AAAI 2026 Papers — Page 22
AAAI Conference on Artificial Intelligence · 4149 papers
Learning-Augmented Ski Rental with Discrete Distribution: A Bayesian Approach
Bosun Kang (Seoul National University), Chenglin Fan (Seoul National University)
Optimization
🎯 What it does: Proposed a ski rental decision framework based on discrete Bayesian inference, which dynamically updates the posterior using prior distributions and decides when to purchase by comparing the expected remaining rent with the one-time purchase cost.
Length-Adaptive Interest Network for Balancing Long and Short Sequence Modeling in CTR Prediction
Zhicheng Zhang (Tsinghua University), Zhenhua Dong (Huawei)
Recommendation SystemTransformerPrompt EngineeringSequential
🎯 What it does: Proposed the Length-Adaptive Interest Network (LAIN), which explicitly incorporates sequence length information into CTR prediction models to balance long- and short-sequence modeling, significantly enhancing CTR prediction performance.
LENS: Learning to Segment Anything with Unified Reinforced Reasoning
Lianghui Zhu (Huazhong University of Science & Technology), Xinggang Wang (vivo Mobile Communication Co., Ltd)
SegmentationTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose the LENS framework to achieve segmentation tasks under text prompts, using a unified reinforcement learning strategy for chained reasoning and segmentation optimization during testing.
Less Is Better: Sparse Instance Learning for Cross-Domain Few-Shot Object Detection
Yali Huang (Zhengzhou University), Hichem Sahbi (Sorbonne University)
Object DetectionDomain AdaptationMeta LearningTransformerContrastive LearningImage
🎯 What it does: Proposed the SI-ViTO framework to achieve sparse instance learning for cross-domain few-shot object detection tasks
Less Is More: Rethinking Parameter-Efficient Fine-Tuning from a Subtractive Perspective
Tianqi Jiang (Tianjin University), Qinghua Hu (University of Electronic Science and Technology of China)
ClassificationComputational EfficiencyImage
🎯 What it does: Proposes a parameter-efficient fine-tuning method called SFP from a subtractive perspective, which replaces redundant or interfering modules in pre-trained models with lightweight filter blocks initialized via pseudo-inverse. Only the filter block is fine-tuned while the remaining parameters are frozen, significantly improving task adaptation performance.
Less Is More: Sparse and Cooperative Perturbation for Point Cloud Attacks
Keke Tang (Guangzhou University), Zhihong Tian (Guangzhou University)
ClassificationAdversarial AttackPoint Cloud
🎯 What it does: Proposed a sparse and collaborative perturbation framework SCP to generate efficient adversarial examples on point cloud classification models, emphasizing the collaborative effect of point sets rather than single-point perturbation;
Less Is More: Vision Representation Compression for Efficient Video Generation with Large Language Models
Yucheng Zhou (University of Macau), Yu Cheng (Chinese University of Hong Kong)
GenerationConvolutional Neural NetworkTransformerLarge Language ModelVideoText
🎯 What it does: This paper addresses the issue of video token redundancy in large language models (LLM) during autoregressive video generation, proposing the Vision Representation Compression (VRC) framework. VRC introduces learnable compressors and decompressors in the video token sequence, compressing the video representation into a shorter sequence, performing autoregressive prediction in the compressed space, and avoiding error accumulation through next-segment prediction.
Let the Model Learn to Feel: Mode-Guided Tonality Injection for Symbolic Music Emotion Recognition
Haiying Xia (Guangxi Normal University), Shuxiang Song (Guangxi Normal University)
RecognitionTransformerSupervised Fine-TuningSequential
🎯 What it does: In symbolic music emotion recognition, this paper evaluates MIDIBERT's insufficient perception of musical modes (major/minor), proposes a mode-oriented diagnostic method MoGE, and designs the MoFi framework, which leverages FiLM to inject mode knowledge into the Transformer's lower layers to enhance emotion recognition performance.
Let the Void Be Void: Robust Open-Set Semi-Supervised Learning via Selective Non-Alignment
You Rim Choi (Seoul National University), Hyung-Sin Kim (Seoul National University)
ClassificationAnomaly DetectionContrastive LearningImage
🎯 What it does: Propose the SkipAlign framework, combining selective non-alignment to achieve open-set semi-supervised learning, enabling simultaneous improvement in ID classification and OOD detection.
Let’s Think with Images Efficiently! An Interleaved-Modal Chain-of-Thought Reasoning Framework with Dynamic and Precise Visual Thoughts
Xu Liu (Harbin Institute of Technology), Libo Qin (Central South University)
SegmentationComputational EfficiencyLarge Language ModelVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose a new Interleaved-Modal Chain-of-Thought framework called DAP-ICOT, which achieves more efficient multimodal reasoning through dynamic visual thinking integration and precise visual thinking guidance.
Lethe: Layer- and Time-Adaptive KV Cache Pruning for Reasoning-Intensive LLM Serving
Hui Zeng (Xidian University), Jidong Zhai (Xidian University)
Computational EfficiencyLarge Language ModelText
🎯 What it does: Propose the Lethe framework, which dynamically prunes the KV cache during inference along both hierarchical and temporal dimensions to reduce memory and latency in long-chain reasoning.
Leveraging Dissimilarity Invariance as a Robust Anchor for Learning with Noisy Labels
Wenxiao Fan (Beijing Institute of Technology), Kan Li (Beijing Institute of Technology)
ClassificationContrastive LearningImage
🎯 What it does: Propose the NegScale framework, which leverages negative similarity invariant under label noise as robust anchor points to enhance model learning effectiveness under noisy labels.
Leveraging Failed Samples: A Few-Shot and Training-Free Framework for Generalized Deepfake Detection
Shibo Yao (Beijing Jiaotong University), Chunjie Zhang (Chinese Academy of Sciences)
Anomaly DetectionTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Propose FTNet, a training-free framework that achieves deepfake detection with only a minimal number of samples (just one synthetic sample), leveraging CLIP intermediate layer features to construct a Key-Value cache and discriminates test samples via nearest neighbor classification;
Leveraging Image as Compressed Visual Prompt and Hierarchical Visual Knowledge for Effective Image Utilization in MLLMs
Shezheng Song, Jie Yu (National University Of Defense Technology)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose an IPK mechanism that integrates compressed visual prompts and hierarchical visual knowledge to enhance the utilization efficiency of images in multimodal large language models.
Leveraging Textual Compositional Reasoning for Robust Change Captioning
Kyu Ri Park (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
GenerationTransformerPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This study proposes a change description framework named CORTEX that leverages text composition reasoning enhancement, aiming to improve the recognition and description of fine-grained changes between image pairs through explicit text prompts.
Leveraging Visual Blur Perception Characteristics for EEG Decoding
Wenchao Liu (Harbin Institute of Technology), Haifeng Li (Harbin Institute of Technology)
RetrievalRepresentation LearningConvolutional Neural NetworkContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose a visual decoding framework for EEG based on visual blur perception features, leveraging multi-level Gaussian blur and feature selection to construct personalized visual representations, and employing a forward-constrained spatial convolution EEG encoder with CLIP for contrastive learning.
LexChain: Modeling Legal Reasoning Chains for Chinese Tort Case Analysis
Huiyuan Xie (Tsinghua University), Zhiyuan Liu (Northeastern University)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the LexChain framework and evaluation benchmark to systematically model legal reasoning chains in Chinese civil tort cases, and evaluate and improve various large language models.
LexInstructEval: Lexical Instruction Following Evaluation for Large Language Models
Huimin Ren (Li Auto Inc.), Chen Wei (Li Auto Inc.)
TextBenchmark
🎯 What it does: Proposes LexInstructEval, a benchmark evaluation framework for fine-grained lexical instruction following in large language models.
LGAN: An Efficient High-Order Graph Neural Network via the Line Graph Aggregation
Lin Du, Zhao Li (Beijing Normal University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a novel graph neural network called Line Graph Aggregation Network (LGAN), which constructs a line graph on the 1-hop subgraph of each node and performs two types of aggregation: target-neighbor and neighbor-neighbor, to achieve graph-level representation learning.
Libra-MIL: Multimodal Prototypes Stereoscopic Infused with Task-specific Language Priors for Few-shot Whole Slide Image Classification
Zhenfeng Zhuang (Xiamen University), Liansheng Wang (Xiamen University)
ClassificationLarge Language ModelMultimodalityBiomedical Data
🎯 What it does: Propose Libra-MIL, a multi-instance learning framework integrating task-specific text priors, bimodal prototype learning, and 3D optimal transport for whole-slide image classification under few annotations.
LiDAR-GS++: Improving LiDAR Gaussian Reconstruction via Diffusion Priors
Qifeng Chen (Alibaba Group), Sheng Yang (Zhejiang University)
Data SynthesisAutonomous DrivingDiffusion modelAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: Propose the LiDAR-GS++ method, which incorporates a diffusion prior into Gaussian Splatting to achieve real-time high-quality LiDAR relighting, particularly improving the synthesis effect of extrapolated viewpoints.
LiDARCrafter: Dynamic 4D World Modeling from LiDAR Sequences
Alan Liang (National University of Singapore), Wei Tsang Ooi (National University of Singapore)
GenerationAutonomous DrivingLarge Language ModelDiffusion modelPoint Cloud
🎯 What it does: Proposes LiDARCrafter, a controllable 4D LiDAR generation and editing framework capable of generating dynamic LiDAR sequences containing geometric, motion, and structural priors based on natural language instructions, while supporting scene editing.
LidarPainter: One-Step Away from Any Lidar View to Novel Guidance
Yuzhou Ji, Xin Tan (51WORLD)
GenerationAutonomous DrivingDiffusion modelGaussian SplattingImageTextPoint Cloud
🎯 What it does: Proposed a first-order diffusion model called LidarPainter, which generates new perspectives of driving scenes consistent with sparse LiDAR conditions and distorted rendering images, significantly improving the quality of dynamic driving scene reconstruction.
LifeAlign: Lifelong Alignment for Large Language Models with Memory-Augmented Focalized Preference Optimization
Junsong Li (East China Normal University), Liang He (East China Normal University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposed the LifeAlign framework, aiming to enable large language models to achieve lifelong alignment, continuously adapting to multi-task environments without experiencing catastrophic forgetting.
Lifelong Domain Adaptive 3D Human Pose Estimation
Qucheng Peng (University of Central Florida), Chen Chen (University of Central Florida)
Pose EstimationDomain AdaptationConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a lifetime-adaptive 3D human pose estimation framework that can sequentially adapt to multiple target domains by only accessing current target domain data, while retaining memory of previous domains.
Lifelong Language-Conditioned Robotic Manipulation Learning
Xudong Wang (Shenyang Institute of Automation, Chinese Academy of Sciences), Zhi Han (Shenyang Institute of Automation, Chinese Academy of Sciences)
Robotic IntelligenceReinforcement LearningMixture of ExpertsTextBenchmark
🎯 What it does: Propose a language-conditioned robotic manipulation framework for lifelong learning, enabling continuous acquisition of multiple skills while mitigating catastrophic forgetting.
Light but Sharp: SlimSTAD for Real-Time Action Detection from Sensor Data
Wei Cui (Institute for Infocomm Research, Agency for Science, Technology and Research), Bing Li (University of Electronic Science and Technology of China)
Object DetectionComputational EfficiencyConvolutional Neural NetworkGraph Neural NetworkTime SeriesSequential
🎯 What it does: Proposes SlimSTAD, a lightweight framework specifically designed for sensor-based temporal action detection, enabling high-accuracy, low-latency real-time action localization and classification on edge devices.
Light-IF: Endowing LLMs with Generalizable Reasoning via Preview and Self-Checking for Complex Instruction Following
Chenyang Wang (Harbin Institute of Technology), Liang Xu (Chinese Language Understanding Evaluation)
Supervised Fine-TuningReinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the Light-IF framework, which enhances LLMs' compliance with complex instructions through preview and self-inspection thinking modes;
Lightning Fast Caching-based Parallel Denoising Prediction for Accelerating Talking Head Generation
Jianzhi Long (University of Sydney), Dacheng Tao (Nanyang Technological University)
GenerationComputational EfficiencyConvolutional Neural NetworkDiffusion modelScore-based ModelVideo
🎯 What it does: Proposes a training-agnostic acceleration framework called LightningCP, which significantly speeds up speaker avatar generation based on diffusion models by leveraging caching and parallel inference, and reduces the computational complexity of attention mechanisms through Decoupled Foreground Attention (DFA).
Lightweight Adaptive Topological Layout and Semantic Mapping in Vision-and-Language Navigation on Websites
Pingrui Lai (Shanghai Jiao Tong University), Hua Yang (Shanghai Jiao Tong University)
Computational EfficiencyConvolutional Neural NetworkGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision-Language-Action ModelMultimodalityGraph
🎯 What it does: Propose a lightweight framework ATLAS that dynamically constructs web topology graphs and unifies semantic representations for Vision-and-Language web navigation and question answering.
Lightweight Optimal-Transport Harmonization on Edge Devices
Maria Larchenko, Georgy Derevyanko (Glam AI)
Image HarmonizationConvolutional Neural NetworkImage
🎯 What it does: Developed a lightweight color harmonization method called MKL-Harmonizer that can achieve real-time AR object color matching on edge devices.
LILAD: Learning In-context Lyapunov-stable Adaptive Dynamics Models
Amit Jena (Texas A&M University), Le Xie (Harvard University)
OptimizationMeta LearningTransformerLarge Language ModelSequentialBenchmarkPhysics Related
🎯 What it does: Proposed the LILAD framework, combining In-Context Learning (ICL) with Lyapunov stability to achieve adaptive and stable system identification.
Linear Time Algorithms for Individually Fair k-means via Multi-Swap Local Search
Beirong Cui (Central South University), Junyu Huang (Central South University)
OptimizationTabular
🎯 What it does: Propose a linear-time multi-swap local search algorithm and design a collaborative initialization framework for k-means clustering under individual fairness constraints.
LiNeXt: Revisiting LiDAR Completion with Efficient Non-Diffusion Architectures
Wenzhe He (Hunan University), Kenli Li (Hunan University)
Autonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Propose a lightweight non-diffusion point cloud completion network LiNeXt, achieving real-time 3D LiDAR scene completion
LinProVSR: Linguistics-Knowledge Guided Progressive Disambiguation Network for Visual Speech Recognition
Feng Xue (Hefei University of Technology), Dan Guo (Hefei University of Technology)
RecognitionConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Propose a fuzzy sample construction and evolutionary contrastive disambiguation network based on language priors to address phoneme-visual ambiguity in visual speech recognition.
LiR3AG: A Lightweight Rerank Reasoning Strategy Framework for Retrieval-Augmented Generation
Guo Chen, Tao Jia (Communication University of China)
RetrievalComputational EfficiencyKnowledge DistillationTransformerPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Systematic analysis of retrieval-augmented generation (RAG) in multi-hop question answering, revealing two primary reasoning strategies of reasoning models (context-based reasoning and internal knowledge-based reasoning), and designing a lightweight LIR-AG framework (Retriever-Reranker-Reasoning Constructor) to replicate and improve these reasoning strategies on non-reasoning LLMs.
Listen like a Teacher: Mitigating Whisper Hallucinations Using Adaptive Layer Attention and Knowledge Distillation
Kumud Tripathi (Sony Research India), Pankaj Wasnik (Sony Research India)
RecognitionKnowledge DistillationTransformerAudio
🎯 What it does: Reduce hallucination errors in noisy environments and improve semantic accuracy in the Whisper model using a two-stage approach.
Listening Between the Frames: Bridging Temporal Gaps in Large Audio-Language Models
Hualei Wang (Beijing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology, Chinese Academy of Sciences), Xiangdong Wang (Beijing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology, Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelTextMultimodalityAudio
🎯 What it does: Propose TimeAudio, an improved large audio-language model focusing on fine-grained temporal localization and long audio understanding.
LiteGE: Lightweight Geodesic Embedding for Efficient Geodesics Computation and Non-Isometric Shape Correspondence
Yohanes Yudhi Adikusuma (University of Texas at Austin), Ying He (Nanyang Technological University)
Computational EfficiencyRepresentation LearningPoint CloudMesh
🎯 What it does: Propose LiteGE, which uses PCA to compress the Unsigned Distance Field (UDF) and obtain a lightweight shape descriptor, directly predicting geodesic distances on 3D surfaces and achieving shape correspondence.
LiteLong: Resource-Efficient Long-Context Data Synthesis for LLMs
Junlong Jia (Beihang University), Binghui Guo (Xiaohongshu)
Data SynthesisComputational EfficiencyTransformerLarge Language ModelAgentic AITextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a resource-efficient long-text data synthesis framework called LiteLong, which utilizes the BISAC classification structure and a multi-agent debate mechanism to generate diverse topics, then retrieves and concatenates them into 128K-token training samples using BM25.
LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding
Xiaodong Wang (Peking University), Peixi Peng (Peking University)
Large Language ModelSupervised Fine-TuningAgentic AIVision Language ModelMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed LiViBench, an interactive live video full-modal evaluation benchmark covering audio, speech, and real-time comments, constructed with 3,168 videos and 3,175 multiple-choice questions based on a semi-automated annotation workflow.
LLA: Enhancing Security and Privacy for Generative Models with Logic-Locked Accelerators
You Li (Northwestern University), Hai Zhou (Northwestern University)
GenerationSafty and PrivacyTransformerText
🎯 What it does: Proposes LLA, a generative model locking framework that integrates software and hardware, preventing model theft, tampering, and information leakage by embedding keys in neurons and implementing lightweight logic locking on hardware accelerators.
LLaVA-MS-PIT: Multi-Modal Schema-Guided Progressive Instruction Tuning for Multi-Modal Event Extraction
Hui Zhang (Central China Normal University), Wei Emma Zhang (University of Adelaide)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: Proposes the LLaVA-MS-PIT framework, which achieves multimodal event extraction through progressively instruction-tuned multimodal event patterns.
LLaVA-UHD v2: Exploiting Hierarchical Vision Granularity in MLLMs via Inverse Semantic Pyramid
Yipeng Zhang (Tsinghua University), Maosong Sun (National University Of Singapore)
TransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose Hiwin Transformer for enhancing hierarchical visual feature representation in multimodal large language models, constructing an inverse semantic pyramid and compressing features using hierarchical window attention.
LLaVA³: Representing 3D Scenes Like a Cubist Painter to Boost 3D Scene Understanding of VLMs
Doriand Petit (Université Paris-Saclay), Loïc Barthe (Université Paris-Saclay)
Object DetectionSegmentationRepresentation LearningVision Language ModelNeural Radiance FieldPoint CloudMesh
🎯 What it does: Propose LLaVA 3, which enables frozen VLMs to achieve 3D scene understanding on multi-view images by decomposing 3D scenes into object hierarchies and generating 360-degree 2D visual descriptions.
LLM Collaboration with Multi-Agent Reinforcement Learning
Shuo Liu (Northeastern University), Christopher Amato (Northeastern University)
OptimizationTransformerLarge Language ModelReinforcement LearningAgentic AIText
🎯 What it does: Proposed and implemented a multilingual model collaboration framework based on collaborative reinforcement learning, using the MAGRPO algorithm to jointly fine-tune multiple models, enabling them to effectively collaborate in writing and coding tasks to generate high-quality content.
LLM Collaborative Filtering: User-Item Graph as New Language
Huachi Zhou (Hong Kong Polytechnic University), Xiao Huang (Hong Kong Polytechnic University)
Recommendation SystemComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningGraph
🎯 What it does: Convert the user-item graph into a 'new language,' generate graph sentences via random walks, compress the corpus using similarity sampling, fine-tune on an LLM, and aggregate hidden representations to obtain user and item embeddings for efficient recommendation.
LLM-Aligned Geographic Item Tokenization for Local-Life Recommendation
Hao Jiang (Kuaishou Technology), Guorui Zhou (Independent Researcher)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningPrompt EngineeringContrastive LearningTextTabularSequential
🎯 What it does: Proposed the LGSID framework, which first aligns the geographic knowledge of large language models (LLMs) through reinforcement learning, and then achieves hierarchical geographic project tokenization to enhance geographic perception and semantic representation in local life recommendations;
LLM-CAS: Dynamic Neuron Perturbation for Real-Time Hallucination Correction
Jusheng Zhang (Sun Yat-sen University), Keze Wang (Sun Yat-sen University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Train a hierarchical reinforcement learning agent to apply temporary dynamic perturbations to neuron activations of large language models (LLMs) in real-time during inference, correcting hallucinated content in generated outputs.
LLM-Enhanced Energy Contrastive Learning for Out-of-Distribution Detection in Text-Attributed Graphs
Xiaoxu Ma (Tianjin University), Chen Zhao (Baylor University)
ClassificationAnomaly DetectionGraph Neural NetworkLarge Language ModelContrastive LearningTextGraphChain-of-Thought
🎯 What it does: Proposed a framework named LECT that integrates large language models (LLM) with energy contrastive learning for node-level discrete detection and classification on text-attribute graphs.
LLM-Free Image Captioning Evaluation in Reference-Flexible Settings
Shinnosuke Hirano (Keio University), Komei Sugiura (Keio University)
Computational EfficiencyRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed a LLM-free supervised image caption evaluation metric called Pearl, capable of unified assessment in both reference-based and reference-free settings.
LLM-Guided Quantified SMT Solving over Uninterpreted Functions
Kunhang Lv (SKLCS and Key Laboratory of System Software, ISCAS), Jian Zhang (SKLCS and Key Laboratory of System Software, ISCAS)
OptimizationComputational EfficiencyLarge Language ModelPrompt EngineeringBenchmarkChain-of-Thought
🎯 What it does: Leverage large language models (LLMs) to provide semantic guidance for quantified uninterpreted functions (UF) in nonlinear real arithmetic formulas, generating more effective instantiations and significantly improving SMT solving efficiency.
LLM-Oriented Token-Adaptive Knowledge Distillation
Xurong Xie (Zhejiang University), Jiangning Zhang (Zhejiang University)
Knowledge DistillationLarge Language ModelText
🎯 What it does: Proposed an adaptive knowledge distillation framework called AdaKD based on tokens, which dynamically adjusts the distillation process by leveraging the real-time learning difficulty of each token.
LLM2CLIP: Powerful Language Model Unlocks Richer Cross-Modality Representation
Weiquan Huang (Tongji University), Liang Hu (Tongji University)
RetrievalRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Integrate large language models (LLMs) into CLIP by first using caption contrastive learning to convert the LLM into a text encoder, then fine-tuning the pre-trained CLIP with a lightweight adapter to enhance cross-modal performance.
LLMC+: Benchmarking Vision-Language Model Compression with a plug-and-play Toolkit
Chengtao Lv (Nanyang Technological University), Wenya Wang (SenseTime Research)
CompressionTransformerImageVideoTextMultimodalityBenchmark
🎯 What it does: Propose LLMC+, a pluggable compression benchmark and toolbox supporting over 20 algorithms and five VLM families, with systematic evaluation of compression effectiveness in spatial/temporal redundancy, fine-grained tasks, and multi-turn dialogues.
LLMdoctor: Token-Level Flow-Guided Preference Optimization for Efficient Test-Time Alignment of Large Language Models
Tiesunlong Shen (Nanyang Technological University), Erik Cambria (Xi'an Jiaotong University)
OptimizationComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningFlow-based ModelContrastive LearningText
🎯 What it does: Propose a new test-time alignment framework called LLMdoctor, which utilizes token-level rewards and flow-guided preference optimization (TFPO) to real-time guide frozen large language models;
LLMs Unleashed: Generating Protocol Code from RFC Specifications
Junfeng Long, Biao Han (National University of Defense Technology)
GenerationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Implementing a fully automatic generation system from RFC documents to compilable protocol code based on large language models
LLMTM: Benchmarking and Optimizing LLMs for Temporal Motif Analysis in Dynamic Graphs
Bing Hao (Tianjin University), Ruijie Wang (Beihang University)
OptimizationTransformerAgentic AIPrompt EngineeringGraphBenchmarkChain-of-Thought
🎯 What it does: Propose the LLMTM benchmark to evaluate the ability of LLMs in tasks such as identifying, detecting, and constructing temporal motifs in dynamic graphs, and develop tool-enhanced LLM agents with structure-aware schedulers to achieve self-adaptive balance between accuracy and computational cost.
LMGL-WD: LLM-Guided Multi-Task Graph Learning for Category-Level Warehouse Demand Prediction in E-Commerce
Wenjun Lyu (JD Logistics), Desheng Zhang (JD Logistics)
OptimizationGraph Neural NetworkTransformerLarge Language ModelGraphTime Series
🎯 What it does: Proposed an LLM-guided multi-task graph learning framework, LMGL-WD, for category demand forecasting at the e-commerce warehouse level.
Local Guidance for Configuration-Based Multi-Agent Pathfinding
Tomoki Arita (National Institute of Advanced Industrial Science and Technology), Keisuke Okumura (National Institute of Advanced Industrial Science and Technology)
OptimizationBenchmark
🎯 What it does: Propose introducing local guidance in configuration-based multi-agent path planning, improving the LaCAM algorithm to enhance solution quality and real-time performance.
Localization-Anchored Instance Discrimination for Domain Adaptive Person Search
Linfeng Qi, Jiqing Zhang (Hebei University)
Object DetectionRetrievalDomain AdaptationConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: Proposed a location-anchor-based instance identification framework LAID for domain adaptation in person search tasks.
LOG-Nav: Efficient Layout-Aware Object-Goal Navigation with Hierarchical Planning
Jiawei Hou (Fudan University), Taiping Zeng (Fudan University)
Computational EfficiencyRepresentation LearningRobotic IntelligenceTransformerLarge Language ModelVision Language ModelNeural Radiance FieldSimultaneous Localization and MappingMultimodalityPoint Cloud
🎯 What it does: Propose LOG-Nav, a hierarchical planning framework for layout-aware object navigation, which enables efficient path planning and execution from global to local levels in unexplored multi-room indoor environments using LLM-driven agents.
Logic Unseen: Revealing the Logical Blindspots of Vision-Language Models
Yuchen Zhou (Sun Yat-sen University), Tat-Seng Chua (National University of Singapore)
TransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: This paper proposes LogicBench benchmark and LogicCLIP training framework to evaluate and enhance the understanding of logical relationships in vision-language models (VLM).
LoGIC: Multi-LoRA Guided Importance Consensus for Multi-Task Pruning in Vision Transformers
Yu-Hong Chou (National Taiwan University), Ming-Syan Chen (National Taiwan University)
ClassificationObject DetectionSegmentationComputational EfficiencyTransformerImageBenchmark
🎯 What it does: Proposes the LoGIC framework for joint structured sparsification of shared trunk and multi-task LoRA modules in multi-task Vision Transformers (ViT), enabling efficient deployment.
Logical Characterizations of GNNs with Mean Aggregation
Moritz Schönherr (Leipzig University), Carsten Lutz (Leipzig University)
Explainability and InterpretabilityRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper studies the expressive power of graph neural networks (GNNs) using arithmetic mean (mean) as the aggregation function, and provides its correspondence with various modal logics under different settings (non-uniform and uniform).
LogicCat: A Chain-of-Thought Text-to-SQL Benchmark for Complex Reasoning
Liutao (Zhengzhou University), Min Peng (Wuhan University)
TransformerPrompt EngineeringTextTabularBenchmarkChain-of-Thought
🎯 What it does: Developed the LogicCat text-to-SQL benchmark, focusing on multi-step reasoning and chain-of-thought (CoT) thinking;
LoGoSeg: Integrating Local and Global Features for Open-Vocabulary Semantic Segmentation
Junyang Chen (Southeast University), Yiguo Qiao (Southeast University)
SegmentationTransformerVision Language ModelImageMultimodality
🎯 What it does: Proposed a single-stage open-vocabulary semantic segmentation framework called LoGoSeg, integrating object prior, region alignment, and dual-stream feature fusion to enhance pixel-level text alignment and segmentation accuracy.
LoKI: Low-Damage Knowledge Implanting of Large Language Models
Runyu Wang, Tianbo Ji (South China University Of Technology)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose LoKI, a parameter-efficient fine-tuning framework based on the internal knowledge storage mechanism of Transformers, which identifies low-impact weights through knowledge vector attribution and fine-tunes these weights in a layer-balanced manner;
Long-form RewardBench: Evaluating Reward Models for Long-form Generation
Hui Huang (Harbin Institute of Technology), Tiejun Zhao (Harbin Institute of Technology)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose Long‑form RewardBench—the first benchmark for evaluating reward models in long-text generation, covering five subtasks: QA, RAG, Chat, Writing, and Reasoning; simultaneously design the Long‑form Needle‑in‑a‑Haystack test to investigate the impact of error positions and lengths on reward models.
LongLLaDA: Unlocking Long Context Capabilities in Diffusion LLMs
Xiaoran Liu (Fudan University), Xipeng Qiu (Fudan University)
GenerationRetrievalTransformerLarge Language ModelDiffusion modelTextBenchmark
🎯 What it does: Systematically evaluate the performance of diffusion-based large language models (dLLMs) in long contexts and propose a training-free context extension method called LongLLaDA
LongSplat: Online Generalizable 3D Gaussian Splatting from Long Sequence Images
Guichen Huang (Beijing Institute of Technology), Yunde Jia (Shenzhen MSU-BIT University)
Computational EfficiencyTransformerGaussian SplattingImageSequential
🎯 What it does: Propose LongSplat, an online real-time 3D Gaussian splatting framework designed for long-sequence images;
LongT2IBench: A Benchmark for Evaluating Long Text-to-Image Generation with Graph-structured Annotations
Zhichao Yang (Xidian University), Leida Li (Xidian University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextGraphBenchmarkChain-of-Thought
🎯 What it does: Constructed a long-text-image generation alignment evaluation benchmark, LongT2IBench (14K long-text-image pairs with graph-structured annotations), and proposed the LongT2IExpert evaluator, which leverages a multimodal large language model to achieve alignment scoring and structured explanations through hierarchical chain-of-thought reasoning;
Look as You Think: Unifying Reasoning and Visual Evidence Attribution for Verifiable Document RAG via Reinforcement Learning
Shuochen Liu (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
RetrievalExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the Chain-of-Evidence (CoE) reasoning paradigm and the Look-As-You-Think (LAT) reinforcement learning framework to achieve verifiable step-by-step visual evidence attribution for Visual Document Retrieval-Augmented Generation (VD-RAG).
Look Closer! An Adversarial Parametric Editing Framework for Hallucination Mitigation in VLMs
Jiayu Hu (Chongqing University), Zhongshi He (Chongqing University)
Explainability and InterpretabilityAdversarial AttackPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed and implemented a framework called ALEAHallu based on adversarial parameter editing, aimed at reducing hallucinations (content inconsistent with visual inputs) generated by visual language models when producing text.
Look-Back: Implicit Visual Re-focusing in MLLM Reasoning
Shuo Yang (Peking University), Li Yuan (Peking University)
Large Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: In this study, the authors propose an implicit visual refocusing method called Look‑Back, which leverages multimodal large language models (MLLM) to automatically refocus image information during reasoning without the need for explicitly injecting visual prompts;
LookFlow: Training-Free and Efficient High-Resolution Image Synthesis via Dynamic Lookahead Guidance Flow
Yuan Zhou (Chinese Academy of Sciences), Shiming Xiang (Chinese Academy of Sciences)
GenerationData SynthesisSuper ResolutionComputational EfficiencyTransformerFlow-based ModelRectified FlowImageTextMultimodalityOrdinary Differential Equation
🎯 What it does: This paper proposes a training-free high-resolution image synthesis framework called LookFlow, which can rapidly and high-quality reconstruct low-resolution images into high-resolution images without additional training.
LoopLLM: Transferable Energy-Latency Attacks in LLMs via Repetitive Generation
Xingyu Li (National Interdisciplinary Research Center of Engineering Physics), Jia-Li Yin (Fuzhou University)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes LoopLLM, an energy-latency attack framework that induces large language models (LLMs) to generate repetitive content to maximize energy consumption and latency.
LoRA in LoRA: Towards Parameter-Efficient Architecture Expansion for Continual Visual Instruction Tuning
Chang Che (Hefei University Of Technology), Zenglin Shi (Tsinghua University)
Computational EfficiencyKnowledge DistillationRepresentation LearningTransformerMultimodalityBenchmark
🎯 What it does: Proposes the LiLoRA method, achieving parameter-efficient structural expansion in Continuous Visual Instruction Tuning (CVIT) by sharing the LoRA matrix A, decomposing B, and introducing a cosine regularization-based base stability loss.
LORETTA: A Low Resource Framework to Poison Continuous Time Dynamic Graphs
Himanshu Pal (International Institute of Information Technology), Charu Sharma (International Institute of Information Technology)
Adversarial AttackGraph
🎯 What it does: Proposes LORETTA, a model-free gradient, low-resource continuous-time dynamic graph poisoning framework that first sparsifies important edges and then inserts adversarial edges using degree-preserving negative sampling;
Loss-Guided Auxiliary Agents for Overcoming Mode Collapse in GFlowNets
Idriss Malek (Mohamed Bin Zayed University of Artificial Intelligence), Salem Lahlou (Mohamed Bin Zayed University of Artificial Intelligence)
GenerationReinforcement LearningFlow-based ModelText
🎯 What it does: In this work, the authors propose a novel auxiliary GFlowNet training framework called LGGFN, which leverages the training loss of the main model to directly drive the exploration of the auxiliary agent, aiming to overcome the mode collapse problem in GFlowNet training.
Lost in Benchmarks? Rethinking Large Language Model Benchmarking with Item Response Theory
Hongli Zhou (Harbin Institute of Technology), Tiejun Zhao (Harbin Institute of Technology)
TextBenchmark
🎯 What it does: This paper proposes the PSN-IRT framework based on a pseudo twin network to evaluate the performance of large language models on various benchmark tests and conduct systematic analysis across 11 mainstream LLM benchmarks.
Lost in Time? A Meta-Learning Framework for Time-Shift-Tolerant Physiological Signal Transformation
Qian Hong (Renmin University of China), Zijing Zeng (OPPO)
Meta LearningConvolutional Neural NetworkTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: Propose a meta-learning framework called ShiftSyncNet that uses a time-shift correction network to perform waveform transformation on asynchronous physiological signals.
Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-tuning
Hyunseok Seung (University of Wisconsin - Madison), Hyunsuk Ko (Hanyang University)
OptimizationComputational EfficiencyTransformerSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Proposed a zeroth-order optimization method called LOREN for LLM fine-tuning, which can significantly reduce memory usage and accelerate convergence without using backpropagation.
LPPG-RL: Lexicographically Projected Policy Gradient Reinforcement Learning with Subproblem Exploration
Ruiyu Qiu (Zhejiang University), Zhijiang Shao (Zhejiang University)
OptimizationReinforcement Learning
🎯 What it does: Proposed the LPPG-RL framework for lexicographic multi-objective reinforcement learning (Lexicographic Multi-Objective RL) in continuous spaces, achieving strict satisfaction of priority constraints through gradient projection and subproblem exploration.
LR-AdaInSeg:Adaptive Instance Segmentation of Incomplete 3D Scenes Driven by Low-Rank Networks
Qin Zhang (Ningbo University), Xulun Ye (Ningbo University)
SegmentationDomain AdaptationComputational EfficiencyConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: This work proposes LR-AdaInSeg, an adaptive instance segmentation framework for unseen classes in incomplete 3D scenes.
LSAP-PV: High-Fidelity Palm Vein Image Synthesis via Layered Spectral Absorption Projection-Guided Diffusion Model
Sheng Shang (Hefei University Of Technology), Wei Jia (Hefei University Of Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Propose a 3D vascular tree model based on multi-layer spectral absorption projection (LSAP) and combine it with a conditional diffusion model to generate high-fidelity finger and palm vein images.
LSD-3D: Large-Scale 3D Driving Scene Generation with Geometry Grounding
Julian Ost (Princeton University), Felix Heide (Princeton University)
GenerationData SynthesisAutonomous DrivingDiffusion modelScore-based ModelGaussian SplattingTextMultimodalityPoint Cloud
🎯 What it does: Propose a geometry-guided diffusion distillation method capable of generating large-scale, controllable, and high-quality 3D driving scenes with accurate geometry and textures from text, maps, and other conditions.
LSHFed: Robust and Communication-Efficient Federated Learning with Locally-Sensitive Hashing Gradient Mapping
Guanjie Cheng (Zhejiang University), Shuiguang Deng (Zhejiang University)
ClassificationCompressionFederated LearningSafty and PrivacyImage
🎯 What it does: Propose a federated learning framework named LSHFed, which achieves malicious gradient detection and communication compression through LSH gradient mapping while ensuring gradient privacy.
LUCID: Learning-Enabled Uncertainty-Aware Certification of Stochastic Dynamical Systems
Ernesto Casablanca (Newcastle University), Sadegh Soudjani (Max Planck Institute for Software Systems)
Autonomous DrivingOptimizationBenchmark
🎯 What it does: Propose the LUCID tool, which learns control barrier certificates from limited transfer data of black-box systems to quantify safety guarantees.
LUMIN: A Longitudinal Multi-modal Knowledge Decomposition Network for Predicting Breast Cancer Recurrence
Chunyao Lu (Netherlands Cancer Institute), Ritse Mann (Netherlands Cancer Institute)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerContrastive LearningMultimodalityTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: Developed a longitudinal multi-modal knowledge decomposition network called LUMIN, utilizing follow-up mammograms and electronic health records (EHR) to predict breast cancer recurrence.
LungNoduleAgent: A Collaborative Multi-Agent System for Precision Diagnosis of Lung Nodules
Cheng Yang (Hangzhou Dianzi University), Ruiquan Ge (Chongqing University of Posts and Telecommunications)
ClassificationObject DetectionAgentic AIPrompt EngineeringMixture of ExpertsVision Language ModelImageTextComputed TomographyRetrieval-Augmented Generation
🎯 What it does: Propose a collaborative multi-agent system named LungNoduleAgent for precise diagnosis of lung nodules in lung CT scans, covering three modules: nodule localization, local report generation, and malignancy assessment.
LWGANet: Addressing Spatial and Channel Redundancy in Remote Sensing Visual Tasks with Light-Weight Grouped Attention
Wei Lu (Anhui University), Si-Bao Chen (Anhui University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose a lightweight remote sensing visual backbone called LWGANet, which uniformly models and optimizes spatial and channel redundancies in remote sensing images.
M-Loss: Quantifying Model Merging Compatibility with Limited Unlabeled Data
Tiantong Wang (Nanyang Technological University), Wei Yang Bryan Lim (Nanyang Technological University)
Computational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a metric based on unlabeled data called M-Loss, which quantifies the gap between parameter average merging and output average ensembling, and uses it to guide model merging and pruning;
M2FMoE: Multi-Resolution Multi-View Frequency Mixture-of-Experts for Extreme-Adaptive Time Series Forecasting
Yaohui Huang (Central South University), Ruipeng Dong (Central South University)
Mixture of ExpertsTime Series
🎯 What it does: Proposes an extreme event adaptive time series prediction model called M2FMoE, which captures differences between regular and extreme patterns through multi-resolution, multi-perspective frequency Mixture-of-Experts;
M2I2: Learning Efficient Multi-Agent Communication via Masked State Modeling and Intention Inference
Chuxiong Sun (Chinese Academy of Sciences), Wei Wang (Beijing University of Posts and Telecommunications)
Computational EfficiencyMeta LearningTransformerReinforcement LearningAuto EncoderBenchmark
🎯 What it does: Propose a multi-agent communication framework named M2I2, which employs masked state modeling (MAE) and inverse models (predicting joint actions) for self-supervised learning, enhancing message integration and partner intent inference.
M²VAE: Multi-Modal Multi-View Variational Autoencoder for Cold-start Item Recommendation
Chuan He (Zhejiang University of Technology), Xin-Wei Yao (Zhejiang University of Technology)
Recommendation SystemMixture of ExpertsAuto EncoderContrastive LearningMultimodality
🎯 What it does: Propose a multimodal multiview variational autoencoder, M2VAE, to address the cold start problem in item recommendation.
M3ashy: Multi-Modal Material Synthesis via Hyperdiffusion
Chenliang Zhou (University of Cambridge), Cengiz Oztireli (University of Cambridge)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelNeural Radiance FieldImageMultimodality
🎯 What it does: Propose a multimodal material synthesis framework M3ashy based on neural fields and superdiffusion, which can generate high-quality measured BRDF materials under given material types, text descriptions, or reference images.
M3SR: Multi-Scale Multi-Perceptual Mamba for Efficient Spectral Reconstruction
Yuze Zhang, Victor C. M. Leung (Shenzhen University)
RestorationAuto EncoderImageBenchmark
🎯 What it does: Propose a multi-scale, multi-perceptual Mamba architecture (M3SR), which extracts features simultaneously in the spatial, frequency, and spectral domains through the MPF (multi-perceptual fusion) module, and integrates the U-Net structure to achieve global-middle-local multi-scale information fusion for RGB image to hyperspectral image reconstruction.
M3Time: LLM-Enhanced Multi-Modal, Multi-Scale, and Multi-Frequency Multivariate Time Series Forecasting
Shuning Jia (Tsinghua University), Chun Yuan (Tsinghua University)
Convolutional Neural NetworkTransformerLarge Language ModelMultimodalityTime Series
🎯 What it does: Studied how to leverage the semantic prior of large language models and multi-scale, multi-frequency, and multi-modal modeling to enhance multivariate time series forecasting.
M3UCD: A Multi-task Multimodal Metaphor Understanding Challenge Dataset for LLMs
Tianlong Zheng (Xinjiang Technical Institute of Physics and Chemistry, Chinese Academy of Sciences), Turghun Osman (Xinjiang Technical Institute of Physics and Chemistry, Chinese Academy of Sciences)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodalityBenchmark
🎯 What it does: Constructed the largest-scale, fine-grained annotated multimodal metaphor understanding dataset M3UCD, containing 15,345 samples and 12 manually annotated attributes; simultaneously systematically evaluated LLMs' performance on multimodal tasks and proposed a unified multitask collaborative learning framework MCLF to enhance the metaphor understanding capability of multimodal LLMs.