal large language models in GUI localization tasks.
Infinite-Story: A Training-Free Consistent Text-to-Image Generation
Jihun Park (DGIST), Sunghoon Im (DGIST)
GenerationTransformerPrompt EngineeringVision Language ModelImageTextBenchmark
🎯 What it does: Propose Infinite-Story, a consistency text-image generation framework that requires no fine-tuning during training, designed for multi-prompt narrative scenarios, leveraging a scale-wise autoregressive model to achieve identity and style consistency.
InfoCLIP: Bridging Vision-Language Pretraining and Open-Vocabulary Semantic Segmentation via Information-Theoretic Alignment Transfer
Muyao Yuan (Xi'an Jiaotong University), Yudeng Xin (China Telecom)
SegmentationTransformerVision Language ModelImageText
🎯 What it does: Propose InfoCLIP, which utilizes information-theoretic methods to achieve fine-grained pixel-text alignment, compression, and mutual information distillation, enabling efficient fine-tuning for open-vocabulary semantic segmentation.
InfoCom: Kilobyte-Scale Communication-Efficient Collaborative Perception with Information Bottleneck
Quanmin Wei (Southwest Jiaotong University), Xiao Wu (Wuhan University of Technology)
Autonomous DrivingComputational EfficiencyConvolutional Neural Network
🎯 What it does: Propose the InfoCom framework, achieving cooperative perception at kilobyte-level communication through the information bottleneck theory.
InfoDecom: Decomposing Information for Defending Against Privacy Leakage in Split Inference
Ruijun Deng (Fudan University), Qiang Duan (Pennsylvania State University)
Safty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Propose the InfoDecom method, which mitigates data reconstruction attacks and reduces privacy leakage by decomposing redundant information during segmentation inference.
InfoQ: Mixed-Precision Quantization via Global Information Flow
Mehmet Emre Akbulut (Politecnico di Milano), Manuel Roveri (Politecnico di Milano)
ClassificationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Propose InfoQ, a mixed-precision quantization framework based on information flow, which estimates the impact of hierarchical quantization on global information flow through a single forward pass and allocates bit-widths via integer linear programming;