AAAI 2025 Papers — Page 17
AAAI Conference on Artificial Intelligence · 3028 papers
Learnability of Parameter-Bounded Bayes Nets
Arnab Bhattacharyya (University of Warwick), Dimitrios Myrisiotis (CNRS@CREATE Ltd.)
🎯 What it does: It is proven that finding a Bayesian network with a known distribution and finite parameters is still NP-hard, and an upper bound on sample complexity under parameter constraints is provided.
Learned Image Transmission with Hierarchical Variational Autoencoder
Guangyi Zhang (Zhejiang University), Runmin Zhang (Zhejiang University)
GenerationCompressionTransformerAuto EncoderImage
🎯 What it does: A joint source-channel coding (HJSCC) framework based on Hierarchical Variational Autoencoder (VAE) is proposed, supporting adaptive rate wireless image transmission in the presence of feedback links.
Learning 2D Invariant Affordance Knowledge for 3D Affordance Grounding
Xianqiang Gao (University of Science and Technology of China), Bin Zhao (Shanghai AI Laboratory)
Object DetectionRobotic IntelligenceConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes the MIFAG framework, which extracts invariant affordance knowledge from multiple human-machine interaction images and integrates it into 3D point clouds to achieve precise localization of functional areas of 3D objects.
Learning Causal Transition Matrix for Instance-dependent Label Noise
Jiahui Li (Zhejiang University), Jun Zhou (Ant Group)
ClassificationAnomaly DetectionImage
🎯 What it does: This paper proposes the introduction of a causal transfer matrix by constructing a causal graph to address the issue of instance-dependent noisy labels, designing an end-to-end framework to achieve noise separation and label recovery.
Learning Complex Heterogeneous Multimodal Fake News via Social Latent Network Inference
Mingxin Li (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
ClassificationAnomaly DetectionGraph Neural NetworkTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a complex heterogeneous multimodal fake news detection framework HML based on social latent network inference and self-supervised multimodal learning. It can estimate the influence of news through an event-level Hawkes process and construct a potential heterogeneous graph in the absence of real propagation chains, and then classify fake news by combining self-supervised enhanced multimodal features.
Learning Complexity of Gradient Descent and Conjugate Gradient Algorithms
Xianqi Jiao (Xi'an Jiaotong University), Zhiping Chen (Xi'an Jiaotong University)
Optimization
🎯 What it does: This paper studies the learning complexity of gradient descent (GD) and conjugate gradient (CG) algorithms in data-driven algorithm selection and proposes a new cost function based on the cumulative error of each iteration.
Learning Concept Prerequisite Relation via Global Knowledge Relation Optimization
Miao Zhang (Hubei University), Zhifei Li (Hubei University)
Graph Neural NetworkGraphTabular
🎯 What it does: This paper proposes a model called GKROM based on global knowledge relationship optimization for learning prerequisite relationships of concepts.
Learning Cross-Domain Representations for Transferable Drug Perturbations on Single-Cell Transcriptional Responses
Hui Liu (Nanjing Tech University), Shikai Jin (Nanjing Tech University)
Representation LearningDrug DiscoveryAuto EncoderBiomedical Data
🎯 What it does: Proposes the XTransferCDR framework to predict transcriptional responses at the single-cell level caused by drug and genetic perturbations through cross-domain representation learning.
Learning Deep Dissipative Dynamics
Yuji Okamoto (Kyoto University), Ryosuke Kojima (RIKEN BDR)
Time SeriesSequentialPhysics RelatedOrdinary Differential Equation
🎯 What it does: Learn the dynamics of time-series data through neural networks, and use the nonlinear KYP lemma to construct a differentiable projection that maps any dynamics to a dissipative subspace, thereby ensuring the system's dissipation (internal stability, input-output stability, and energy conservation) during the training process.
Learning Disentangled Equivariant Representation for Explicitly Controllable 3D Molecule Generation
Haoran Liu (Texas A&M University), Martin Renqiang Min (NEC Laboratories America)
GenerationDrug DiscoveryGraph Neural NetworkAuto EncoderGraph
🎯 What it does: Proposes an E(3)-equivariant Wasserstein autoencoder that can explicitly control the generation of 3D drug molecules in terms of attributes and structural context;
Learning Dynamic Similarity by Bidirectional Hierarchical Sliding Semantic Probe for Efficient Text Video Retrieval
Yang Liu (Sichuan University), Jiancheng Lv (Stevens Institute of Technology)
RetrievalContrastive LearningVideoText
🎯 What it does: This paper proposes a text-video retrieval framework based on a bidirectional hierarchical sliding semantic probe, which can dynamically compute the similarity between video and text without cross-modal interaction, with a complexity of O(n).
Learning Fine-Grained Alignment for Aerial Vision-Dialog Navigation
Yifei Su (Carnegie Mellon University), Liang Wang (Carnegie Mellon University)
Object DetectionRobotic IntelligenceTransformerContrastive LearningImage
🎯 What it does: This paper proposes a drone visual dialogue navigation method for fine-grained entity-landmark alignment, enhancing navigation effectiveness through semantic grid representation and three auxiliary tasks.
Learning Fine-grained Domain Generalization via Hyperbolic State Space Hallucination
Qi Bi (Wuhan University), Gui-Song Xia (Wuhan University)
Domain AdaptationImage
🎯 What it does: This paper proposes Hyperbolic State Space Hallucination (HSSH), which enhances fine-grained domain generalization by performing style hallucination in the state space and applying consistency constraints in hyperbolic space.
Learning from Mistakes: Self-correct Adversarial Training for Chinese Unnatural Text Correction
Xuan Feng (Jinan University), Liang Chang (Guangxi University of Electronic Technology)
GenerationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningGenerative Adversarial NetworkContrastive LearningText
🎯 What it does: A self-correcting adversarial training framework called LIMIT is proposed, which enhances the robustness of error correction and adversarial resistance for non-natural text in both Chinese and English through a generative correction mechanism, adversarial sample generation based on self-prediction, and decoding intervention strategies.
Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling
Michael Heck (Heinrich Heine University Dusseldorf), Milica Gasic (Heinrich Heine University Dusseldorf)
Meta LearningText
🎯 What it does: A self-learning unsupervised instant meta-loss recalibration method called STORM is proposed, which can learn from noisy labels without a clean validation set;
Learning from Summarized Data: Gaussian Process Regression with Sample Quasi-Likelihood
Yuta Shikuri (Tokio Marine Holdings)
Tabular
🎯 What it does: The study investigates how to use Gaussian process regression for learning and inference when only summary data that captures representative features, summary statistics, and sample sizes is available, and proposes a sample pseudo-likelihood method.
Learning Generalized Residual Exchange-Correlation-Uncertain Functional for Density Functional Theory
Sizhuo Jin (Nanjing University of Science and Technology), Jun Li (Nanjing University of Science and Technology)
Supervised Fine-TuningPhysics Related
🎯 What it does: A Residual Bayesian Network framework is proposed to learn the mean and variance of exchange-correlation (XC) functionals in DFT, thereby improving traditional XC approximations.
Learning Joint Behaviors with Large Variations
Tianxu Li (Nanjing University of Aeronautics and Astronautics), Kun Zhu (Nanjing University of Aeronautics and Astronautics)
Reinforcement LearningBenchmark
🎯 What it does: A new multi-agent exploration method called DJBD is proposed, which enhances the exploration capability of cooperative learning by maximizing the travel distance between joint actions through learning a representation function that satisfies the 1-Lipschitz constraint.
Learning Local Neighborhoods of Non-Gaussian Graphical Models
Sarah Liaw (California Institute of Technology), Ricardo Baptista (California Institute of Technology)
Graph Neural NetworkReinforcement LearningBiomedical Data
🎯 What it does: An expandable algorithm L-SING is proposed to learn the local neighborhood of continuous non-Gaussian distribution graphical models using transport mapping;
Learning More Expressive General Policies for Classical Planning Domains
Simon Ståhlberg (RWTH Aachen University), Hector Geffner (Universitat Pompeu Fabra)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a parameterized relational graph neural network R-GNN[t], which enhances the expressiveness of 2-GNN to the level of 3-GNN through a parameter t-based transformation of state atoms, while maintaining feasible time and space complexity; it is used to learn the value function of general planning domains, thereby obtaining generalizable planning strategies.
Learning Multiple User Distributions for Recommendation via Guided Conditional Diffusion
Cheng Wu (Tsinghua University), Peng Jiang (Kuaishou Inc.)
Recommendation SystemTransformerDiffusion modelTabular
🎯 What it does: A recommendation model based on guided conditional diffusion, GCDR, is proposed, which learns user uncertainty through multi-distribution.
Learning Nash Equilibrium of Markov Potential Games with a Shared Constraint via Primal-Dual Optimization
Songtao Feng (University of Florida), Jie Fu (University of Florida)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: This paper proposes a primal-dual learning algorithm for Markov Potential Games (MPG) with shared constraints, which can converge to an approximate Nash equilibrium under the Slater condition.
Learning Optimal Auctions with Correlated Value Distributions
Da Huo (Shanghai Jiao Tong University), Fan Wu (Shanghai Jiao Tong University)
OptimizationReinforcement Learning
🎯 What it does: A deep learning auction mechanism designed for single-item auctions that can utilize value correlation while maintaining strategic proofness—Conditional Auction Net (CAN).
Learning Personalized Decision Support Policies
Umang Bhatt (New York University), Ameet Talwalkar (Carnegie Mellon University)
Reinforcement LearningImageTabular
🎯 What it does: An interactive tool named Modiste has been developed, utilizing context-free bandit learning for personalized decision support strategies, allowing different decision-makers to automatically receive the most appropriate support (model predictions, expert consensus, or no support) when faced with different inputs;
Learning Physics Informed Neural ODEs with Partial Measurements
Paul Ghanem (Northeastern University), Deniz Erdogmus (Northeastern University)
OptimizationComputational EfficiencyData-Centric LearningTime SeriesSequentialPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper proposes a physical information neural ODE learning framework for partially measured systems, utilizing sequential alternating second-order optimization (Newton) to estimate hidden states and update model parameters at each time step, ultimately obtaining the complete dynamic equations.
Learning Regularization for Graph Inverse Problems
Moshe Eliasof (University of Cambridge), Eldad Haber (University of British Columbia)
Graph Neural NetworkGraph
🎯 What it does: A general framework based on graph neural networks is proposed to solve Graph Inverse Problems (GRIP), and it is validated on various tasks.
Learning Robust and Privacy-Preserving Representations via Information Theory
Binghui Zhang (Illinois Institute of Technology), Binghui Wang (Illinois Institute of Technology)
Safty and PrivacyRepresentation LearningAdversarial AttackTabular
🎯 What it does: This paper proposes an information-theoretic framework called ARPRL, which learns representations that are both adversarially robust and privacy-preserving, and are generalizable to any downstream task.
Learning Set Functions with Implicit Differentiation
Gözde Özcan (Northeastern University), Stratis Ioannidis (Northeastern University)
Recommendation SystemAnomaly DetectionOptimizationTabular
🎯 What it does: This paper studies a framework for learning set functions under the supervision of an optimal subset oracle and improves the gradient computation method based on variational inference.
Learning Strategy Representation for Imitation Learning in Multi-Agent Games
Shiqi Lei (Institute of Automation Chinese Academy of Sciences), Jinkyoo Park (Korea Advanced Institute of Science and Technology)
Representation LearningRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: A STRIL framework is proposed to filter offline data in multi-agent zero-sum games by learning strategy representations and indicators to enhance imitation learning performance.
Learning Structural Causal Models from Ordering: Identifiable Flow Models
Minh Khoa Le (Deakin University), Truyen Tran (Deakin University)
Flow-based ModelBiomedical DataMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: Using observational data and known causal ordering, we design and train identifiable flow models (CFM, S-CFM, P-CFM) to learn structural causal models (SCM), supporting observational, intervention, and counterfactual reasoning.
Learning Theorem Rationale for Improving the Mathematical Reasoning Capability of Large Language Models
Yu Sheng (Chinese Academy of Sciences), Daniel Dajun Zeng (Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: By constructing a dataset of question-theorem-answer triples and performing instruction fine-tuning, we propose Explicit Learning Theorem Reasoning (Theorem Rationale, TR) to enhance the mathematical reasoning capabilities of large language models.
Learning to Collaborate with Unknown Agents in the Absence of Reward
Zuyuan Zhang (George Washington University), Tian Lan (George Washington University)
Reinforcement Learning
🎯 What it does: This study proposes a framework (STUN) for collaborating with unknown agents in the absence of a reward function, by actively inferring the potential rewards of unknown agents and adapting strategies based on the inferred results without online retraining.
Learning to Generate Gradients for Test-Time Adaptation via Test-Time Training Layers
Qi Deng (South China University of Technology), Xiping Hu (Artificial Intelligence Research Institute Shenzhen MSU-BIT University)
Domain AdaptationOptimizationImage
🎯 What it does: A Meta Gradient Generator (MGG) based on learning optimization is proposed, which generates reliable gradients during the adaptive process at testing time, thereby replacing traditional manual optimizers;
Learning to Manipulate Under Limited Information
Wesley H. Holliday (University of California), Eric Pacuit (University of Maryland)
Supervised Fine-TuningTabular
🎯 What it does: Train a large-scale multilayer perceptron to learn strategic voting on eight voting rules with only limited voting information.
Learning to Prompt with Text Only Supervision for Vision-Language Models
Muhammad Uzair Khattak (MBZ University of AI), Federico Tombari (Khalifa University)
ClassificationObject DetectionSegmentationDomain AdaptationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: ProText is proposed, which utilizes text descriptions generated by large language models for text-supervised prompt learning to enhance the zero-shot generalization ability of CLIP in visual tasks.
Learning to Rewind via Iterative Prediction of Past Weights for Practical Unlearning
Jinhyeok Jang (Electronics and Telecommunications Research Institute), Chan-Hyun Youn (Korea Advanced Institute of Science and Technology)
Convolutional Neural NetworkDiffusion modelImage
🎯 What it does: A machine forgetting method based on inverse weight prediction, InvWNN, is proposed, which can eliminate model memory with only the data to be forgotten.
Learning Together Securely: Prototype-Based Federated Multi-Modal Hashing for Safe and Efficient Multi-Modal Retrieval
Ruifan Zuo (Qilu University of Technology), Xiaofeng Qu (University of Jinan)
RetrievalFederated LearningSafty and PrivacyMultimodality
🎯 What it does: Proposes Prototype-based Federated Multi-modal Hashing (PFMH), which combines federated learning with multi-modal hashing to achieve secure and efficient multi-modal retrieval.
Learning Valid Dual Bounds in Constraint Programming: Boosted Lagrangian Decomposition with Self-Supervised Learning
Swann Bessa (Polytechnique Montreal), Quentin Cappart (Polytechnique Montreal)
OptimizationGraph Neural NetworkTabular
🎯 What it does: Utilizing self-supervised learning, the graph neural network directly predicts Lagrange multipliers, thereby generating effective and compact dual lower bounds in constraint programming, replacing or accelerating traditional subgradient iterations;
Learning Verified Safe Neural Network Controllers for Multi-Agent Path Finding
Mingyue Zhang (Southwest University), Wu Chen (Southwest University)
Knowledge DistillationRobotic IntelligenceReinforcement Learning from Human FeedbackConvolutional Neural NetworkReinforcement LearningTabular
🎯 What it does: This paper proposes a verification-safe multi-agent neural control method (VSMANC), which addresses the collision avoidance problem in multi-agent path planning by jointly training decentralized control barrier functions (DCBF) and controllers, and achieving safety guarantees through formal verification and counterexample retraining.
Learning Visually Grounded Domain Ontologies via Embodied Conversation and Explanation
Jonghyuk Park (University of Edinburgh), Subramanian Ramamoorthy (University of Edinburgh)
Object DetectionSegmentationExplainability and InterpretabilityRobotic IntelligenceAgentic AIImage
🎯 What it does: Proposes an interpretable interactive learning framework that allows agents to gradually build a visual domain ontology and improve visual recognition through corrections from teachers based on their explanations.
Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space
Linchao Pan (Shenzhen University), Jinbao Wang (Shenzhen University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: To address the problem of noisy labels in an open world, a dual representation space joint learning framework is proposed, which learns features of the prototype space and class-independent space through a projection network and a one-to-many network.
Learnware Specification via Label-Aware Neural Embedding
Wei Chen (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationOptimizationConvolutional Neural NetworkImageText
🎯 What it does: A learnware normalization method named LANE is proposed, which utilizes label information and random neural networks to map training data into a neural embedding space, thereby generating more accurate model specifications.
Less Is More: Adaptive Program Repair with Bug Localization and Preference Learning
Zhenlong Dai (Zhejiang University), Jingyuan Chen (Zhejiang University)
AI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the AdaPR task and designs the AdaPatcher two-stage framework (Bug Locator + Program Modifier) to achieve adaptive program repair, generating patches that meet specifications and have minimal modifications.
Less Is More: Token Context-Aware Learning for Object Tracking
Chenlong Xu (Guangxi Normal University), Shuxiang Song (University of Chinese Academy of Sciences)
Object TrackingTransformerVideo
🎯 What it does: We propose LMTrack—a tracking framework based on Token Context that automatically collects and updates important reference tokens from videos, achieving efficient cross-frame association and localization using a unidirectional attention mechanism.
Leveraging Anatomical Consistency for Multi-Object Detection in Ultrasound Images via Source-free Unsupervised Domain Adaptation
Bin Pu (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
Object DetectionDomain AdaptationGraph Neural NetworkImageBiomedical DataUltrasound
🎯 What it does: To address the problem of multi-object (organ) detection under source-free unsupervised domain adaptation (SF-UDA), an AATS framework utilizing anatomical consistency (topological and morphological) is proposed, which can achieve ultrasound image detection transfer across hospitals and devices using only source domain pre-trained weights.
Leveraging Asynchronous Spiking Neural Networks for Ultra Efficient Event-Based Visual Processing
DingYi Zeng (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
ClassificationComputational EfficiencyGraph Neural NetworkSpiking Neural NetworkImageGraph
🎯 What it does: This paper proposes an Asynchronous Spiking Graph Convolutional Network (ASGCN), which achieves event-driven low-latency and low-energy visual processing by directly mapping event streams to dynamic graphs and asynchronously executing spiking convolutions on K-hop subgraphs.
Leveraging Attention to Effectively Compress Prompts for Long-Context LLMs
Yunlong Zhao (Chinese Academy of Sciences), Bo Xu (Chinese Academy of Sciences)
RetrievalCompressionTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposes the AttnComp method, which utilizes cross-attention and self-attention of LLMs to achieve prompt compression, generating compact contexts that align with queries.
Leveraging Consistent Spatio-Temporal Correspondence for Robust Visual Odometry
Zhaoxing Zhang (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
Pose EstimationAutonomous DrivingRecurrent Neural NetworkSimultaneous Localization and MappingOptical FlowImageVideo
🎯 What it does: Proposes the STVO framework, which utilizes spatiotemporal consistent optical flow matching to achieve multi-frame visual odometry.
Leveraging Constraint Violation Signals for Action Constrained Reinforcement Learning
Janaka Chathuranga Brahmanage (Singapore Management University), Akshat Kumar (Singapore Management University)
Reinforcement LearningFlow-based ModelSequential
🎯 What it does: A flow model based on constraint violation signals, CV-Flow, has been developed to directly map sampled potential actions to the actionable action space in reinforcement learning with action constraints, and it is combined with Soft Actor-Critic (SAC) for safe control.
Leveraging First and Zeroth-Order Gradient to Address Imbalanced Black-Box Prompt Tuning via Minimax Optimization
Haozhen Zhang (Jilin University), Yi Chang (Jilin University)
OptimizationTransformerPrompt EngineeringText
🎯 What it does: A black-box prompt tuning framework (BPT-FZG) is designed to handle imbalanced data in downstream tasks by maximizing AUC and transforming it into a non-convex-concave extremum problem.
Leveraging Group Classification with Descending Soft Labeling for Deep Imbalanced Regression
Ruizhi Pu (Western University), Boyu Wang (Western University)
ClassificationOptimizationMixture of ExpertsContrastive LearningText
🎯 What it does: This paper proposes a group classification and multi-expert regression framework based on symmetric decreasing soft labels, utilizing group contrastive learning and soft labels to enhance deep imbalance regression performance.
Leveraging Large Language Models for Node Generation in Few-Shot Learning on Text-Attributed Graphs
Jianxiang Yu (East China Normal University), Xuecang Zhang (Huawei Technologies)
ClassificationGenerationData SynthesisMeta LearningGraph Neural NetworkLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: A lightweight framework LLM4NG is proposed, which generates node text samples using large language models in few-shot scenarios, and integrates these generated nodes into the original graph through an edge predictor, achieving node classification without modifying the original graph data.
Leveraging Large Vision-Language Model as User Intent-Aware Encoder for Composed Image Retrieval
Zelong Sun (Renmin University of China), Zhiwu Lu (Renmin University of China)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes the CIR-LVLM framework based on a large vision-language model, using LVLM as a user intent-aware encoder to perform combined image retrieval.
Leveraging RGB-D Data with Cross-Modal Context Mining for Glass Surface Detection
Jiaying Lin (City University of Hong Kong), Rynson W. H. Lau
Object DetectionSegmentationConvolutional Neural NetworkImageMultimodality
🎯 What it does: A new framework for glass surface detection using RGB-D data is proposed.
Leveraging the Dual Capabilities of LLM: LLM-Enhanced Text Mapping Model for Personality Detection
Weihong Bi (Beijing University of Posts and Telecommunications), Mingying Xu (North China University of Technology)
ClassificationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper proposes an LLM-enhanced Text Mapping Model (ETM) that encodes posts using a small model, extracts long text embeddings with a lightweight LLM, fuses user vectors through cross-attention, generates multi-dimensional MBTI label explanations with a powerful LLM, and optimizes the mapping of user vectors to labels using contrastive learning, thereby achieving personality detection.
LIBA: Language Instructed Multi-granularity Bridge Assistant for 3D Visual Grounding
Yuan Wang (Tsinghua University), Shengjin Wang (Tsinghua University)
Object DetectionSegmentationTransformerLarge Language ModelPoint Cloud
🎯 What it does: The LIBA framework is proposed, utilizing multi-granularity bridging adapters, cross-scale object modulation, and LLM-guided hierarchical query selection to achieve cross-scale alignment and localization of 3D vision and language.
LiD-FL: Towards List-Decodable Federated Learning
Hong Liu (Sichuan University), Jiancheng Lv (Sichuan University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: A list-decodable federated learning framework LiD-FL is proposed, where the server maintains a set of models, randomly selects client updates, and ensures that at least one model is valid through a voting mechanism.
LiDAR-LLM: Exploring the Potential of Large Language Models for 3D LiDAR Understanding
Senqiao Yang (Peking University), Shanghang Zhang
Object DetectionAutonomous DrivingTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalityPoint Cloud
🎯 What it does: This paper proposes a framework called LiDAR-LLM that integrates 3D LiDAR data with large language models (LLM) to achieve linguistic understanding and reasoning of outdoor 3D scenes.
Lifelong Scalable Generative System via Online Maximum Mean Discrepancy
Fei Ye (University of Electronic Science and Technology of China), Adrian G. Bors (University of York)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: In an unsupervised, task-agnostic continual learning scenario, a Dynamic Expansion Memory Unit (DEMU) is proposed, which determines whether to add new memory buffers to preserve key data through Maximum Mean Discrepancy (MMD);
Lifting Scheme-Based Implicit Disentanglement of Emotion-Related Facial Dynamics in the Wild
Xingjian Wang (Zhejiang University), Li Chai (Zhejiang University)
ClassificationRecognitionConvolutional Neural NetworkTransformerVideoMultimodality
🎯 What it does: The IFDD framework is proposed, utilizing a learnable wavelet lifting scheme for two-stage implicit decomposition of video expression dynamics, separating emotion-related dynamic features from global context through the Inter-frame Static-dynamic Splitting Module (ISSM) and the Lifting-based Aggregation-disentanglement Module (LADM);
Light-T2M: A Lightweight and Fast Model for Text-to-motion Generation
Ling-An Zeng (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationCompressionComputational EfficiencyConvolutional Neural NetworkDiffusion modelTextMultimodality
🎯 What it does: A lightweight text-to-motion generation model, Light-T2M, is proposed, significantly reducing the number of parameters and inference latency while maintaining or improving generation quality.
LightPROF: A Lightweight Reasoning Framework for Large Language Model on Knowledge Graph
Tu Ao (Beijing University of Posts and Telecommunications), Zhen Cai (National University of Singapore)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: The LightPROF framework is proposed, which enhances KGQA reasoning capabilities by generating soft prompts for small-scale LLMs through a three-step process of retrieval, embedding, and inference using the structural information of knowledge graphs.
Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval
Jiaxing Li (Guangzhou University), Jie Wen (Harbin Institute of Technology)
RetrievalKnowledge DistillationContrastive LearningMultimodality
🎯 What it does: A lightweight Contrastive Distillation Hashing (LCDH) model is proposed for online cross-modal retrieval, which can achieve the transfer of semantic relevance extracted by the CLIP + attention module from an offline teacher network to a lightweight student network through knowledge distillation and similarity matrix approximation.
Lightweight Yet Fine-Grained: A Graph Capsule Convolutional Network with Subspace Alignment for Shared-Account Sequential Recommendation
Jinyu Zhang (Shandong University of Science and Technology), Yanwei Yu (Ocean University of China)
Recommendation SystemGraph Neural NetworkContrastive LearningSequential
🎯 What it does: Proposes LightGCN2N, a lightweight graph capsule convolutional network with subspace alignment, to address the shared account sequential recommendation problem.
Like an Ophthalmologist: Dynamic Selection Driven Multi-View Learning for Diabetic Retinopathy Grading
Xiaoling Luo (Shenzhen University), Linlin Shen (Shenzhen University)
ClassificationExplainability and InterpretabilityTransformerMixture of ExpertsImage
🎯 What it does: A multi-view grading method for diabetic retinopathy that simulates the diagnostic process of ophthalmologists has been designed.
Limitations in Employing Natural Language Supervision for Sensor-Based Human Activity Recognition - And Ways to Overcome Them
Harish Haresamudram (Georgia Institute of Technology), Thomas Ploetz (Optum AI)
RecognitionRetrievalLarge Language ModelSupervised Fine-TuningContrastive LearningMultimodalityTime Series
🎯 What it does: This study explores the use of Natural Language Supervision (NLS) for human activity recognition with wearable sensors, and proposes solutions to the challenges of sensor heterogeneity and the scarcity of text descriptions.
Linear Equations with Min and Max Operators: Computational Complexity
Krishnendu Chatterjee (Institute of Science and Technology Austria), Jakub Svoboda (Institute of Science and Technology Austria)
Optimization
🎯 What it does: This paper studies and systematically analyzes the computational complexity of linear equation systems with minimum/maximum operations (LEMM) under different constraint conditions.
Linear Streaming Bandit: Regret Minimization and Fixed-Budget Epsilon-Best Arm Identification
Yuming Shao (Tsinghua University), Zhixuan Fang (Shanghai Qi Zhi Institute)
Reinforcement LearningTabular
🎯 What it does: A linear streaming bandit model is proposed, along with a multi-channel cumulative reward minimization and fixed budget ε-optimal arm identification algorithm designed for this model.
Linking Industry Sectors and Financial Statements: A Hybrid Approach for Company Classification
Guy Stephane Waffo Dzuyo (Forvis Mazars), Luis Belmar-Letelier (Forvis Mazars)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextTabularFinance Related
🎯 What it does: This paper develops various financial statement representation methods and constructs a company industry classification model based on machine learning and language models, exploring the enhancement of classification performance through textual information.
LiON: Learning Point-Wise Abstaining Penalty for LiDAR Outlier DetectioN Using Diverse Synthetic Data
Shaocong Xu (Tsinghua University), Yilun Chen (Tsinghua University)
Anomaly DetectionAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes the LiON method, which combines selective classification and point-level abandonment penalty to address the issue of LiDAR point cloud anomaly detection, and generates diverse realistic anomalies through ShapeNet.
List Update with Prediction
Yossi Azar (Tel Aviv University), Varun Suriyanarayana (Cornell University)
TabularSequential
🎯 What it does: A randomized list update algorithm based on machine learning prediction has been designed and implemented, which guarantees a 1+ε smoothness on any request sequence and maintains good robustness in the presence of large errors.
LiteSearch: Efficient Tree Search with Dynamic Exploration Budget for Math Reasoning
Ante Wang (Xiamen University), Dong Yu (Tencent AI Lab)
Large Language ModelTextTabularChain-of-Thought
🎯 What it does: LiteSearch is proposed, an efficient tree search algorithm that utilizes dynamic node-level exploration budgets to solve mathematical reasoning tasks on large language models.
Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning
Amr Abourayya (University Hospital Essen), Michael Kamp (University Hospital Essen)
Federated LearningSafty and PrivacyTabularBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A Federated Collaborative Training (FEDCT) framework is proposed, which only shares hard labels from publicly available unlabeled datasets as pseudo-labels for local training, thereby enhancing privacy protection while maintaining model quality.
LLaVA Needs More Knowledge: Retrieval Augmented Natural Language Generation with Knowledge Graph for Explaining Thoracic Pathologies
Ameer Hamza (Kyung Hee University), Seong Tae Kim (Kyung Hee University)
GenerationRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodalityComputed TomographyElectronic Health RecordsRetrieval-Augmented Generation
🎯 What it does: This paper proposes a knowledge graph retrieval-augmented visual-language framework (KG-RAG) for generating natural language explanations (NLE) of chest X-ray images.
LLM Agents Can Be Choice-Supportive Biased Evaluators: An Empirical Study
Nan Zhuang (Zhejiang University), Qi Liu (South China University of Technology)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper systematically studies the choice support bias of large language model (LLM) agents in decision evaluation through two types of experiments: memory-oriented and evaluation-oriented, and compares it with highly educated human subjects.
LLM-DR: A Novel LLM-Aided Diffusion Model for Rule Generation on Temporal Knowledge Graphs
Kai Chen (National University of Defense Technology), Yalong Xie (Hunan University of Humanities, Science and Technology)
GenerationData SynthesisRecommendation SystemRecurrent Neural NetworkTransformerLarge Language ModelDiffusion modelGraphTime SeriesSequential
🎯 What it does: A rule-based temporal knowledge graph extrapolation method called LLM-DR is proposed, which utilizes a diffusion model to generate high-quality logical rules consistent with the original rule distribution under class-conditional guidance, and combines semantic constraints from large language models for rule generation and evaluation, further enhancing the predictive performance of temporal knowledge graphs.
LLM-Powered User Simulator for Recommender System
Zijian Zhang (Jilin University), Peng Jiang (Kuaishou Technology)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningTextSequential
🎯 What it does: A user simulator based on large language models has been constructed, combining logical reasoning and statistical learning to explicitly simulate the 'like/dislike' interactions in recommendation systems.
LLM-RG4: Flexible and Factual Radiology Report Generation Across Diverse Input Contexts
Zhuhao Wang (Tsinghua University), Hongen Liao (Tsinghua University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataElectronic Health Records
🎯 What it does: A radiology report generation framework LLM-RG4 is proposed, which can flexibly adapt to various clinical input scenarios, and a corresponding dataset MIMIC-RG4 is constructed.
LLM+AL: Bridging Large Language Models and Action Languages for Complex Reasoning About Actions
Adam Ishay (Arizona State University), Joohyung Lee (Samsung Research)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: A framework that combines large language models (LLM) with action language (AL) is proposed (LLM+AL) for complex action reasoning tasks.
LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation
Mushui Liu (Zhejiang University), Changjie Fan (Netease Inc.)
GenerationTransformerLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: This study proposes the LLM4GEN framework, which integrates the semantic representations of large language models (LLMs) into text-to-image diffusion models through cross-adapter modules, enhancing the generation quality and semantic alignment of complex prompts.
LLM4RSR: Large Language Models as Data Correctors for Robust Sequential Recommendation
Yatong Sun (Northeastern University), Xinghua Qu (Bytedance)
Recommendation SystemKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringSequential
🎯 What it does: By using large language models (LLM) to semantically correct unreliable input-target pairs in sequence recommendation training samples, the accuracy of robust sequence recommendations is improved.
LLMEmb: Large Language Model Can Be a Good Embedding Generator for Sequential Recommendation
Qidong Liu (Xi'an Jiaotong University), Yefeng Zheng (Tencent)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningSequential
🎯 What it does: This paper proposes LLMEmb, a framework that utilizes large language models to generate project embeddings for sequential recommendation systems.
LNS2+RL: Combining Multi-agent Reinforcement Learning with Large Neighborhood Search in Multi-agent Path Finding
Yutong Wang (National University of Singapore), Guillaume Sartoretti (National University of Singapore)
OptimizationRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningAgentic AISequential
🎯 What it does: In high-density multi-agent path planning tasks, the authors propose the LNS2+RL algorithm, which adaptively switches between low-speed high-quality planning based on MARL and high-speed low-quality planning based on PP+SIPPS during the iterative process, significantly reducing the number of conflicts while maintaining scalability.
Local Causal Discovery for Structural Evidence of Direct Discrimination
Jacqueline Maasch (Stevens Institute of Technology), Fei Wang (Cornell Tech)
Tabular
🎯 What it does: The LD3 local causal discovery method is proposed to identify structural evidence of direct discrimination.
Local Causal Discovery Without Causal Sufficiency
Zhaolong Ling (Anhui University), Kui Yu (Hefei University of Technology)
GraphTabularAgriculture Related
🎯 What it does: A local causal discovery algorithm named LatentLCD has been developed to learn the local MAG of the target variable under non-causal sufficiency conditions.
Local Conditional Controlling for Text-to-Image Diffusion Models
Yibo Zhao (Zhejiang University), Boxi Wu (Tencent)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A local control mechanism is proposed, allowing users to apply image conditions only to specified areas in the text-to-image diffusion model, while the remaining areas rely solely on text prompts.
Locally Convex Global Loss Network for Decision-Focused Learning
Haeun Jeon (Korea Advanced Institute of Science and Technology), Woo Chang Kim (Korea Advanced Institute of Science and Technology)
Recommendation SystemOptimizationData-Centric LearningReinforcement Learning from Human FeedbackTabularTime SeriesFinance Related
🎯 What it does: This paper proposes the Locally Convex Global Loss Network (LCGLN), a surrogate model that can learn the global task loss in a single pass across various decision-focused learning (DFL) tasks, and validates its effectiveness on three types of stochastic decision problems: inventory, budget allocation, and asset portfolio.
Locate Anything on Earth: Advancing Open-Vocabulary Object Detection for Remote Sensing Community
Jiancheng Pan (Tsinghua University), Xiaomeng Huang (Tsinghua University)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: This paper proposes the open vocabulary object detection task 'Locate Anything on Earth (LAE)' in the field of remote sensing, and constructs the LAE-1M dataset containing approximately one million instances through the LAE-Label Engine. It also designs and trains the LAE-DINO model, enhancing the open object detection capability of remote sensing images.
Logarithmic Regret for Linear Markov Decision Processes with Adversarial Corruptions
Canzhe Zhao (Shanghai Jiao Tong University), Shuai Li (Shanghai Jiao Tong University)
Reinforcement Learning
🎯 What it does: This study explores the logarithmic regret problem of reinforcement learning (RL) in linear Markov decision processes (MDPs) with adversarial disturbances. An algorithm called Double Weighted Least Squares Value Iteration with Upper Confidence Bound (DW-LSVI-UCB) is proposed to simultaneously learn the unknown reward parameters and transition parameters.
Logic Induced High-Order Reasoning Network for Event-Event Relation Extraction
Peixin Huang (National University of Defense Technology), Weidong Xiao (National University of Defense Technology)
Graph Neural NetworkTransformerText
🎯 What it does: This paper proposes a high-order reasoning network driven by logical constraints (LogicERE) to extract temporal and sub-event relationships between document-level events, generating a unified event evolution graph.
Logic-Q: Improving Deep Reinforcement Learning-based Quantitative Trading via Program Sketch-based Tuning
Zhiming Li (Nanyang Technological University), Danny Dongning Sun (Peng Cheng Lab)
OptimizationReinforcement LearningTabularTime SeriesFinance Related
🎯 What it does: A logic-guided deep reinforcement learning framework called Logic-Q is proposed, which embeds abstract market trend knowledge into existing DRL trading strategies through program sketches and performs post-tuning of the strategies during trading.
LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction
Er Jin (RWTH Aachen University), Johannes Stegmaier (RWTH Aachen University)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextChain-of-Thought
🎯 What it does: Proposes LogicAD, which utilizes AVLM to extract text features for one-shot logical anomaly detection and provides interpretable anomaly explanations.
LoGoFair: Post-Processing for Local and Global Fairness in Federated Learning
Li Zhang (Zhejiang University), Xiaolin Zheng (Zhejiang University)
OptimizationFederated LearningTabular
🎯 What it does: The LoGoFair framework is proposed, achieving local and global fairness in federated learning through post-processing.
LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views
Ziyun Zou (Xiamen University), Xiangrong Liu (Xiamen University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Designed and implemented the LOHA framework, utilizing self-supervised graph contrastive learning with direct comparison of low-pass and high-pass spectral filters, and introduced spectral signal trends as composite features to avoid feature separation of the same node between two views.
LOMA: Language-assisted Semantic Occupancy Network via Triplane Mamba
Yubo Cui (Northeastern University), Zheng Fang (Northeastern University)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkVision Language ModelMultimodalityPoint Cloud
🎯 What it does: This paper proposes a 3D semantic occupancy prediction network LOMA based on a visual-language framework, which utilizes VLM to generate 3D language features and fuse them with visual features, achieving finer semantic and geometric completion.
Long-Tailed Out-of-Distribution Detection: Prioritizing Attention to Tail
Yina He (Xiamen University), Zhiming Luo (Xiamen University)
ClassificationAnomaly DetectionContrastive LearningImage
🎯 What it does: This paper proposes an OOD detection framework called PATT under long-tail distribution, focusing on balancing features and classifiers through attention enhancement for tail classes and implicit semantic enhancement, as well as feature calibration during inference.
Long-Term EEG Partitioning for Seizure Onset Detection
Zheng Chen (Osaka University), Jimeng Sun (University of Illinois Urbana-Champaign)
RecognitionAnomaly DetectionRecurrent Neural NetworkGraph Neural NetworkTime SeriesBiomedical Data
🎯 What it does: A two-stage SODor framework is proposed to explicitly identify the onset points of seizures through subsequence clustering.
Look Around Before Locating: Considering Content and Structure Information for Visual Grounding
Shiyi Zheng (Guangxi University), Qingbao Huang (Guangxi University)
RecognitionObject DetectionTransformerVision Language ModelImageText
🎯 What it does: A semi-structured reasoning framework based on Transformer, called SSRVG, is proposed for visual localization tasks.
Look Back for More: Harnessing Historical Sequential Updates for Personalized Federated Adapter Tuning
Danni Peng (Institute of High Performance Computing), Qingsong Wei (Institute of High Performance Computing)
Federated LearningTransformerSupervised Fine-TuningImage
🎯 What it does: The pFedSeq framework is proposed, which utilizes a server-side sequential learner to update the sequence of client historical adapters, achieving personalized federated adapter fine-tuning.
Look Before You Leap: Enhance Attention and Vigilance Regarding Harmful Content with GuidelineLLM
Shaoqing Zhang (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A defense framework named GuidelineLLM is proposed, which utilizes an auxiliary LLM to identify risks in queries before answering and generate safety guidelines. These guidelines are then provided along with the original query to the target LLM, thereby reducing the probability of generating harmful content.