AAAI 2026 Papers — Page 27
AAAI Conference on Artificial Intelligence · 4149 papers
On Logical Extrapolation for Mazes with Recurrent and Implicit Networks
Brandon Knutson (Colorado School of Mines), Daniel McKenzie (Colorado School of Mines)
Explainability and InterpretabilityRecurrent Neural Network
🎯 What it does: Investigated the logical extrapolation capabilities of recursive and implicit networks in maze-solving tasks, examined whether they truly learned scalable algorithms, and analyzed their potential convergence dynamics.
On Modality Weighting and Specificity for Multi-Modal Entity Alignment
Yu Xing (Nanjing University), Tieke He (Nanjing University)
Knowledge DistillationRepresentation LearningGraph Neural NetworkTransformerMixture of ExpertsContrastive LearningMultimodality
🎯 What it does: Propose a multi-modal entity alignment framework HUMEA based on hierarchical Mixture of Experts (MoE) and unimodal distillation, which can adaptively weight different modalities while preserving modality-specific information.
On Model and Data Scaling for Skeleton-based Self-Supervised Gait Recognition
Adrian Cosma (National University of Science and Technology POLITEHNICA Bucharest), Emilian Radoi (National University of Science and Technology POLITEHNICA Bucharest)
RecognitionHyperparameter SearchTransformerContrastive LearningVideoGraph
🎯 What it does: In this paper, the authors conduct experimental research on the scalability of skeleton-based self-supervised gait recognition models, quantifying the impact of data volume, model size, and computational resources on zero-shot recognition performance.
On Robustness of Linear Classifiers to Targeted Data Poisoning
Nakshatra Gupta, Venkatesh R (IIT Bombay)
ClassificationAdversarial AttackImageTextTabularBenchmark
🎯 What it does: This paper studies the robustness against targeted data poisoning under linear classifiers, proposes a black-box attack model, and defines robustness metrics. It proves that solving robustness is NP-complete and provides efficient methods for computing upper and lower bounds.
On Stealing Graph Neural Network Models
Marcin Podhajski (Institute of Fundamental Technological Research Polish Academy of Sciences), Tomasz Paweł Michalak (Institute of Fundamental Technological Research Polish Academy of Sciences)
Safty and PrivacyAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: Propose a strategy in GNN model stealing attacks under extreme query budgets, first locally acquiring the encoder and then utilizing query selection to maximize information extraction.
On the Approximation Ratio of Optimal Fixed-Price Mechanisms for Single and Multi-Unit Bilateral Trade
Giordano Giambartolomei (King's College London), Bart de Keijzer (King's College London)
Optimization
🎯 What it does: This paper improves mathematical programming techniques to re-estimate the approximate ratio of optimal social welfare under fixed-price mechanisms in single-item and multi-item bilateral trade, enhancing the lower and upper bounds for single items, and providing better lower bounds for multi-items (especially two items) and improvements in symmetric cases.
On the Calibration of Image Semi-Supervised Learning Models
Mehrab Mustafy Rahman (University of Illinois Chicago), Cornelia Caragea (University of Illinois Chicago)
ClassificationImage
🎯 What it does: Proposes CalibrateMix, a targeted mixup strategy based on training dynamics and sample difficulty, to enhance confidence calibration and classification accuracy in semi-supervised learning models.
On the Edge of Core (Non-)Emptiness: An Automated Reasoning Approach to Approval-Based Multi-Winner Voting
Ratip Emin Berker (Carnegie Mellon University), Lirong Xia (University of Adelaide)
Optimization
🎯 What it does: This paper studies the existence of core stability in approval-based multi-elections and proposes a method based on mixed integer linear programming (MILP) to determine whether the core is empty.
On the Evaluation of Capability Estimation Methods for Large Language Models
Qiang Hu (Tianjin University), Yongqiang Lyu (University of Luxembourg)
Large Language ModelTextBenchmark
🎯 What it does: This paper proposes AEBench, an unlabeled capability estimation benchmark for large language models (LLMs), covering 12 AutoEval methods.
On the Impact of Weight Quantization on Deep Neural Network Uncertainty
Shuang Liang (Nanjing University), Shao-Qun Zhang (Nanjing University)
Explainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Studied the impact of weight quantization on uncertainty in deep neural networks, proposed the Exact Moment Propagation (EMP) estimator, and designed the Moment Alignment (MOMA) optimization method.
On the Information Processing of One-Dimensional Wasserstein Distances with Finite Samples
Cheongjae Jang (Hanyang University), Yung-Kyun Noh (Seoul National University)
Time SeriesBiomedical Data
🎯 What it does: This paper theoretically analyzes and empirically validates the data information processing capability of the one-dimensional Wasserstein distance under finite samples, demonstrating its ability to simultaneously capture point differences (rate) and support differences.
On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD
Tongcheng Zhang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: Studied the learning dynamics of stochastic gradient descent (SGD) with label noise in two-layer linear networks, revealing a two-phase transition from lazy to rich regimes, and generalized this idea to Sharpness-Aware Minimization (SAM).
On the Misalignment Between Data Learnability and Forgettability in Machine Unlearning
Zijie Pan (City University of Macau), Wanlei Zhou (City University of Macau)
ClassificationImageText
🎯 What it does: Investigate the mismatch between the learning process and the difficulty of later forgetting, propose Unlearning Gradient Sensitivity (UGS) and Learnability-Forgettability Divergence (LFD) metrics, and design Dual-Aware Training (DAT) to align the two during training;
On the Probabilistic Learnability of Compact Neural Network Preimage Bounds
Luca Marzari (University of Verona), Alessandro Farinelli (University of Verona)
Safty and PrivacyReinforcement Learning from Human FeedbackBenchmark
🎯 What it does: Propose a probabilistic method called RF-ProVe based on random forests and active resampling to approximate the preimage boundary of neural networks with high confidence.
On the Robustness of Bandit Multiple Testing
Zhengyu Zhou (Wuhan University), Weiwei Liu (Wuhan University)
Optimization
🎯 What it does: This paper proposes a robust Bandit framework for multiple testing under adversarial contamination, and presents two adaptive sampling strategies (based on trimmed mean and e-process) to achieve anytime control of FDR and improve TPR.
On the Superimposed Noise Accumulation Problem in Sequential Knowledge Editing of Large Language Models
Ding Cao (University of Science and Technology of China), Guangzhong Sun (University of Science and Technology of China)
TransformerLarge Language ModelText
🎯 What it does: Studied the problem of cumulative noise accumulation in sequential knowledge editing of large language models, and proposed the DeltaEdit method to reduce noise and improve editing success rates.
OncoCoT: A Temporal-causal Chain-of-Thought Dataset for Oncologic Decision-Making
Peiru Yang (Tsinghua University), Yongfeng Huang (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataElectronic Health RecordsBenchmarkChain-of-Thought
🎯 What it does: Constructed the OncoCoT dataset, a long-chain-of-thought dataset tailored for tumor diagnosis and treatment, along with its benchmark OncoEval.
One for All: Synthesis-Free Fingerprint Learning for Attribution of In-the-Wild Synthetic Images
Jianwei Fei (University of Florence), Alessandro Piva (University of Florence)
RecognitionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a synthesis-free open-source method for attributing the origin of synthetic images, which uses a random encoder-decoder to simulate fingerprints of trillions of generative models, and learns frequency-domain fingerprint features to verify unknown models.
One-Shot Refiner: Boosting Feed-forward Novel View Synthesis via One-Step Diffusion
Yitong Dong (Zhejiang University), Guofeng Zhang (Hangzhou VIVO Information Technology Co Ltd)
GenerationConvolutional Neural NetworkTransformerSupervised Fine-TuningDiffusion modelGenerative Adversarial NetworkGaussian SplattingImage
🎯 What it does: Propose an end-to-end framework combining a dual-domain detail perception module and a feature-guided one-step diffuser for generating high-quality novel views from sparse, unposed images.
One-Step Generative Policies with Q-Learning: A Reformulation of MeanFlow
Zeyuan Wang (National University of Defense Technology), Yanwei Fu (Fudan University)
Reinforcement LearningFlow-based ModelBenchmark
🎯 What it does: Introduce a first-order generation strategy, achieving a one-step direct mapping from noise to actions through residual reconstruction of MeanFlow, compatible with Q-learning in offline reinforcement learning.
One2Seq: One-Token Wise Decoder for Efficient Scene Text Recognition
Zhibin Ma (Sun Yat-sen University), Xugong Qin (Shenzhen University)
RecognitionTransformerImageText
🎯 What it does: Propose a one-token-wise decoder called One2Seq, which performs autoregressive decoding using a single context token to address issues such as attention drift, slow decoding speed, and lack of global context in traditional AR decoders.
OnEDIT: Online Editing with Decoupled Implicit Task for Large Language Models
Chae-Won Lee (Hanyang University), Joon-Hyuk Chang (Hanyang University)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a method called OnEDIT that decouples online editing from implicit tasks for continuous instruction tuning without task IDs.
OneFont: A Unified Agent for End-to-End Font Creation
Yingxin Lai (Xiamen University), Shaozi Li (Xiamen University)
GenerationGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringDiffusion modelGenerative Adversarial NetworkImageTextBenchmark
🎯 What it does: Propose OneFont, a unified end-to-end font generation agent capable of parsing user intent through free-text dialogue and automatically scheduling glyph synthesis and refinement modules;
OneLIP: Unlocking and Improving Long-Text Representations of CLIP via One-Stage Adaptation
Renjie Pan (Shanghai Jiao Tong University), Hua Yang (Shanghai Jiao Tong University)
ClassificationRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Design a one-stage long-text CLIP adaptation framework called OneLIP, enabling CLIP to directly process texts of arbitrary length while maintaining performance on short texts.
OneSug: The Unified End-to-End Generative Framework for E-commerce Query Suggestion
Xian Guo (Kuaishou Technology), Han Li (Kuaishou Technology)
GenerationRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringText
🎯 What it does: Propose a unified end-to-end generative framework called OneSug for query suggestions in e-commerce search.
Online Capacitated General Matching with Knapsack
Ruoyu Wu (University of Sydney), Hequn Wang (University of Sydney)
OptimizationGraph
🎯 What it does: This paper proposes an online capacity general matching with knapsack (OCGMK) problem and designs a new online capacity-knapsack allocation (OCKA) algorithm to maximize rewards without knowing future arrival orders and reward distributions.
Online Conformal Selection with Accept-to-Reject Changes
Kangdao Liu (University of Macau), Hongxin Wei (Southern University of Science and Technology)
TextTabularBiomedical Data
🎯 What it does: Propose an online acceptable-reject modification conformal selection method OCS-ARC, satisfying the ARC constraint of irrevocably selected candidates and controlling FDR
Online Cross-Modal Hashing with Expanding Label Space
Wentao Fan (Nanjing University), Huaxiong Li (Nanjing University)
RetrievalMultimodality
🎯 What it does: Designed an online multi-modal hashing method called OH-ELS, which can simultaneously handle sample growth and category expansion to achieve cross-modal retrieval;
Online Fair Allocations with Binary Valuations and Beyond
Yuanyuan Wang (University of Macau), Tianze Wei (City University of Hong Kong)
Optimization
🎯 What it does: This paper studies an online fair allocation model where the future item arrival order is completely unknown. For goods and chores, it proposes deterministic algorithms that achieve approximate EF1, MMS, and USW under different valuation functions (submodular binary, personalized binary, etc.), and provides matching lower bounds and infeasibility results.
Online Linear Regression with Paid Stochastic Features
Nadav Merlis (Technion Israel Institute of Technology), Nicolò Cesa-Bianchi (Technion Israel Institute of Technology)
Optimization
🎯 What it does: This paper studies the online linear regression problem where features are corrupted by noise and learners can pay to reduce noise, designing an algorithm that simultaneously determines the payment level and linear predictor, and provides the optimal degradation rate.
Online Multi-LLM Selection via Contextual Bandits Under Unstructured Context Evolution
Manhin Poon (City University of Hong Kong), Jinhang Zuo (City University of Hong Kong)
OptimizationTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Studies online multi-LLM selection, proposing sequential decision-making using context-robust bandits under unstructured context evolution;
Online Multi-Relational Clustering with Dominant View Mining
Zhengzhong Zhu (Sichuan University), Jiangping Zhu (Sichuan University)
Graph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose an end-to-end online multi-relational graph clustering framework OMC-DVM, achieving clustering through dynamic main view mining and unified contrastive learning;
Online Robust Planning Under Model Uncertainty: A Sample-Based Approach
Tamir Shazman, Vadim Indelman (Technion Israel Institute of Technology)
OptimizationReinforcement Learning
🎯 What it does: Propose a new online robust planning algorithm RSS (Robust Sparse Sampling) for real-time decision-making in Markov Decision Processes (MDP) under model uncertainty.
OnlineBootKNN: An Unsupervised Framework for Detecting Anomalies in Spectral Data Streams
Nicolas Rojas Varela (University Clermont Auvergne), Engelbert Mephu Nguifo (University Clermont Auvergne)
Anomaly DetectionAuto EncoderTime SeriesPhysics Related
🎯 What it does: Studies unsupervised anomaly detection in spectral data streams, first evaluating existing methods on multi-dimensional table streams, then proposing and implementing a framework named OnlineBootKNN specifically for real-time detection of spectral anomalies.
Open-World 3D Scene Graph Generation for Retrieval-Augmented Reasoning
Yu Fei (Liaoning University of Technology), Lechao Cheng (Liaoning University of Technology)
Object DetectionGenerationRetrievalLarge Language ModelVision Language ModelImageTextMultimodalityGraphRetrieval-Augmented Generation
🎯 What it does: Developed an open-world 3D scene graph generation and retrieval-enhanced reasoning framework capable of dynamically constructing object and relationship graphs from multi-frame RGB-D sequences, and supporting four interactive tasks (text/image question answering, visual localization, instance retrieval, and task planning) through vector retrieval.
Open-World Deepfake Attribution via Confidence-Aware Asymmetric Learning
Haiyang Zheng (University of Trento), Zhun Zhong (University of Trento)
ClassificationAnomaly DetectionSupervised Fine-TuningImageBenchmark
🎯 What it does: This paper proposes a framework named Confidence-Aware Asymmetric Learning (CAL) for simultaneously identifying known and unknown forgery methods in open-world deepfake attribution tasks.
Open-World Object Counting in Videos
Niki Amini-Naieni (University of Oxford), Andrew Zisserman (University of Oxford)
Object DetectionObject TrackingVision Language ModelVideoText
🎯 What it does: The study proposes an open-world video object counting task, where the input is a video along with text or visual examples, and the output is unique instance counts for each frame and the entire video.
OpenDriveVLA: Towards End-to-end Autonomous Driving with Large Vision Language Action Model
Xingcheng Zhou (Technical University of Munich), Alois Knoll (Technical University of Munich)
Autonomous DrivingTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: Developed an end-to-end autonomous driving vision-language action model based on a large language model, capable of generating feasible driving trajectories from multi-perspective images, 3D instance perception, vehicle status, and language instructions.
OpenScan: A Benchmark for Generalized Open-Vocabulary 3D Scene Understanding
Youjun Zhao, Rynson W. H. Lau (City University of Hong Kong Hong Kong University of Science and Technology)
SegmentationVision Language ModelMultimodalityPoint CloudBenchmark
🎯 What it does: Studied the extended open vocabulary 3D scene understanding task (GOV-3D), constructed a large-scale benchmark named OpenScan containing eight language attribute dimensions, and systematically evaluated the performance of existing OV-3D models on attribute understanding.
OPERA: A Reinforcement Learning--Enhanced Orchestrated Planner-Executor Architecture for Reasoning-Oriented Multi-Hop Retrieval
Yu Liu (Chinese Academy of Sciences), Zhiyuan Ma (Huazhong University of Science and Technology)
RetrievalLarge Language ModelReinforcement LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a multi-agent planning-execution architecture called OPERA for multi-hop retrieval tasks, which can dynamically decompose problems, rewrite queries, and filter retrieval results.
OPLoRA: Orthogonal Projection LoRA Prevents Catastrophic Forgetting During Parameter-Efficient Fine-Tuning
Yifeng Xiong (University of California Irvine), Xiaohui Xie (University of California Irvine)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a new parameter-efficient fine-tuning method called OPLoRA, which introduces bilateral orthogonal projection into the low-rank updates of LoRA to prevent interference between the updates and the principal singular subspace of pre-trained weights, thereby avoiding catastrophic forgetting.
Opt3DGS: Optimizing 3D Gaussian Splatting with Adaptive Exploration and Curvature-Aware Exploitation
Ziyang Huang (Wuhan University), Shunping Ji (Wuhan University)
OptimizationComputational EfficiencyGaussian SplattingImageStochastic Differential Equation
🎯 What it does: Propose the Opt3DGS framework, which enhances the rendering quality of 3D Gaussian Splatting by performing global exploration through adaptive weighted SGLD, followed by fine convergence guided by local quasi-Newton directions using Adam.
OptiHive: Ensemble Selection for LLM-Based Optimization via Statistical Modeling
Maxime Bouscary (Massachusetts Institute of Technology), Saurabh Amin (Massachusetts Institute of Technology)
OptimizationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language Model
🎯 What it does: The OptiHive framework leverages LLM to simultaneously parallelly generate candidate solvers, problem instances, and verification tests, filtering out interpretable components and then selecting the optimal optimization solver through statistical inference.
Optimal Look-back Horizon for Time Series Forecasting in Federated Learning
Dahao Tang (University of Sydney), Dong Yuan (University of Sydney)
Data SynthesisFederated LearningTime Series
🎯 What it does: Propose a theoretical framework for adaptively selecting the look-back horizon in time series prediction under federated learning scenarios, combining synthetic data generator (SDG) and intrinsic representation space;
Optimal Welfare in Noncooperative Network Formation Under Attack
Natan Doubez (École Polytechnique), Marcus Wunderlich (University of Augsburg)
Optimization
🎯 What it does: This paper studies a non-cooperative network formation model in the presence of attackers and immune mechanisms, analyzing the social welfare of networks formed by self-interested agents under different attack strategies, and provides tight bounds for these welfare levels;
Optimally Auditing Adversarial Agents
Sanmay Das (Virginia Polytechnic Institute and State University), Yuang Zhang (George Mason University)
OptimizationReinforcement Learning
🎯 What it does: This study designs and solves an optimal audit strategy within a master-slave game framework to maximize the leader's utility or social welfare when adversarial agents select the worst equilibrium;
Optimization and Robustness-Informed Membership Inference Attacks for LLMs
Zichen Song, Yao Shu (Sungkyunkwan University)
OptimizationSafty and PrivacyAdversarial AttackLarge Language ModelText
🎯 What it does: Propose a membership inference attack framework called OR-MIA, which leverages gradient norm and perturbation robustness features to attack large language models.
Optimization Method for Surrogate Function in Spiking Neural Networks Based on Membrane Potential Distribution
Qi Sun (Xidian University), Biao Hou (Xidian University)
OptimizationComputational EfficiencySpiking Neural NetworkImage
🎯 What it does: Propose an adaptive Surrogate optimization framework (MPO) based on membrane potential distribution, enhancing gradient stability and energy efficiency of SNN training through dynamically aligning Surrogate width with membrane potential variance, regularizing membrane potential distribution, and compensating discretization errors.
Optimization of Multi-Agent Flying Sidekick Traveling Salesman Problem over Road Networks
Ruixiao Yang (Massachusetts Institute of Technology), Chuchu Fan (Massachusetts Institute of Technology)
OptimizationGraphTabular
🎯 What it does: This paper proposes a MA-FSTSP model for multi-vehicle and multi-UAV collaborative delivery, and designs a three-phase integrated algorithm to solve this NP-hard problem.
Optimized Algorithms for Text Clustering with LLM-Generated Constraints
Chaoqi Jia (RMIT University), Kok-Leong Ong (Western Sydney University)
OptimizationTransformerLarge Language ModelText
🎯 What it does: Propose a text clustering framework based on LLM automatically generating set-based must-link and cannot-link constraints, and design a penalty-based local search clustering algorithm to utilize these constraints.
Optimized Distortion in Linear Social Choice
Luise Ge (Washington University in St. Louis), Yevgeniy Vorobeychik (Washington University in St. Louis)
OptimizationTabular
🎯 What it does: The study investigates distortion in social choice under a linear utility model, proposes new voting rules (such as Maximum Coordinate Majority (MCP), Linear Stable Lottery (LSLR), and Pure Stable Lottery (PSLR)), designs instance-optimal deterministic and randomized algorithms, and evaluates them on real-world data.
Optimizing LoRA Allocation of MoE with the Alignment of Topic Correlation
Hengyuan Xu (Southeast University), Zijie Xu (Southeast University)
OptimizationRepresentation LearningGraph Neural NetworkLarge Language ModelMixture of ExpertsText
🎯 What it does: Propose a LoRA allocation framework called TopicLoRA based on topic priors, which uses a topic knowledge graph to guide the expert routing in Mixture-of-Experts (MoE), and achieves semantic consistency through KL alignment;
OPTION: An Online Pricing Strategy for Asynchronous Federated Learning Against Free-Riding Attacks
Bangqi Pan (Wuhan University of Science and Technology), Guanghui Wen (Wuhan University of Science and Technology)
OptimizationFederated LearningImage
🎯 What it does: This paper proposes an online pricing strategy called OPTION, aiming to effectively suppress free-riding attacks and enhance system fairness and model performance by dynamically pricing model download costs and upload rewards in asynchronous federated learning (AFL).
OptMark: Robust Multi-bit Diffusion Watermarking via Inference Time Optimization
Jiazheng Xing (Zhejiang University), Mike Zheng Shou (National University of Singapore)
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: In images generated by diffusion models, copyright protection and user tracking are achieved by optimizing to insert visible multi-bit watermarks into intermediate latent variables during inference;
OptScale: Probabilistic Optimality for Inference-time Scaling
Youkang Wang (PolySmart Group), Xiao-Yong Wei (Sichuan University)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes the OPTSCALE framework for inference time scaling based on probabilistic optimality, which dynamically determines the optimal number of samples to enhance the inference performance of large language models.
OR-R1: Automating Modeling and Solving of Operations Research Optimization Problem via Test-Time Reinforcement Learning
Zezhen Ding (Hong Kong University of Science and Technology), Tianlong Chen (University of North Carolina)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Propose the OR-R1 framework, which uses a small amount of labeled data for Supervised Fine-Tuning (SFT) and further trains on unlabeled data through Test-Time Group Relative Policy Optimization (TGRPO), automatically completing the modeling and solving of operations research optimization problems.
ORACLE: Optimizing Reasoning Abilities of Large Language Models via Constraint-Led Synthetic Data Elicitation
Zhuojie Yang (Sun Yat-sen University), Keze Wang (Sun Yat-sen University)
OptimizationData-Centric LearningTransformerSupervised Fine-TuningTextBenchmark
🎯 What it does: Designed and implemented the ORACLE framework, generating and verifying high-quality training data for multi-step reasoning through templated prompts and a symbolic reasoning engine to enhance the reasoning capabilities of large language models.
Order-Preserving Dimension Reduction for Multimodal Semantic Embedding
Chengyu Gong (New York University), Dongfang Zhao (Pacific Northwest National Laboratory)
Computational EfficiencyRepresentation LearningTransformerMultimodality
🎯 What it does: Propose the Order-Preserving Dimension Reduction (OPDR) method to reduce high-dimensional multimodal embeddings while preserving the kNN ordering.
Ordered Local Momentum for Asynchronous Distributed Learning Under Arbitrary Delays
Chang-Wei Shi (Nanjing University), Wu-Jun Li (Nanjing University)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new asynchronous distributed learning algorithm called OrLoMo, which achieves efficient training of Momentum SGD (MSGD) under local updates in environments with arbitrary delays.
Ordered Objectives in Maximum Satisfiability
Jeremias Berg (University of Helsinki), Matti Järvisalo (University of Helsinki)
OptimizationBenchmark
🎯 What it does: This paper studies ordered objectives appearing in Maximum Satisfiability (MaxSAT) instances, systematically defines the concepts of ordered and approximately ordered objectives, and explores their structural properties and impacts on core-based fundamental solving algorithms;
Ordinal Secretaries with Advice
Hasti Nourmohammadi (University of Alberta), Xiaoqi Tan (University of Alberta)
Optimization
🎯 What it does: Studied the ordinal secretary problem with position prediction, proposed deterministic and randomized waiting and accepting algorithms, and extended to multi-choice and re-employment variants.
OrgaCast: A Trustworthy Spatiotemporal Diffusion Model for Fluorescence Organoid Forecasting
Dawei Gao (University of North Texas), Yunhe Feng (University of North Texas)
GenerationExplainability and InterpretabilityConvolutional Neural NetworkTransformerDiffusion modelImageMultimodalityTime SeriesBiomedical Data
🎯 What it does: Propose a multimodal conditional diffusion model called OrgaCast for predicting the temporal development of fluorescent microscopy images derived from cardiac-like embryonic stem cells and generating corresponding uncertainty confidence maps.
Organ-Aware Routing Mixture-of-Retrieval Augmented Generation for Fetal Ultrasound Reporting
Bin Pu (Hunan University), Kenli Li (Southern Medical University)
GenerationAnomaly DetectionTransformerMixture of ExpertsVision Language ModelContrastive LearningBiomedical DataUltrasoundRetrieval-Augmented Generation
🎯 What it does: Proposed the FetusR dataset and the ORM-RAG model for multi-organ fetal ultrasound report generation;
Orion: Steering Personalized Web Agents via Global-Micro Profiling and Adaptive Intent Tracking
Die Hu (Chinese Academy of Sciences), Bingzhen Wu (Chinese Academy of Sciences)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AITextRetrieval-Augmented Generation
🎯 What it does: This paper proposes Orion, a framework that enhances the personalization understanding and interaction efficiency of large language model (LLM)-driven web agents through global-micro personalization configuration and adaptive intent tracking.
ORTCL: Towards Continual Learning of Time Series Foundation Models on Streaming Data via Orthogonal Rotation
Li Lin (Southeast University), Kaiwen Xia (Southeast University)
Representation LearningTime Series
🎯 What it does: Propose the ORTCL method, which achieves lossless continual learning on time series base models using an orthogonal rotation matrix;
Orthogonal Spatial-temporal Distributional Transfer for 4D Generation
Wei Liu (Anhui University of Finance and Economics), Wynne Hsu (National University of Singapore)
GenerationData SynthesisKnowledge DistillationDiffusion modelGaussian SplattingVideoPoint Cloud
🎯 What it does: This paper proposes a framework that transfers 3D spatial priors and video temporal priors separately into a 4D diffusion model, achieving high-quality 4D content generation.
ORVIT: Near-Optimal Online Distributionally Robust Reinforcement Learning
Debamita Ghosh (University of Central Florida), Yue Wang (University of Central Florida)
Reinforcement LearningTabular
🎯 What it does: Online learning of distributionally robust Markov decision processes in a single unknown environment, proposing the f-ORVIT algorithm to optimize performance in extreme scenarios.
Oscillation Inversion: Training-Free Image and Video Enhancement Through Oscillated Latents in Large Flow Models
Yan Zheng (University of Texas at Austin), Zhangyang Wang (University of Texas at Austin)
RestorationGenerationFlow-based ModelRectified FlowAuto EncoderImageVideo
🎯 What it does: Investigate the oscillatory behavior in large-scale flow models during the inversion process, proposing the Oscillation Inversion method to achieve training-agnostic image and video enhancement and editing.
OscuFit: Learning to Fit Osculating Implicit Quadrics for Point Clouds
Rao Fu (Nanyang Technological University), Jianmin Zheng (Nanyang Technological University)
OptimizationGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a learning-based framework for estimating local surface differential properties of point clouds, utilizing osculating implicit quadrics to simultaneously output normals and curvature.
OSVBench: Benchmarking LLMs on Specification Generation Tasks for Operating System Verification
Shangyu Li, Jiasi Shen (Georgia Institute of Technology)
GenerationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose the OSVBENCH benchmark to evaluate large language models in generating complete formal specifications required for correctness verification of operating system kernel functions;
OT-ALD: Aligning Latent Distributions with Optimal Transport for Accelerated Image-to-Image Translation
Zhanpeng Wang (Dalian University of Technology), Zhongxuan Luo (Dalian University of Technology)
Image TranslationDiffusion modelScore-based ModelImage
🎯 What it does: OT-ALD proposes to match the potential distributions of the source and target domains via optimal transport, achieving efficient and higher quality image-to-image translation.
OTARo: Once Tuning for All Precisions Toward Robust On-Device LLMs
Shaoyuan Chen (Houmo AI), Qiang Wu (Houmo AI)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose the OTARo scheme, achieving a unified single fine-tuning to obtain LLMs supporting multiple bit-widths, addressing the non-reusability of traditional quantization when switching between different precisions.
Other Vehicle Trajectories Are Also Needed: A Driving World Model Unifies Ego-Other Vehicle Trajectories in Video Latent Space
Jian Zhu (Li Auto Inc), Xianpeng Lang (Li Auto Inc)
GenerationData SynthesisAutonomous DrivingTransformerDiffusion modelAuto EncoderWorld ModelVideo
🎯 What it does: Proposed the EOT-WM world model, which unifies the representation of ego vehicle and other vehicles' trajectories in video space, achieving controllable video generation and future scene prediction
OTI: A Model-free and Visually Interpretable Measure of Image Attackability
Jiaming Liang (University of Macau), Chi-Man Pun (University of Macau)
SegmentationExplainability and InterpretabilityAdversarial AttackImageBiomedical Data
🎯 What it does: Propose a model-free and visually interpretable image adversarial susceptibility metric called OTI, which directly evaluates the degree to which an image is vulnerable to adversarial perturbations based on the texture intensity and area ratio of semantic objects.
Otter: Mitigating Background Distractions of Wide-Angle Few-Shot Action Recognition with Enhanced RWKV
Wenbo Huang (Southeast University), Miki Haseyama (Southeast University)
RecognitionTransformerVideo
🎯 What it does: Investigate the background interference problem in wide-view videos for few-shot action recognition, and propose the Otter model, which enhances the subject through the Compound Segmentation Module (CSM) and reconstructs temporal relationships via the Temporal Reconstruction Module (TRM) to improve performance.
OursFed: Provable Group Fairness-Aware Federated Learning Against Distrust and Fragility
Yun Xin (Wuhan University of Science and Technology), Kehao Wang (Wuhan University of Technology)
Federated LearningTabular
🎯 What it does: Propose the OursFed framework to address group fairness (GF) in federated learning, ensuring fairness guarantees in scenarios of low trust and data vulnerability.
Out-of-Context Misinformation Detection via Variational Domain-Invariant Learning with Test-Time Training
Xi Yang (Xidian University), Hong Han (Xidian University)
Domain AdaptationAnomaly DetectionTransformerLarge Language ModelVision Language ModelAuto EncoderContrastive LearningMultimodality
🎯 What it does: Proposes a framework combining variational domain-invariant learning with test-time training (VDT) for cross-domain detection of out-of-context (OOC) misinformation in the news domain.
Out-of-Distribution Detection with Positive and Negative Prompt Supervision Using Large Language Models
Zhixia He (Tianjin University), Linlin Yu (University of Arkansas)
Anomaly DetectionGraph Neural NetworkLarge Language ModelPrompt EngineeringImageMultimodality
🎯 What it does: This paper proposes a three-stage PNPS framework, which first generates class-specific positive and negative prompts using a large language model (LLM), then optimizes the prompts through learnable text and visual matrices, and finally aggregates the semantic supervision of the prompts via a cross-modal graph neural network, propagating them to visual features to enhance image OOD detection performance.
Out-of-Distribution Generalization with a SPARC: Racing 100 Unseen Vehicles with a Single Policy
Bram Grooten (Sony AI), Peter R. Wurman (Sony AI)
Domain AdaptationAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningReinforcement LearningSequential
🎯 What it does: Propose a single-stage adaptive reinforcement learning framework, SPARC, for achieving robust control in out-of-distribution (OOD) environments.
Outlier Matters: Efficient Long-to-Short Reasoning via Outlier-Guided Model Merging
Qiyuan Zhu (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
Computational EfficiencyKnowledge DistillationLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed the ORCA (Outlier-aware Reasoning Conciseness Adaptive Merge) framework, which achieves efficient fusion of base models and reasoning models through activation outlier analysis in large models, significantly shortening the Chain-of-Thought (CoT) output length while maintaining or even improving reasoning accuracy.
OW-DAR: Dual-Granularity Adaptive Reconstruction-Error Modeling for Open-World Object Detection
Linhua Ye (South China University of Technology), Ronghua Luo (South China University of Technology)
Object DetectionAnomaly DetectionConvolutional Neural NetworkImageBenchmark
🎯 What it does: Developed an open-world object detection framework named OW-DAR based on dual-grained reconstruction error modeling.
OWL: Unsupervised 3D Object Detection by Occupancy Guided Warm-up and Large Model Priors Reasoning
Xusheng Guo (Xiamen University), Chenglu Wen (Xiamen University)
Object DetectionAutonomous DrivingLarge Language ModelPoint Cloud
🎯 What it does: Propose the OWL framework, achieving unsupervised 3D object detection through Occupancy-Guided Preheating (OGW), Instance-Guided Reasoning (ICR), and Weight Adaptive Self-training (WAS).
OwlCap: Harmonizing Motion-Detail for Video Captioning via HMD-270K and Caption Set Equivalence Reward
Chunlin Zhong (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
GenerationLarge Language ModelReinforcement LearningVision Language ModelVideoTextMultimodality
🎯 What it does: Propose OwlCap through two-phase HMD-270K dataset construction with CSER reward-based reinforcement learning to address the imbalance between motion and detail in video captioning.
OX-MABSR: A Benchmark for Open-domain Explainable Multimodal Aspect-Based Sentiment Reasoning
Xinjing Liu (Beijing University of Posts and Telecommunications), Ruifan Li (Beijing University of Posts and Telecommunications)
Explainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose an open-domain explainable multi-modal aspect-based sentiment reasoning task (OX-MABSR) and construct a corresponding high-quality dataset (OX-MABSR-Bench), while designing a multi-modal LLM framework (MABSR-LLM) to accomplish open-vocabulary aspect-sentiment pair prediction, dual-layer (perceptual-cognitive) explanation generation, and reasoning path generation.
Oxytrees: Model Trees for Bipartite Learning
Pedro Ilídio (KU Leuven), Celine Vens (Universidade de SãO Paulo)
Drug DiscoveryBiomedical Data
🎯 What it does: Proposed the Oxytrees model tree, which uses a proxy matrix to compress the interaction matrix, significantly accelerating the training and inference of bilateral learning.
P-SLCR: Unsupervised Point Cloud Semantic Segmentation via Prototypes Structure Learning and Consistent Reasoning
Lixin Zhan (National University of Defense Technology), Xuehu Duan (National University of Defense Technology)
SegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes an unsupervised point cloud semantic segmentation framework called P-SLCR, which achieves consistent structural learning and inference through the construction of a learnable prototype library.
P2S: Probabilistic Process Supervision for General-Domain Reasoning Question Answering
Wenlin Zhong, Kun Kuang (Worcester Polytechnic Institute)
Large Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: In open-domain reasoning tasks without verifiable rewards, Probabilistic Process Supervision (P2S) is proposed, providing large language models with dense process-level rewards through a self-supervised probabilistic approach.
PA-FAS: Towards Interpretable and Generalizable Multimodal Face Anti-Spoofing via Path-Augmented Reinforcement Learning
Yingjie Ma (Shenzhen University), Zitong Yu (Great Bay University)
Anomaly DetectionExplainability and InterpretabilityLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Propose the PA-FAS framework, combining reasoning path enhancement and answer shuffling, to achieve interpretable and generalizable learning for multi-modal face anti-spoofing under limited annotations.
Pacing Equilibria in Second-Price Auctions with Few Buyers
Yonglei Yan (Beijing Institute of Technology), Zhengyang Liu (Beijing Institute of Technology)
OptimizationFinance Related
🎯 What it does: This paper proposes a polynomial-time algorithm that can accurately compute the second-price surge equilibrium (SPPE) in a second-price auction market with a fixed number of buyers;
PADiff: Predictive and Adaptive Diffusion Policies for Ad Hoc Teamwork
Hohei Chan, Mengchen Zhao (South China University Of Technology)
Robotic IntelligenceReinforcement LearningDiffusion model
🎯 What it does: Propose PADiff, an adaptive prediction strategy based on diffusion models, to address the problem of immediate collaboration with unknown teammates in AHT (Ad Hoc Teamwork).
PAGE: A Unified Approach for Federated Graph Unlearning
Yuming Ai (Beijing Institute of Technology), Guoren Wang (Beijing Institute of Technology)
Federated LearningSafty and PrivacyComputational EfficiencyKnowledge DistillationGraph Neural NetworkGenerative Adversarial NetworkGraphBenchmark
🎯 What it does: Achieve multi-scenario graph data forgetting (Meta Unlearning and Client Unlearning) in a federated graph learning environment
PAGPL: Privacy-Aware Graph Prompt Learning Scheme via Adaptive Perturbation-Estimated Topology Recovery
Ju Jia (Southeast University), Guang Cheng (Southeast University)
Safty and PrivacyRepresentation LearningGraph Neural NetworkPrompt EngineeringContrastive LearningGraph
🎯 What it does: A privacy-aware graph prompt learning framework named PAGPL is developed for efficient graph prompt learning on differentially private perturbed graphs.
Pairing-free Group-level Knowledge Distillation for Robust Gastrointestinal Lesion Classification in White-Light Endoscopy
Qiang Hu (Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology), Zhiwei Wang (Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology)
ClassificationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningBiomedical Data
🎯 What it does: This paper proposes PaGKD, a group-level knowledge distillation framework that does not require paired images, to transfer diagnostic information from narrow-band imaging (NBI) to white-light imaging (WLI) models, thereby improving the performance of gastrointestinal lesion classification.
Palimpsest: Reconciling the CISS Trilemma for Incremental Nuclei Segmentation
Jiajia Li (Shenzhen University), Huisi Wu (Shenzhen University)
SegmentationBiomedical Data
🎯 What it does: Proposes the Palimpsest framework for incremental nucleus segmentation without requiring examples, while maintaining the model size unchanged.
PANDA – Patch and Distribution-Aware Augmentation for Long-Tailed Exemplar-Free Continual Learning
Siddeshwar Raghavan (Purdue University), Fengqing Zhu (Purdue University)
Representation LearningData-Centric LearningTransformerVision Language ModelImage
🎯 What it does: Proposes a Patch-and-Distribution-Aware Augmentation (PANDA) framework to address the two-tier imbalance problem in continuous learning without sample replay.
Panda: Test-Time Adaptation with Negative Data Augmentation
Ruxi Deng (University of Illinois Urbana-Champaign), Jingrui He (University of Illinois Urbana-Champaign)
Domain AdaptationImage
🎯 What it does: Proposed a test-time adaptation method called Panda, which suppresses prediction bias caused by image distortion through negative data augmentation (NDA), and can be easily integrated into existing TTA frameworks.
PanFlow: Decoupled Motion Control for Panoramic Video Generation
Cheng Zhang (Monash University), Jianfei Cai (Monash University)
GenerationTransformerSupervised Fine-TuningDiffusion modelOptical FlowVideo
🎯 What it does: Developed a PanFlow framework based on motion decoupling for generating 360° videos with precise motion control from a single image and optical flow conditions.
PanFoMa: A Lightweight Foundation Model and Benchmark for Pan-Cancer
Xiaoshui Huang (Shanghai Jiao Tong University), Jian Zhang (Jiangxi University of Finance and Economics)
Drug DiscoveryTransformerMultimodalityBiomedical DataBenchmark
🎯 What it does: Propose a lightweight hybrid model called PanFoMa, which processes whole-transcriptome single-cell data through a two-stage approach combining local Transformer and global Mamba, and constructs a benchmark named PanFoMaBench containing 3.5 million cells across 34 cancer types.
Pano-GS: Perception-Aware Gaussian Optimization with Gradient Consistency and Multi-Criteria Densification for High-Quality Rendering
Yang Deng (Peking University), Ronggang Wang (Peking University)
GenerationOptimizationGaussian SplattingImage
🎯 What it does: Propose Pano-GS, a 3D Gaussian splatting optimization framework oriented towards perception, which enhances texture details and rendering quality through gradient consistency constraint loss, multi-criteria densification strategies, and stochastic perturbation exploration.
PanoNav: Mapless Zero-Shot Object Navigation with Panoramic Scene Parsing and Dynamic Memory
Qunchao Jin (Hong Kong University of Science and Technology), Changhao Chen (Hong Kong University of Science and Technology)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: Propose a fully RGB-only, map-free, open-vocabulary goal navigation framework called PanoNav, which leverages panoramic perspectives and multimodal large language models (LLMs) for scene parsing, and enhances target localization efficiency through a dynamic bounded memory queue to guide decision-making.