AAAI 2025 Papers — Page 21
AAAI Conference on Artificial Intelligence · 3028 papers
Offline Safe Reinforcement Learning Using Trajectory Classification
Ze Gong (Singapore Management University), Pradeep Varakantham (Singapore Management University)
Safty and PrivacyReinforcement LearningContrastive LearningTabular
🎯 What it does: This paper proposes transforming the offline safe reinforcement learning problem into a trajectory classification problem. It first divides the pre-collected dataset into two categories of trajectories: 'acceptable' and 'unacceptable', and then directly optimizes the policy by training a trajectory classifier to generate acceptable trajectories and avoid unacceptable ones.
Offline-to-Online Hyperparameter Transfer for Stochastic Bandits
Dravyansh Sharma (Toyota Technological Institute at Chicago), Arun Suggala (Google DeepMind)
Hyperparameter SearchReinforcement LearningImage
🎯 What it does: A hyperparameter transfer framework is proposed that learns from offline task data and transfers to online stochastic bandit tasks.
OGP-Net: Optical Guidance Meets Pixel-Level Contrastive Distillation for Robust Multi-Modal and Missing Modality Segmentation
Aniruddh Sikdar (Indian Institute of Science), Suresh Sundaram (Indian Institute of Science)
SegmentationKnowledge DistillationConvolutional Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: This study investigates the use of RGB and IR images for semantic segmentation in multimodal and missing modality scenarios, and proposes a new fusion network.
OLiDM: Object-aware LiDAR Diffusion Models for Autonomous Driving
Tianyi Yan (University of Macau), Jianbing Shen (Li Auto Inc)
GenerationData SynthesisAutonomous DrivingDiffusion modelPoint Cloud
🎯 What it does: This paper proposes an object-aware LiDAR data generation framework based on diffusion models, called OLiDM, which can generate high-quality and controllable point clouds at both object and scene levels.
OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition
Yiheng Yu (Zhejiang University of Technology), Xuhua Yang (Zhejiang University of Technology)
RecognitionKnowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: A framework for continuous sign language recognition called OLMD is proposed, focusing on long-term motion aggregation and multi-directional motion decoupling to enhance the model's understanding and recognition capabilities of complex sign language actions.
Omni-Query Active Learning for Source-Free Domain Adaptive Cross-Modality 3D Semantic Segmentation
Jianxiang Xie (Xiamen University), Yanyun Qu (Xiamen University)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkMultimodalityPoint Cloud
🎯 What it does: Proposes the OmniQuery framework, which combines active learning with source-free domain adaptation to achieve cross-modal 3D semantic segmentation.
OmniCount: Multi-label Object Counting with Semantic-Geometric Priors
Anindya Mondal (University of Surrey), Anjan Dutta (University of Surrey)
Object DetectionSegmentationImage
🎯 What it does: This paper proposes OmniCount, a training-independent, open-vocabulary, multi-label object counting method that can count multiple categories of objects at once.
OmniMark: Efficient and Scalable Latent Diffusion Model Fingerprinting
Jianwei Fei (University of Macau), Jiantao Zhou (University of Macau)
GenerationData SynthesisComputational EfficiencyAdversarial AttackConvolutional Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: A lightweight method called OmniMark is proposed for the Latent Diffusion Model (LDM), which allows each generated image to carry an invisible fingerprint without affecting the generation quality, enabling model attribution and accountability tracking.
OmniSR: Shadow Removal Under Direct and Indirect Lighting
Jiamin Xu (Hangzhou Dianzi University), Gang Xu (Zhejiang University)
RestorationData SynthesisTransformerImage
🎯 What it does: A method for removing both direct and indirect light shadows has been developed, and a synthetic dataset containing over 30,000 pairs of shadowed and non-shadowed images has been constructed.
On Action Theories with Iterable First-Order Progression
Daxin Liu (Nanjing University), Jens Claßen (Roskilde University)
🎯 What it does: This paper studies the progression problem in action theory within scenario calculus, proposing a progression method that can be iteratively executed in first-order logic, and defines a new class of action theories;
On Corruption-Robustness in Performative Reinforcement Learning
Vasilis Pollatos (Archimedes Athena Research Center), Goran Radanovic (Max Planck Institute for Software Systems)
OptimizationReinforcement LearningSequential
🎯 What it does: A robust repeated training framework is proposed for performative reinforcement learning environments with Huber ε-contamination, along with a convergence theory proof.
On Designing the Optimal Integrated Ad Auction in E-commerce Platforms
Yuchao Ma (Renmin University of China), Changyuan Yu (Baidu Inc)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: This paper studies the optimal auction mechanism for integrating advertisements and organic products on the same search results page in e-commerce platforms, aiming to balance platform revenue and user experience.
On Effects of Steering Latent Representation for Large Language Model Unlearning
Huu-Tien Dang (Japan Advanced Institute of Science and Technology), Naoya Inoue (Japan Advanced Institute of Science and Technology)
Adversarial AttackData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper achieves effective 'unlearning' of undesirable knowledge in models by shifting the representations of forgotten samples in the intermediate layers of large language models (LLMs) in a random direction (RMU), combined with an improved method of adaptive coefficients (Adaptive RMU), while enhancing robustness against adversarial jailbreak attacks.
On Finding Hubs in High Dimensions with Sampling
Huiwen Dong (Beijing Normal University), Ninh Pham (University of Auckland)
OptimizationComputational EfficiencyTime Series
🎯 What it does: By employing random sampling kNN computation on high-dimensional datasets, SamHub efficiently identifies hub points with a time complexity of O(sn).
On Local Overfitting and Forgetting in Deep Neural Networks
Uri Stern (Hebrew University of Jerusalem), Daphna Weinshall (Hebrew University of Jerusalem)
OptimizationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a new local overfitting metric - the validation set forgetting rate, and utilizes this metric to construct a lightweight implicit ensemble method to recover knowledge that the network has forgotten during later training phases.
On Oversquashing in Graph Neural Networks Through the Lens of Dynamical Systems
Alessio Gravina (University of Pisa), Carola-Bibiane Schönlieb (University of Cambridge)
Graph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: This paper studies and proposes a differential equation-based graph neural network model called SWAN, which introduces antisymmetric parameterization in both spatial and weight domains to achieve global and local non-dissipative characteristics, thereby addressing the oversquashing problem in traditional message-passing networks.
On Probabilistic Truncation in Privacy-preserving Machine Learning
Lijing Zhou (Huawei Technology), Yu Yu (Shanghai Jiao Tong University)
Safty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper systematically analyzes the errors of probabilistic truncation in Privacy-Preserving Machine Learning (PPML) and proposes methods such as deterministic truncation protocols, ReLU channel scaling techniques, and zero-keeping random mappings, significantly reducing communication overhead and improving inference speed.
On Shallow Planning Under Partial Observability
Randy Lefebvre (Laval University), Audrey Durand (Laval University)
Reinforcement Learning
🎯 What it does: This paper discusses the impact of shallow planning (i.e., shorter discount factors) on the bias-variance trade-off in reinforcement learning within partially observable environments, and provides a new planning loss bound based on MDP structural parameters.
On the Asymptotic Optimality of Confidence Interval Based Algorithms for Fixed Confidence MABs
Kushal Kejriwal (Indian Institute of Technology Bombay), Jayakrishnan Nair (Indian Institute of Technology Bombay)
Optimization
🎯 What it does: This paper addresses the problem of best arm identification under fixed confidence, conducting a quantitative analysis of the asymptotic sample complexity and sampling ratio of the classic LUCB algorithm and its greedy variant LUCB Greedy, and proving that its upper bound differs from the information theoretical lower bound by a constant factor as δ approaches 0.
On the Convergence of Tâtonnement for Linear Fisher Markets
Tianlong Nan (Columbia University), Christian Kroer (Columbia University)
OptimizationTabular
🎯 What it does: This study investigates the convergence of the t‑tatonnement process in linear Fisher markets and quasi-linear Fisher markets, proving that with sufficiently small step sizes, the process converges linearly to an approximate equilibrium.
On the Distortion of Committee Election with 1-Euclidean Preferences and Few Distance Queries
Dimitris Fotakis (National Technical University of Athens), Panagiotis Patsilinakos (Paris-Dauphine University)
🎯 What it does: The study examines the metric distortion of the k-committee election problem with limited distance queries in one-dimensional Euclidean space.
On the Expressiveness and Length Generalization of Selective State Space Models on Regular Languages
Aleksandar Terzic (IBM Research), Abbas Rahimi (IBM Research)
Recurrent Neural NetworkTransformerSequential
🎯 What it does: This paper proposes a new Selective Dense State Space Model (SD-SSM) for perfectly learning and generalizing regular language (finite state machine) tasks within a single-layer model.
On the Hardness of Training Deep Neural Networks Discretely
Ilan Doron-Arad (Technion - Israel Institute of Technology)
🎯 What it does: This paper conducts an in-depth theoretical study on the complexity of training discrete parameters (D-NNT) and continuous parameters (C-NNT) in deep neural networks, providing hardness lower bounds concerning dimensions such as network depth, width, and dataset size, and offers a pseudo-polynomial solving algorithm for two-layer networks.
On the Logic of Theory Base Change: Reformulation of Belief Bases
Eduardo L. Fermé (Universidade da Madeira), Maria Vanina Martinez (Artificial Intelligence Research Institute)
🎯 What it does: A new belief base reformulation operation is proposed, its construction method is defined, and its axiomatic characteristics are provided. It is implemented based on two fundamental contractions: partial meet contraction and kernel contraction, and several non-redundant variants are discussed.
On the Modelling of Constraints with Tractable Logical Operators
Ruiwei Wang (National University of Singapore), Roland H. C. Yap
🎯 What it does: This paper studies a framework of Constraint Representation Pair, exploring the implementation of decidable negation operations within the constraint representation. Two new sub-forms, op-BCT and op-OMVD, are proposed under this framework, proving that they are more concise than traditional OMDD while supporting decidable logical operations (∧, ∨, ¬).
On the Power of Convolution-Augmented Transformer
Mingchen Li (University of Michigan), Samet Oymak (University of Michigan)
TransformerText
🎯 What it does: This paper proposes a Convolution-Enhanced Transformer (CAT) architecture, which embeds convolutional filters into the keys, values, and queries of the attention layer to enhance memory, copying, and length generalization capabilities.
On the Power of Randomization for Obviously Strategy-Proof Mechanisms
Shiri Ron (Weizmann Institute of Science), Daniel Schoepflin (Rutgers University)
Optimization
🎯 What it does: Designed and analyzed various randomized obvious strategy proof (OSP) mechanisms to achieve approximate social welfare maximization in multi-auction environments (multi-unit auctions, additive auctions, unit demand auctions).
On the Power of Strategic Corpus Enrichment in Content Creation Games
Haya Nachimovsky (Technion - Israel Institute of Technology), Moshe Tennenholtz (Technion - Israel Institute of Technology)
Recommendation SystemOptimization
🎯 What it does: This paper proposes achieving stable multi-query content competition by adding a small number of fixed virtual documents to the corpus of search/recommendation systems, and demonstrates that it can achieve pure Nash equilibrium and high user welfare.
On the Relationship Between Monotone and Squared Probabilistic Circuits
Benjie Wang (University of California, Los Angeles), Guy Van den Broeck (University of California, Los Angeles)
ImageTabular
🎯 What it does: Proposes Inception PCs (Deep Sum-Square-Sum Structure), unifying and extending the expressive power of monotonic PCs and square PCs;
On the Robustness of Distributed Machine Learning Against Transfer Attacks
Sebastien Andreina (NEC Labs Europe), Ghassan Karame (Ruhr University Bochum)
ClassificationOptimizationFederated LearningAdversarial AttackImage
🎯 What it does: The paper studies a distributed machine learning framework that simultaneously distributes model training and inference across multiple nodes, introducing multi-dimensional heterogeneity in data, architecture, optimizer, scheduler, etc., and evaluates its robustness against transfer attacks.
On the Trainability and Classical Simulability of Learning Matrix Product States Variationally
Afrad Basheer (University of Technology Sydney), Hakop Pashayan (Freie Universität Berlin)
Physics Related
🎯 What it does: This paper studies the trainability and classical simulability when using the matrix product state (MPS) ansatz for quantum state approximation learning in variational quantum algorithms. It proves that global observables lead to barren plateaus, while local observables can avoid them; and through theoretical derivation and numerical simulation, it shows that under local observables, the MPS ansatz has an effective subspace, which may provide possibilities for classical simulation with low quantum resources.
One for Dozens: Adaptive REcommendation for All Domains with Counterfactual Augmentation
Huishi Luo (Beihang University), Deqing Wang (Beihang University)
Recommendation SystemMixture of ExpertsTabular
🎯 What it does: This study proposes a multi-domain recommendation framework called AREAD, which can handle dozens of domains and enhances recommendation performance through hierarchical expert fusion, expert mask pruning, and popularity-based counterfactual augmentation.
One Node One Model: Featuring the Missing-Half for Graph Clustering
Xuanting Xie (University of Electronic Science and Technology of China), Wenyu Chen (University of Electronic Science and Technology of China)
Graph Neural NetworkContrastive LearningGraph
🎯 What it does: A graph clustering framework based on 'one model per node' (FPGC) is proposed, which selects clustering-related features through the squeeze-excitation mechanism and enhances low-order feature interactions using feature crossing.
One Token Can Help! Learning Scalable and Pluggable Virtual Tokens for Retrieval-Augmented Large Language Models
Yutao Zhu (Renmin University of China), Ji-Rong Wen (Renmin University of China)
GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A pluggable and extensible virtual token method called SPRING is designed to enhance the performance of LLM in retrieval-augmented generation (RAG) scenarios while maintaining the general generation capability of the original model.
One-Shot Reference-based Structure-Aware Image to Sketch Synthesis
Rui Yang (Shaanxi Normal University), Xiaojun Wu (Shaanxi Normal University)
Image TranslationGenerationData SynthesisDiffusion modelImage
🎯 What it does: This study proposes a training-free reference style sketch generation method called Ref2Sketch-SA, which can automatically synthesize sketches that retain structure and conform to the reference style based on a given content image and a single reference sketch.
OneBatchPAM: A Fast and Frugal K-Medoids Algorithm
Antoine de Mathelin (Universite Paris-Saclay), Nicolas Vayatis (Universite Paris-Saclay)
OptimizationComputational EfficiencyImageTabular
🎯 What it does: A OneBatchPAM algorithm is proposed to quickly estimate the k-medoids objective using a single batch (size O(log n)), significantly reducing computational and storage costs.
Online and Streaming Algorithms for Constrained k-Submodular Maximization
Fabian Christian Spaeh (Boston University), Huy Nguyen (Northeastern University)
Optimization
🎯 What it does: An online/streaming algorithm with a single pass is proposed to maximize any (including non-monotonic) k-submodular function under cardinality and knapsack constraints.
Online Fraud Detection via Test-Time Retrieval-Based Representation Enrichment
Yiran Qiao (Chinese Academy of Sciences), Xiang Ao
Anomaly DetectionTransformerTabularSequentialRetrieval-Augmented Generation
🎯 What it does: To address the issue of concept drift in online fraud detection, a Test-time Retrieval Enhanced Representation (TRE) method is proposed. It constructs a retrieval database using real-time test data and dynamically enhances input representations during the inference phase through nearest neighbor retrieval, weighted aggregation, and an adaptive k mechanism, thereby improving the model's robustness and accuracy.
Online Guidance Graph Optimization for Lifelong Multi-Agent Path Finding
Hongzhi Zang (Tsinghua University), Jiaoyang Li (Monash University)
OptimizationConvolutional Neural NetworkGraph
🎯 What it does: An online guidance strategy is proposed, which dynamically updates the guidance map using real-time traffic information to enhance the throughput of lifelong multi-agent path planning.
Online Learning of Coalition Structures by Selfish Agents
Saar Cohen (Bar Ilan University), Noa Agmon (Bar Ilan University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes an algorithm for self-interested agents to learn alliance structures online, aiming to achieve approximately Nash stable allocations through repeated feedback under unknown preferences.
Online MDP with Prototypes Information: A Robust Adaptive Approach
Shuo Sun (University of California Berkeley), Zuo-Jun Max Shen (University of California Berkeley)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies online robust Markov decision processes (MDP) and designs a robust learning algorithm RPO-AAS using a known finite number of transition kernel prototypes, and proposes a non-robust algorithm NRPO-NPC based on recent prototypes.
Online Nonsubmodular Optimization with Delayed Feedback in the Bandit Setting
Sifan Yang (Nanjing University), Lijun Zhang (Nanjing University)
OptimizationTabular
🎯 What it does: This paper studies online non-stochastic optimization in a bandit setting with delayed feedback and proposes three new algorithms (DBGD-NF, DOGD-NF, BDBGD-NF) to reduce regret.
Online Prompt Selection for Program Synthesis
Yixuan Li (University of Edinburgh), Elizabeth Polgreen (University of Edinburgh)
OptimizationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper studies an online learning framework called CYANEA, which dynamically selects the best LLM and prompt combination or symbolic solver to address program synthesis tasks using a multi-armed bandit approach.
OODML: Whole Slide Image Classification Meets Online Pseudo-Supervision and Dynamic Mutual Learning
Tingting Zheng (Harbin Institute of Technology), Zhongyuan Wang (Wuhan University)
ClassificationTransformerImageMagnetic Resonance Imaging
🎯 What it does: The OODML (Online Pseudo-Supervision and Dynamic Mutual Learning) framework is proposed, which enhances model performance for whole slide image (WSI) classification through online pseudo-label generation and mutual learning of feature representations.
OOTDiffusion: Outfitting Fusion Based Latent Diffusion for Controllable Virtual Try-On
Yuhao Xu (Xiao-i Research), Arlene Chen (Xiao-i Research)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A virtual try-on method OOTDiffusion based on latent diffusion models is proposed, which can generate realistic try-on images while maintaining details.
Open Models, Closed Minds? On Agents Capabilities in Mimicking Human Personalities through Open Large Language Models
Lucio La Cava (University of Calabria), Andrea Tagarelli (University of Calabria)
Large Language ModelPrompt EngineeringText
🎯 What it does: This study investigates the inherent personalities of 12 open-source LLMs in MBTI and BFI personality assessments, as well as their imitation abilities under conditional prompts and role prompts.
Open-Set Cross-Network Node Classification via Unknown-Excluded Adversarial Graph Domain Alignment
Xiao Shen (Hainan University), Xi Zhou (Hainan University)
ClassificationDomain AdaptationGraph Neural NetworkGenerative Adversarial NetworkGraph
🎯 What it does: This paper proposes a framework called UAGA for open cross-network node classification, aimed at completing node classification and unknown detection when the target network contains unknown categories not present in the source network.
Open-Set Heterogeneous Domain Adaptation: Theoretical Analysis and Algorithm
Thai-Hoang Pham (Ohio State University), Ping Zhang (Ohio State University)
Domain AdaptationRepresentation LearningImageText
🎯 What it does: This paper proposes the Open Set Heterogeneous Domain Adaptation (OSHeDA) problem and provides corresponding theoretical analysis and a representation learning-based algorithm called RLOSHeDA.
Open-world Radio Frequency Fingerprint Identification via Augmented Semi-supervised Learning
Zehua Han (Beihang University), Wenrui Ding (Beihang University)
ClassificationRecognitionTransformerContrastive LearningTime Series
🎯 What it does: Under open world conditions, a self-supervised pre-training based radio frequency fingerprint identification framework called OpenRFI is proposed, along with a dedicated Roinformer feature extraction model.
OpenViewer: Openness-Aware Multi-View Learning
Shide Du (Fuzhou University), Wenzhong Guo (Fuzhou University)
Explainability and InterpretabilityRepresentation LearningImage
🎯 What it does: A multi-view learning framework for open environments, OpenViewer, is proposed to address interpretability and generalization issues.
OpenVIS: Open-vocabulary Video Instance Segmentation
Pinxue Guo (Fudan University), Wenqiang Zhang (Fudan University)
Object DetectionObject TrackingSegmentationRecurrent Neural NetworkTransformerVision Language ModelContrastive LearningVideo
🎯 What it does: The InstFormer framework is proposed to achieve Open Vocabulary Video Instance Segmentation (OpenVIS), capable of simultaneously detecting, segmenting, and tracking instances of any category in videos.
Operationalising Rawlsian Ethics for Fairness in Norm Learning Agents
Jessica Woodgate (University of Bristol), Nirav Ajmeri (University of Bristol)
Reinforcement Learning
🎯 What it does: Proposes the RAWL E agent, which combines Rawls' maximin principle and uses reward shaping to enable reinforcement learning agents to learn and generate ethical norms in multi-agent harvesting scenarios, enhancing social welfare and fairness;
Optimal and Efficient Binary Questioning for Accelerated Annotation
Franco Marchesoni-Acland (Universite Paris-Saclay), Gabriele Facciolo (City University of Hong Kong)
OptimizationImage
🎯 What it does: The paper proposes a fast annotation framework based on binary question answering, utilizing a predictor to estimate label probabilities, modeling the annotation process as a Huffman coding problem, and providing a feasible approximate lookahead + heuristic algorithm for rapid annotation.
Optimal Auction Design for Mixed Bidders
Xiaohui Bei (Nanyang Technological University), Xiang Yan (Huawei)
Optimization
🎯 What it does: This study investigates the optimal auction design for mixed bidders, particularly focusing on auction mechanisms that consider both utility maximizers (UMs) and value maximizers (VMs).
Optimal Bounds for Dissatisfaction in Perpetual Voting
Alexander Kozachinskiy (Centro Nacional de Inteligencia Artificial), Tomasz Steifer (Institute of Fundamental Technological Research)
Optimization
🎯 What it does: This paper studies how to design a voting method in continuous voting to ensure that the number of disappointments for each voter remains within a small range.
Optimal Classification Trees for Continuous Feature Data Using Dynamic Programming with Branch-and-Bound
Cătălin E. Brița (University of Amsterdam), Emir Demirović (Delft University of Technology)
ClassificationOptimizationComputational EfficiencyTabular
🎯 What it does: This paper proposes a dynamic programming and branch-and-bound algorithm called ConTree, which constructs optimal classification trees on continuous feature data within a given size constraint.
Optimal Control Operator Perspective and a Neural Adaptive Spectral Method
Mingquan Feng (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationTabularOrdinary Differential Equation
🎯 What it does: An optimal control method based on the perspective of Instance-Solution Control Operator is proposed, and a new neural operator architecture, Neural Adaptive Spectral Method (NASM), is implemented for direct one-time solving of OCP.
Optimally Solving Simultaneous-Move Dec-POMDPs: The Sequential Central Planning Approach
Johan Peralez (University Lyon), Jilles S. Dibangoye (University of Groningen)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes re-modeling the traditional Dec-POMDP with simultaneous movement into a sequentially moving Dec-POMDP, and under this framework, designs a new centralized training and decentralized execution algorithm (oSARSA seq) to achieve the scalability of ϵ-optimal solutions.
Optimising Spatial Teamwork Under Uncertainty
Gregory Everett (University of Southampton), Sarvapali D. Ramchurn (Sentient Sports)
OptimizationRobotic IntelligenceGraph Neural NetworkReinforcement LearningTime Series
🎯 What it does: A method for evaluating and optimizing multi-agent teamwork based on spatial coordination is proposed, utilizing Multi-Agent Markov Decision Processes (MMDP) combined with Monte Carlo Tree Search (MCTS) and linear programming to make dynamic decisions on player actions in football defense, thereby reducing opponent threats.
Optimize Incompatible Parameters Through Compatibility-aware Knowledge Integration
Zheqi Lv (National University of Singapore), Fei Wu (National University of Singapore)
Recommendation SystemOptimizationText
🎯 What it does: A compatibility-aware knowledge integration (CKI) method is constructed through compatibility assessment and parameter splicing, optimizing incompatible parameters using the complementary parameters of multiple pre-trained models without adding extra parameters.
Optimized Gradient Clipping for Noisy Label Learning
Xichen Ye (Shanghai University), Cheng Jin (Fudan University)
OptimizationImage
🎯 What it does: A dynamic threshold-based gradient clipping method called OGC is proposed to suppress the negative impact of noisy labels on gradient distribution during the training process.
Optimizing Human Pose Estimation Through Focused Human and Joint Regions
Yingying Jiao (Jilin University), Zhuoyue Xu (Zhejiang Gongshang University)
Pose EstimationTransformerVideo
🎯 What it does: This paper proposes a dual-stream framework VREMD, which significantly improves the accuracy of video human pose estimation by utilizing human-keypoint masks enhancement and a bidirectional motion decoupling module.
Optimizing Label Assignment for Weakly Supervised Person Search
Haiyang Zhu (Xidian University), Nannan Wang (Xidian University)
Object DetectionOptimizationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A weakly supervised person search label optimization framework is proposed, integrating context-aware clustering, optimal transport-based prototype matching, and dual memory pool enhancement.
Optimizing Quantized Diffusion Models via Distillation with Cross-Timestep Error Correction
Yanxi Li (University of Sydney), Chengbin Du (University of Sydney)
GenerationOptimizationKnowledge DistillationDiffusion modelImage
🎯 What it does: The study proposes a Cross-Time Step Error Correction (CTEC) technique to quantify the error accumulation problem in diffusion models.
Orchestrating the Symphony of Prompt Distribution Learning for Human-Object Interaction Detection
Mingda Jia (Peking University), Yun Zheng (Alibaba Group)
RecognitionObject DetectionTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes InterProDa, a method for HOI detection based on prompt distribution learning. By treating the prompt of each category as a Gaussian distribution, it achieves nearly infinite modeling of intra-class patterns and captures inter-class dependencies, thereby enhancing the expressive power of detection queries.
Orpheus: Engineering Multiagent Systems via Communicating Agents
Matteo Baldoni (Università degli Studi di Torino), Amit K. Chopra (Lancaster University)
Agentic AI
🎯 What it does: This paper proposes the Orpheus programming model, which combines information protocols with cognitive programming to achieve a high-level abstraction of multi-agent systems.
OT-StainNet: Optimal Transport Driven Semantic Matching for Weakly Paired H&E-to-IHC Stain Transfer
Xianchao Guan (Harbin Institute of Technology), Yongbing Zhang (Harbin Institute of Technology)
Image TranslationGenerationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a virtual IHC staining method called OT-StainNet, which is based on optimal transport-driven semantic matching, capable of achieving accurate staining transfer in weakly paired images where H&E and IHC are difficult to align at the pixel level.
OTIAS: OcTree Implicit Adaptive Sampling for Multispectral and Hyperspectral Image Fusion
Shangqi Deng (Xi'an Jiaotong University), Ping Wei (Xi'an Jiaotong University)
Image TranslationRestorationData SynthesisImage
🎯 What it does: An implicit neural representation based on octrees, called OTIAS, is proposed for the fusion of multispectral and hyperspectral images.
OTLRM: Orthogonal Learning-based Low-Rank Metric for Multi-Dimensional Inverse Problems
Xiangming Wang (Harbin Institute of Technology), Guoqing Chao (Harbin Institute of Technology)
RestorationGenerative Adversarial NetworkImageVideoMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A low-rank t-SVD generative model based on learnable orthogonal transformations is proposed for the recovery of multi-dimensional inverse problems (such as tensor completion, CASSI reconstruction, and MSI denoising).
OTPNet: ODE-inspired Tuning-free Proximal Network for Remote Sensing Image Fusion
Wei Yu (Harbin Institute of Technology), Xin Sun (Harbin Institute of Technology)
RestorationSuper ResolutionImageOrdinary Differential Equation
🎯 What it does: A parameter-free approximation network based on ODE, OTPNet, is proposed for remote sensing image fusion.
Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion
Jingyuan Chen (University of Rochester), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisDiffusion modelVideoBenchmark
🎯 What it does: A parameter-free long video generation framework called Ouroboros-Diffusion is proposed, which improves the denoising strategy of the traditional FIFO-Diffusion to enhance the consistency of video content.
OUS: Bridging Scene Context and Facial Features to Overcome the Rigid Cognitive Problem
Xinji Mai (Fudan University), Wenqiang Zhang (Fudan University)
RecognitionRecurrent Neural NetworkTransformerPrompt EngineeringVideoMultimodality
🎯 What it does: A dynamic facial expression recognition method OUS is proposed, which integrates scene context with facial features to address cognitive biases caused by traditional methods neglecting the scene.
Out of Length Text Recognition with Sub-String Matching
Yongkun Du (Fudan University), Yu-Gang Jiang (Beijing Jiaotong University)
RecognitionTransformerSupervised Fine-TuningImageTextBenchmark
🎯 What it does: This work proposes a long text recognition method based on substring matching called SMTR, and constructs the first long text benchmark LTB to address the issue of recognizing long texts when trained solely on short text datasets.
Out-of-Distribution Detection with Prototypical Outlier Proxy
Mingrong Gong (Shenzhen University), Hui Huang (Great Bay University)
Anomaly DetectionImage
🎯 What it does: Proposes the Prototypical Outlier Proxy (POP) framework, which enhances OOD detection by reshaping the ID/OOD decision boundary through the introduction of virtual OOD prototypes without the need for real or synthetic outlier samples.
Out-of-Distribution Generalization on Graphs via Progressive Inference
Yiming Xu (Xi'an Jiaotong University), Chen Chen (University of Virginia)
Domain AdaptationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposes the GPro model, which learns causal invariant substructures in graph structures through multi-step evolutionary reasoning to achieve OOD generalization;
OV-DQUO: Open-Vocabulary DETR with Denoising Text Query Training and Open-World Unknown Objects Supervision
Junjie Wang (Harbin Institute of Technology), Yong Xu (Chongqing Research Institute of HIT)
Object DetectionTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes the OV-DQUO framework, which effectively alleviates the confidence bias in open vocabulary detection through open-world pseudo-labeling, wildcard matching, and denoised text query training.
Overcoming Heterogeneous Data in Federated Medical Vision-Language Pre-training: A Triple-Embedding Model Selector Approach
Aowen Wang (Zhejiang University), Shiting Wen (NingboTech University)
Federated LearningSafty and PrivacyTransformerVision Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: The PMS-FM framework is proposed to address data heterogeneity and privacy issues in medical audiovisual language pre-training, achieving cross-institutional federated pre-training through a personalized model selector and triple embedding.
p-Mean Regret for Stochastic Bandits
Anand Krishna (National University of Singapore), Vincent Y. F. Tan (National University of Singapore)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a new performance metric for Markov Bandits (MAB) called p-mean regret, which can balance fairness and efficiency.
PA3Fed: Period-Aware Adaptive Aggregation for Improved Federated Learning
Chengxiang Huang (Beijing University of Posts and Telecommunications), Bingyan Liu (Beijing University of Posts and Telecommunications)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: Proposes the PA Fed 3 method, which enhances model accuracy and stability in federated learning by analyzing the fluctuations during the training process of individual clients and utilizing periodic-aware adaptive aggregation.
Paid with Models: Optimal Contract Design for Collaborative Machine Learning
Bingchen Wang (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
OptimizationFederated Learning
🎯 What it does: This study investigates the optimal contract design using models as rewards in collaborative machine learning.
PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation
Sarthak Kumar Maharana (University of Texas at Dallas), Yunhui Guo (University of Texas at Dallas)
Domain AdaptationOptimizationImage
🎯 What it does: A continuous testing time adaptive learning rate (CTTA) method named PALM is proposed, which can automatically select the layers that need fine-tuning and dynamically adjust their learning rates without relying on pseudo-labels, thereby enhancing the model's robustness in domain shift environments.
Pamba: Enhancing Global Interaction in Point Clouds via State Space Model
Zhuoyuan Li (University of Science and Technology of China), Tianzhu Zhang (Sangfor Technologies Inc.)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: Pamba is proposed, a global interactive point cloud semantic segmentation framework that utilizes Mamba (state space model) to achieve linear complexity long-range modeling across the entire point cloud.
PanAdapter: Two-Stage Fine-Tuning with Spatial-Spectral Priors Injecting for Pansharpening
RuoCheng Wu, Liang-Jian Deng (Xi'an Jiaotong University)
Image TranslationRestorationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A two-stage parameter-efficient fine-tuning framework called PanAdapter is proposed, which utilizes the spatial-spectral prior information from a pre-trained image restoration model to achieve high-quality image fusion.
PanoDiT: Panoramic Videos Generation with Diffusion Transformer
Muyang Zhang (Institute of Automation, Chinese Academy of Sciences), Xiaopeng Zhang (Hengyang Normal University)
GenerationData SynthesisTransformerDiffusion modelVideoText
🎯 What it does: A PanoDiT framework based on Diffusion Transformer is proposed to generate high-quality 360° panoramic videos from text prompts.
Parallel Greedy Best-First Search with a Bound on Expansions Relative to Sequential Search
Takumi Shimoda (University of Tokyo), Alex Fukunaga (University of Tokyo)
OptimizationTabularBenchmark
🎯 What it does: A parallel greedy best-first search (GBFS) algorithm named OBAT is proposed and further improved to OBAT S (combining separate generation and evaluation techniques). OBAT ensures that only one 'bench' is expanded at a time under the premise of sharing the Open list, thus achieving an upper bound on the number of expansions compared to sequential GBFS.
Parallel-Learning of Invariant and Tempo-variant Attributes of Single-Lead Cardiac Signals: PLITA
Adrian Atienza (Technical University of Denmark), Sadasivan Puthusserypady (Technical University of Denmark)
ClassificationRecognitionTransformerContrastive LearningTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: The PLITA method is proposed, utilizing parallel learning to capture the invariant and time-varying attributes of single-lead ECG signals.
Parameterized Complexity of Caching in Networks
Robert Ganian (TU Wien), Dimitra Tsigkari (Telefónica Scientific Research)
🎯 What it does: This paper conducts a systematic analysis of the parameterized complexity of the network cache allocation problem, providing results on decidability, FPT, XP, and W[1]-hard/paraNP-hard under various parameters.
Parametric ρ-Norm Scaling Calibration
Siyuan Zhang (Jiangnan University), Linbo Xie (Jiangnan University)
OptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A new ρ-Norm Scaling post-processing calibration method is proposed, and a multi-layer objective function is designed to enhance the model's confidence calibration performance.
Pareto Continual Learning: Preference-Conditioned Learning and Adaption for Dynamic Stability-Plasticity Trade-off
Song Lai (City University of Hong Kong), Gaofeng Meng (Chinese Academy of Sciences)
ClassificationOptimizationConvolutional Neural NetworkImageSequential
🎯 What it does: This paper proposes Pareto Continual Learning (ParetoCL), which treats the stability-plasticity trade-off in continual learning as a multi-objective optimization problem, using a preference-conditioned model to learn within a single network and dynamically adapt to different trade-offs during inference.
Pareto Set Learning for Multi-Objective Reinforcement Learning
Erlong Liu (Nanjing University), Chao Qian (Nanjing University)
Reinforcement Learning
🎯 What it does: The PSL-MORL method is proposed, utilizing Pareto Set Learning and hypernetworks to generate personalized policy networks for any preference weights, achieving complete Pareto front coverage in multi-objective RL.
ParGo: Bridging Vision-Language with Partial and Global Views
An-Lan Wang (ByteDance China), Wei-Shi Zheng (Sun Yat-sen University)
GenerationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: The ParGo project is proposed, which combines visual features through partial views and global views to project them into the language model space, better bridging the visual encoder and large language model.
ParseCaps: An Interpretable Parsing Capsule Network for Medical Image Diagnosis
Xinyu Geng (Harbin Institute of Technology), Jun Xu (Harbin Institute of Technology)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes ParseCaps, a capsule network that achieves interpretable medical image diagnosis through a parse tree structure.
Partial Label Causal Representation Learning for Instance-Dependent Supervision and Domain Generalization
Yizhi Wang (Southeast University), Min-Ling Zhang (University of Newcastle)
ClassificationDomain AdaptationRepresentation LearningAuto EncoderContrastive LearningImage
🎯 What it does: The study utilizes causal representation learning to achieve robustness in classification and domain transfer within instance-dependent partial label learning (IDPLL).
Partial Point Cloud Registration with Multi-view 2D Image Learning
Yue Zhang (Xidian University), Biao Hou (Xidian University)
RecognitionSegmentationOptimizationTransformerContrastive LearningImagePoint Cloud
🎯 What it does: Proposes the IAPReg framework, which fuses 2D features generated from multi-view depth maps with 3D features extracted by KPConv to achieve partial point cloud registration;
Partially Blinded Unlearning: Class Unlearning for Deep Networks from Bayesian Perspective
Subhodip Panda (Indian Institute of Science), Prathosh A.P. (Indian Institute of Science)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a single-step class-level unlearning method (Partially-Blinded Unlearning, PBU) that utilizes only the data of the class to be forgotten to eliminate the information of that class from a pre-trained network while maintaining the performance of other classes.
ParZC: Parametric Zero-Cost Proxies for Efficient NAS
Peijie Dong (Hong Kong University of Science and Technology), Xiaowen Chu (Hong Kong University of Science and Technology)
OptimizationNeural Architecture SearchTransformerImage
🎯 What it does: Proposes the Parametric Zero-Cost Proxies (ParZC) framework, which utilizes node-level zero-cost statistics and a parameterizable mixed model to enhance the ranking quality of NAS.
PAT: Pruning-Aware Tuning for Large Language Models
Yijiang Liu (Nanjing University), Li Du (Nanjing University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A Pruning-Aware Tuning (PAT) framework is proposed, which combines structured pruning and fine-tuning by inserting Hybrid Sparsification Modules (HSM) between the attention and feedforward networks, and implementing a unified sparse mask to achieve dynamic pruning of the hidden dimensions of large language models.
Patch-level Sounding Object Tracking for Audio-Visual Question Answering
Zhangbin Li (Hefei University of Technology), Dan Guo (Hefei University of Technology)
RecognitionObject TrackingGraph Neural NetworkImageTextMultimodalityAudio
🎯 What it does: A Patch-level Sounding Object Tracking (PSOT) method is proposed, which uses a three-stage graph neural network to track and filter visual patches related to answers based on visual motion, audio correspondence, and question semantics.
PatentLMM: Large Multimodal Model for Generating Descriptions for Patent Figures
Shreya Shukla (Indian Institute of Technology Jodhpur), Anand Mishra (Microsoft)
GenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents a dataset of patent images and their brief and detailed descriptions, PATENTDESC-355K, consisting of 355K samples, and builds a patent image description generation model, PATENTLMM, based on this dataset.
Path-Adaptive Matting for Efficient Inference Under Various Computational Cost Constraints
Qinglin Liu (Harbin Institute of Technology), Shengping Zhang (Harbin Institute of Technology)
SegmentationOptimizationComputational EfficiencyKnowledge DistillationImage
🎯 What it does: This paper proposes the Path-Adaptive Matting (PAM) framework, which can dynamically select network paths based on image context and FLOP constraints to achieve efficient inference;