AAAI 2026 Papers — Page 15
AAAI Conference on Artificial Intelligence · 4149 papers
Fine-flow Distilling Coarse-flow Video Generation for Long-Term Driving World Model
Xiaodong Wang (Peking University), Peixi Peng (Peking University)
GenerationAutonomous DrivingTransformerDiffusion modelFlow-based ModelAuto EncoderWorld ModelVideoMultimodality
🎯 What it does: This paper designs and implements a hierarchical long video generation world model, which first learns coarse-grained prediction of large motions (Coarse DiT) and fine-grained prediction of continuous detailed motions (Fine DiT). Subsequently, it uses the fine-grained video stream as a self-supervised signal to distill the coarse-grained flow, ultimately achieving high-quality, temporally consistent driving scene video generation.
Fine-Grained DINO Tuning with Dual Supervision for Face Forgery Detection
Tianxiang Zhang (Jinan University), Hui Gao (Jinan University)
ClassificationAnomaly DetectionTransformerSupervised Fine-TuningMixture of ExpertsImageVideo
🎯 What it does: To address DeepFake detection, this paper proposes a lightweight DeepFake Fine-Grained Adapter (DFF-Adapter), which inserts multi-head LoRA adapters into each Transformer block of the frozen DINOv2 vision Transformer. It jointly trains authenticity discrimination and forgery type classification, achieving fine-grained forgery feature extraction and detection.
Fine-Grained Generalization via Structuralizing Concept and Feature Space into Commonality, Specificity and Confounding
Zhen Wang (Hebei University of Technology), Wenlong Yu (Tianjin University)
ClassificationDomain AdaptationRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a framework named CFSG for fine-grained domain generalization, with the core idea of simultaneously decomposing the concept space and feature space into three subspaces: public, specific, and confusing, and dynamically adapting to varying degrees of domain shift by weighted fusion of the three during inference.
Fine-grained Image Quality Assessment for Perceptual Image Restoration
Xiangfei Sheng (Xidian University), Leida Li (Xidian University)
RestorationVision Language ModelContrastive LearningImage
🎯 What it does: Constructed the FGRestore fine-grained image quality assessment dataset and proposed the FGResQ model for fine-grained evaluation of image restoration quality.
Fine-Grained Image Retrieval via Dual-Vision Adaptation
Xin Jiang (Nanjing University of Science and Technology), Zechao Li (Hong Kong Polytechnic University)
RetrievalKnowledge DistillationRepresentation LearningTransformerImage
🎯 What it does: This paper proposes a Dual Visual Adapter (DVA) framework that achieves fine-grained image retrieval by leveraging a frozen pre-trained ViT model through sample and feature co-adaptation;
Fine-Grained Representation for Lane Topology Reasoning
Guoqing Xu (Chinese Academy of Sciences), Yang Yang (Chinese Academy of Sciences)
Autonomous DrivingTransformerImage
🎯 What it does: Propose an end-to-end fine-grained lane topology reasoning framework called TopoFG, which leverages multi-scale BEV features, position priors, and sequence priors to achieve joint prediction of lane centerlines and topology relationships.
Fine-grained Uncertainty Decomposition in Large Language Models: A Spectral Approach
Nassim Walha (German Cancer Research Center), Florian Buettner (German Cancer Research Center)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: The paper proposes the Spectral Uncertainty framework, which uses von Neumann entropy and functional Bregman information to perform a fine-grained decomposition of aleatoric and epistemic uncertainty in large language models' predictions.
Fine-Tuned LLMs Know They Don’t Know: A Parameter-Efficient Approach to Recovering Honesty
Zeyu Shi (Beihang University), Jianxin Li (Beihang University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose a parameter-efficient recovery method called Honesty-Critical Neurons Restoration (HCNR) to address the decline in honesty of large language models (LLMs) after supervised fine-tuning (SFT), by locating and repairing critical neurons to restore the model's expression of its own knowledge boundaries.
FINE: Factorized Multimodal Sentiment Analysis via Mutual INformation Estimation
Yadong Liu (University of Science and Technology of China), Shangfei Wang (University of Science and Technology of China)
RecognitionRepresentation LearningTransformerMixture of ExpertsContrastive LearningMultimodality
🎯 What it does: Proposes the FINE framework, which decomposes and denoises multimodal emotional representations through mutual information estimation, achieving more robust emotion recognition.
FineRef: Fine-Grained Error Reflection and Correction for Long-Form Generation with Citations
Yixing Peng (University of Science and Technology of China), Quan Wang (Shenzhen University)
GenerationRetrievalTransformerSupervised Fine-TuningReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: Propose the FineRef framework, leveraging a fine-grained error reflection and correction mechanism to enhance the accuracy of long-form citation generation.
FineTec: Fine-Grained Action Recognition Under Temporal Corruption via Skeleton Decomposition and Sequence Completion
Dian Shao (Northwestern Polytechnical University), Like Liu (Northwestern Polytechnical University)
RecognitionGraph Neural NetworkGraphSequential
🎯 What it does: FineTec proposes a complete framework for achieving fine-grained action recognition in skeletal sequences severely damaged over time;
FineVAU: A Novel Human-Aligned Benchmark for Fine-Grained Video Anomaly Understanding
Joao Alexandre Cardeira Pereira (NOVA LINCS), David Semedo (NOVA LINCS)
Anomaly DetectionLarge Language ModelVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Propose the FineVAU benchmark, decomposing video anomaly understanding into three-dimensional evaluation (What, Who, Where) and introducing the FV-Score evaluation metric.
FineXtrol: Controllable Motion Generation via Fine-Grained Text
Keming Shen (Shenzhen University), Linlin Shen (Shenzhen University)
GenerationTransformerVision Language ModelDiffusion modelContrastive LearningTextSequential
🎯 What it does: This study proposes the FineXtrol framework, which generates controllable human motion sequences by jointly utilizing fine-grained text descriptions (control signals) and coarse-grained text;
FinMathBench: A Formula-Driven Benchmark for Evaluating LLMs’ Math Reasoning Capabilities in Finance
Yi He (Ant Group), Haixiang Hu (Ant Group)
TransformerLarge Language ModelTextBenchmarkFinance RelatedChain-of-Thought
🎯 What it does: Proposed a formula-driven, dynamically generated financial mathematics reasoning question-answering benchmark called FinMathBench.
FinMMDocR: Benchmarking Financial Multimodal Reasoning with Scenario Awareness, Document Understanding, and Multi-Step Computation
Zichen Tang, Haocheng Gao (Beijing University of Posts and Telecommunications)
TransformerAgentic AIPrompt EngineeringVision Language ModelTextMultimodalityBenchmarkFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed FinMMDocR, a financial multimodal reasoning benchmark containing multi-scenario, long documents, and multi-step calculations; evaluated the performance of multimodal large language models (MLLMs) and retrieval-augmented generation (RAG) methods on this benchmark.
FinRpt: Dataset, Evaluation System and LLM-based Multi-agent Framework for Equity Research Report Generation
Song Jin (Renmin University of China), Rui Yan (Renmin University of China)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AITextBenchmarkFinance Related
🎯 What it does: Proposed the task of automatically generating Equity Research Reports (ERR), constructed the FinRpt benchmark dataset and evaluation system, and developed the multi-agent framework FinRpt-Gen for ERR generation.
Firing Bits Where It Matters: Spiking-Guided Just Recognizable Distortion Modeling for Machine-Centric Video Coding
Wuyuan Xie (Shenzhen University), Miaohui Wang (Shenzhen University)
CompressionComputational EfficiencySpiking Neural NetworkImage
🎯 What it does: Construct a fine-grained machine-recognizable distortion (JRD) dataset and propose an encoder based on spiking neural networks (SNN), combining three-branch feature decoupling, multi-scale redundancy removal, and spike attention aggregation to achieve efficient QP prediction.
FIRM-MoE:Fine-GrainedExpert Decomposition for Resource-Adaptive MoE Inference
Keyu Chen (Zhejiang University), Shibo He (Zhejiang University of Technology)
Computational EfficiencyLarge Language ModelMixture of ExpertsText
🎯 What it does: Propose a framework called FIRM-MoE for efficiently inferring Mixture-of-Experts (MoE) large language models on memory-constrained edge devices.
First Learn, Then Review: Human-Like Continual Learning for Cross-View Geo-Localization with Limited Field of View
Lei Cheng (Southeast University), Teng Wang (Southeast University)
RetrievalKnowledge DistillationConvolutional Neural NetworkMixture of ExpertsContrastive LearningImage
🎯 What it does: Proposed the HCL-Geo two-stage human-like continual learning framework, addressing the cross-view geolocation task under limited perspectives and unknown directions.
First-Order Error Matters: Accurate Compensation for Quantized Large Language Models
Xingyu Zheng (Beihang University), Xianglong Liu (ETH Zurich)
CompressionTransformerTextBenchmark
🎯 What it does: Propose a new post-training quantization method called FOEM, which improves quantization error compensation by explicitly incorporating first-order gradient information, thereby enhancing the compression efficiency of large language models.
First-Order Representation Languages for Goal-Conditioned RL
Simon Ståhlberg (RWTH Aachen University), Hector Geffner (RWTH Aachen University)
Graph Neural NetworkReinforcement LearningGraphBenchmark
🎯 What it does: Proposes three HER-based goal relabeling methods (State HER, Propositional HER, Lifted HER), and uses them to learn strategies that generalize across large planning instances.
FIXME: Towards End-to-End Benchmarking of LLM-Aided Design Verification
Gwok-Waa Wan (Southeast University), Jun Yang (Texas Tech University)
AI Code AssistantLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Propose the FIXME benchmark, constructing 747 end-to-end functional verification tasks derived from real hardware designs, covering five subsets: specification understanding, reference models, testbenches, assertions, and RTL debugging.
FLAG-4D: Flow-Guided Local-Global Dual-Deformation Model for 4D Reconstruction
Guan Yuan Tan (Monash University), Chee-Ming Ting (National Institute of Technology Warangal)
GenerationData SynthesisRecurrent Neural NetworkTransformerGaussian SplattingOptical FlowVideo
🎯 What it does: FLAG-4D proposes a dual-network framework that utilizes an Instantaneous Deformation Network (IDN) and a Global Motion Network (GMN) to perform fine-grained and globally consistent deformation of 3D Gaussian primitives over time, achieving high-quality 4D reconstruction.
FlashKAT: Understanding and Addressing Performance Bottlenecks in the Kolmogorov-Arnold Transformer
Matthew Raffel (Oregon State University), Lizhong Chen (Oregon State University)
Computational EfficiencyTransformerImage
🎯 What it does: Research and optimize the training process of Kolmogorov‑Arnold Transformer (KAT), proposing FlashKAT for accelerated implementation.
FlashSVD: Memory-Efficient Inference with Streaming for Low-Rank Models
Zishan Shao (Duke University), Hai ¨Helen¨ Li
Computational EfficiencyTransformerText
🎯 What it does: This paper proposes FlashSVD, an end-to-end, low-rank aware streaming inference framework for language models that have already been compressed using SVD, which can significantly reduce activation memory requirements without increasing computational costs.
FlashVideo: Flowing Fidelity to Detail for Efficient High-Resolution Video Generation
Shilong Zhang (University of Hong Kong), Ping Luo (ByteDance)
GenerationTransformerSupervised Fine-TuningFlow-based ModelAuto EncoderVideoText
🎯 What it does: Propose a two-stage text-to-video generation framework called FlashVideo, which first generates content and motion highly aligned with the text at low resolution using a large model, then refines details at high resolution using a lightweight model.
Flexible Concept Bottleneck Model
Xingbo Du (Mohamed bin Zayed University of Artificial Intelligence), Rui Zhang (Renmin University of China)
ClassificationExplainability and InterpretabilityRepresentation LearningLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a Flexible Concept Bottleneck Model (FCBM) that allows dynamic modification of the concept set during training and inference without retraining the entire network; it generates concept-to-label weights using a hypernetwork and implements sparse concept selection via a sparsemax with a learnable temperature; further demonstrating the model's zero-shot generalization on unseen concepts and rapid adaptation to new concepts with only one round of fine-tuning.
FloorPlanFormer: Multi-Task Transformer Network for Floor Plan Recognition with Outer-to-Inner Feature Refinement
Yun Liang (South China Agricultural University), Yishen Lin (South China Agricultural University)
SegmentationTransformerImage
🎯 What it does: Propose a three-stage Transformer architecture called FloorPlanFormer, which utilizes multi-task learning to simultaneously identify outer contours, inner contours, and entrance doors;
Flora: Effortless Context Construction to Arbitrary Length and Scale
Tianxiang Chen (University of Science and Technology of China), Jieping Ye
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose the Flora strategy, which constructs arbitrary-length long contexts by concatenating short instructions without using LLMs or human intervention, and enhances the LLM's ability to process long texts through instruction fine-tuning.
FlorE: Integrating Full Lorentz Group and Directional Offsets for Effective Knowledge Graph Embedding
Zehua Duo (Inner Mongolia University), Guanglai Gao (Inner Mongolia University)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: The paper proposes a novel knowledge graph embedding model called FlorE, which combines the full Lorentz group with directional offset to address the Z-Paradox relation pattern.
Flow-Based Knowledge Transfer for Efficient Large Model Distillation
Xinye Yang (Newcastle University), Yiwei Chen (Jilin University)
ClassificationObject DetectionSegmentationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkFlow-based ModelRectified FlowImageBenchmarkOrdinary Differential Equation
🎯 What it does: Proposes a knowledge distillation framework called FlowDistill based on reversible normalized flows (CNF), which can achieve information-lossless mapping between teacher and student models, thereby more accurately transferring high-dimensional distribution information.
Flow-Induced Diagonal Gaussian Processes
Moule Lin (Trinity College Dublin, University of Dublin), Goetz Botterweck (Trinity College Dublin, University of Dublin)
SegmentationAnomaly DetectionComputational EfficiencyFlow-based ModelImage
🎯 What it does: This paper proposes the Flow-Induced Diagonal Gaussian Processes (FiDGP) framework, which projects neural network weight uncertainty into a low-dimensional induced subspace and combines regularized flow variational posterior to achieve model compression and high-quality uncertainty estimation, while supporting out-of-distribution (OoD) detection in a single forward pass.
FlowAnyTime: Efficient Fine-tuning with Intra-Inter Frame Distillation for All-Weather Optical Flow Estimation
Zixu Wang, Xinbo Zhao (Northwestern Polytechnical University)
Knowledge DistillationTransformerOptical FlowImageVideo
🎯 What it does: Proposes the FlowAnyTime framework, which utilizes the pre-trained CroCo v2 model for optical flow estimation, and performs fine-tuning and cross-frame distillation on key layers to achieve unified optical flow inference under degraded scenarios.
Flowing Backwards: Improving Normalizing Flows via Reverse Representation Alignment
Yang Chen (Nanjing University), Limin Wang (Alibaba Group)
GenerationTransformerFlow-based ModelImage
🎯 What it does: Proposes a Reverse Representation Alignment (R-REPA) strategy, aligning intermediate features with pre-trained visual foundation models on the forward encoding path and performing gradient updates on the reverse generation path to enhance the semantic expression and generation quality of Normalizing Flows (NF); simultaneously introduces a training-free test-time optimization classification method that directly utilizes NF's likelihood gradients for inference.
FlowPath: Learning Data-Driven Manifolds with Invertible Flows for Robust Irregularly-sampled Time Series Classification
YongKyung Oh (University of California Los Angeles), Sungil Kim (Ulsan National Institute of Science and Technology)
ClassificationRecurrent Neural NetworkFlow-based ModelTime SeriesBiomedical DataStochastic Differential Equation
🎯 What it does: Propose the FlowPath framework, which utilizes reversible neural flows to learn continuous, data-driven control paths to improve classification of irregularly sampled time series.
FLRQ: Faster LLM Quantization with Flexible Low-Rank Matrix Sketching
Hongyaoxing Gu (Institute of Software Chinese Academy of Sciences), Fangfang Liu (Institute of Software Chinese Academy of Sciences)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes the FLRQ (Flexible Low-Rank Quantization) algorithm, which uses R1-Sketch for variable-rank low-rank quantization and minimizes quantization error through BLC iteration, significantly reducing storage and inference costs while maintaining high accuracy.
FoAM: Foresight-Augmented Multi-Task Imitation Policy for Robotic Manipulation
Litao Liu (Corenetic AI), Wenzhao Lian (Chinese Academy of Sciences)
Robotic IntelligenceTransformerVision Language ModelAuto EncoderImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes the FoAM multi-task imitation learning strategy, leveraging multi-modal goal conditions (image + text) and foresight augmentation to enhance robotic manipulation performance.
FocusDPO: Dynamic Preference Optimization for Multi-Subject Personalized Image Generation via Adaptive Focus
Qiaoqiao Jin (ByteDance), Jidong Jiang (ByteDance)
GenerationSupervised Fine-TuningVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes FocusDPO, a framework for multi-agent personalized image generation achieved through dynamic focus adjustment and preference optimization.
Focusing on Language: Revealing and Exploiting Language Attention Heads in Multilingual Large Language Models
Xin Liu (Institute of Information Engineering Chinese Academy of Sciences), Xiaoqi Jia (Institute of Information Engineering Chinese Academy of Sciences)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an efficient method called LAHIS to evaluate the importance of each head in multi-head self-attention during multilingual processing, and identifies language-specific and general-purpose attention heads.
Force-Aware 3D Contact Modeling for Stable Grasp Generation
Zhuo Chen (University of Birmingham), Hyung Jin Chang (University of Birmingham)
GenerationOptimizationRobotic IntelligenceAuto EncoderMesh
🎯 What it does: This paper proposes a force-aware 3D contact modeling method for generating stable human grasping postures.
Forecast Then Calibrate: Feature Caching as ODE for Efficient Diffusion Transformers
Shikang Zheng (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)
GenerationComputational EfficiencyTransformerDiffusion modelImageVideoTextBenchmarkOrdinary Differential Equation
🎯 What it does: Proposed a training-agnostic acceleration framework called FoCa, treating feature caching as solving a feature ODE and using BDF2 predictor with Heun corrector to achieve multi-step, stable feature prediction, significantly improving the inference speed of Diffusion Transformer.
ForeDiffusion: Foresight-Conditioned Diffusion Policy via Future View Construction for Robot Manipulation
Weize Xie (Shenzhen University), F. Richard Yu (Shenzhen University)
Robotic IntelligenceConvolutional Neural NetworkDiffusion modelMultimodality
🎯 What it does: Designed and implemented Foresight-Conditioned Diffusion (ForeDiffusion), a robotic manipulation strategy that leverages predicted future views combined with a dual loss, enabling the generation of more stable and accurate action sequences in long-horizon complex tasks.
Forest vs Tree: The (N, K) Trade-off in Reproducible ML Evaluation
Deepak Pandita (Rochester Institute of Technology), Christopher M Homan
OptimizationData-Centric LearningText
🎯 What it does: The study evaluates the trade-off between the number of samples N and the number of annotations per sample K when assessing machine learning models under a fixed total annotation budget, and proposes how to allocate human annotation resources to achieve reliable evaluation.
Forget Less by Learning from Parents Through Hierarchical Relationships
Arjun Ramesh Kaushik (University at Buffalo-SUNY), Venu Govindaraju (University at Buffalo-SUNY)
GenerationRepresentation LearningDiffusion modelImage
🎯 What it does: Propose a framework named FLLP that leverages hierarchical relationships in hyperbolic space for concept transfer, mitigating catastrophic forgetting in custom diffusion models during continual learning.
Forget What Has Seen: Selective Concept Unlearning in Segmentation Foundation Models
Miaozeng Du (Southeast University), Qianshan Wei (Southeast University)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes Selective Concept Unlearning (SCU), a method for achieving selective forgetting of target concepts in segmentation foundation models;
Forgetting by Pruning: Data Deletion in Join Cardinality Estimation
Chaowei He (Soochow University), An Liu (Beijing Jiaotong University)
Computational EfficiencyTabularBenchmark
🎯 What it does: This paper proposes CEP, a chi-square estimation forgetting framework for multi-table machine learning based on distribution-sensitive pruning and domain pruning, designed to quickly delete data under incomplete training scenarios.
Forgetting Knowledge Localization and Isolation for Continual Forgetting of Pre-trained Vision Models
Zhiwen Yang (Hangzhou Dianzi University), Liang Li (Institute Of Computing Technology Chinese Academy Of Sciences)
Representation LearningSupervised Fine-TuningImage
🎯 What it does: The paper addresses the continuous forgetting task by proposing precise forgetting of target knowledge in pre-trained visual models through knowledge layer localization and parameter isolation, while minimizing damage to remaining knowledge.
Formal Abductive Latent Explanations for Prototype-Based Networks
Jules Soria (Université Paris-Saclay), Daniela Cancila (Université Paris-Saclay)
ClassificationExplainability and InterpretabilityImage
🎯 What it does: This paper proposes Abductive Latent Explanations (ALEs), a formal explainable method for constructing explanations in the latent space of case-based reasoning networks;
Formal Verification of Diffusion Auctions
Rustam Galimullin (University of Bergen), Laurent Perrussel (University of Bergen)
Finance Related
🎯 What it does: Proposes two logics (L_n and SL_n) for describing and verifying seller strategies in diffusion auctions, along with corresponding model checking and strategy existence algorithms.
Format as a Prior: Quantifying and Analyzing Bias in LLMs for Heterogeneous Data
Jiacheng Liu (Wuhan University), Tieyun Qian (Wuhan University)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelTextGraphTabularRetrieval-Augmented Generation
🎯 What it does: This paper systematically investigates the 'format bias' present in large language models when processing multi-format information (text, tables, infoboxes, knowledge graphs), and reveals its existence, driving factors, and internal mechanisms through a three-phase empirical analysis.
Format Matters: The Robustness of Multimodal LLMs in Reviewing Evidence from Tables and Charts
Xanh Ho (National Institute of Informatics), Akiko Aizawa (University of Tokyo)
TransformerLarge Language ModelVision Language ModelImageMultimodalityTabularChain-of-Thought
🎯 What it does: Design and evaluate the robustness of multi-modal LLMs in verifying scientific claims using tables and charts as evidence;
Foundation-Adaptive Integrated Refinement for Generalized Category Discovery
Yuwei Bian (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)
ClassificationRepresentation LearningTransformerContrastive LearningImageBenchmark
🎯 What it does: Proposes the Foundation-Adaptive Integrated Refinement (FAIR) framework to address the conflict between known and unknown class discovery in GCD.
Foundations of Formal Reasoning over Knowledge Bases Combining Symbolic and Sub-Symbolic Knowledge
Gianluca Cima (Sapienza University of Rome), Laura Papi (Sapienza University of Rome)
ClassificationComputational EfficiencyGraph
🎯 What it does: Proposes a Hybrid Knowledge Base (HKB) framework that integrates a machine learning binary classifier with ontology (DL-Lite RDFS) logic knowledge, providing model semantics and reasoning tasks.
FoundationSLAM: Unleashing the Power of Depth Foundation Models for End-to-End Dense Visual SLAM
Yuchen Wu (Beihang University), Xiao Bai (Beihang University)
Pose EstimationDepth EstimationOptimizationSimultaneous Localization and MappingOptical FlowVideoBenchmark
🎯 What it does: Proposed a monocular dense SLAM system called FoundationSLAM, which uses geometric priors from large-scale foundation depth models to guide optical flow estimation and achieves multi-view geometric consistency through bidirectional bundle adjustment;
FourierPET: Deep Fourier-based Unrolled Network for Low-count PET Reconstruction
Zheng Zhang (Hong Kong Polytechnic University), Jing Qin (Sun Yat-sen University Cancer Center)
RestorationConvolutional Neural NetworkBiomedical DataPositron Emission Tomography
🎯 What it does: To address the low-count positron emission tomography (PET) reconstruction problem, we propose FourierPET, an ADMM deconvolution network based on Fourier domain analysis, which can separately correct high-frequency phase noise and low-frequency amplitude attenuation in the frequency domain;
FP=XINT: Representing Neural Networks via Low-Bit Series Basis Functions
Boyang Zhang (Institute of Computing Technology, Chinese Academy of Sciences), Fangming Liu (Harbin Institute of Technology)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerImageTextBenchmark
🎯 What it does: This paper proposes a series expansion framework for deep networks, rewriting full-precision models as a linear combination of low-bit weight basis functions, achieving post-training quantization without calibration or fine-tuning.
FPT Approximation Algorithms for TSP on Non-Metric Graphs
Jingyang Zhao (University of Electronic Science and Technology of China), Mingyu Xiao (Anhui University)
Optimization
🎯 What it does: This paper improves two types of parameterized approximation algorithms for the Traveling Salesman Problem (TSP) on non-metric graphs, proposing new fixed-parameter tractable (FPT) approximation algorithms with parameters p (the number of vertices violating the triangle inequality) and q (the size of the minimum destroying set).
FQ-PETR: Fully Quantized Position Embedding Transformation for Multi-View 3D Object Detection
Jiangyong Yu (Houmo AI), Dawei Yang (Southeast University Dalian University Of Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A full integer quantization framework is implemented for the PETR series of multi-view 3D detection models, significantly reducing inference latency and memory usage.
FracSegmentator: Fracture Instance Segmentation with Trauma-Prior-Guided Contrastive Learning
Yanzhen Liu (Beihang University), Yu Wang (Beihang University)
SegmentationConvolutional Neural NetworkTransformerContrastive LearningImageComputed Tomography
🎯 What it does: Developed a two-stage deep learning framework named FracSegmentator for precise segmentation of fracture fragments from CT images.
Fractured Glass, Failing Cameras: Simulating Physics-Based Adversarial Samples for Autonomous Driving Systems
Manav Prabhakar (University of Michigan), Arpan Kusari (University of Michigan)
Object DetectionAutonomous DrivingAdversarial AttackImagePhysics Related
🎯 What it does: This study develops a physics-based adversarial sample generation method to simulate sensor failure caused by camera glass breakage and evaluate its impact on autonomous driving systems.
FRBAT: Conditionally-Visible Physical Backdoor Attack via Fluorescence
Yalun Wu (Beijing Jiaotong University), Jiqiang Liu (Beijing Jiaotong University)
ClassificationRecognitionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Designed a fluorescent physical backdoor attack (FRBAT) visible only under specific lighting conditions, which can implant hidden triggers into traffic sign recognition models and be activated by a UV light when needed.
FreDN: Spectral Disentanglement for Time Series Forecasting via Learnable Frequency Decomposition
Zhongde An (Shanghai University of Finance and Economics), Shouguo Du (Shanghai Municipal Big Data Center)
Time SeriesBenchmark
🎯 What it does: This paper proposes a new frequency domain time series forecasting framework, FreDN, which can learn and separate trend and periodic components in the frequency domain, model the trend in the time domain, and ultimately achieve long-term forecasting.
Free-Form Scene Editor: Enabling Multi-Round Object Manipulation Like in a 3D Engine
Xincheng Shuai (Fudan University), Dacheng Tao (Nanyang Technological University)
Image TranslationGenerationTransformerDiffusion modelImageVideo
🎯 What it does: Proposed Free-Form Scene Editor (FFSE), an autoregressive framework enabling multi-round 3D-aware object editing on real images.
FreeAskWorld: An Interactive and Closed-Loop Simulator for Human-Centric Embodied AI
Yuhang Peng (Tsinghua University), Jiangtao Gong (Tsinghua University)
Autonomous DrivingData-Centric LearningRobotic IntelligenceLarge Language ModelWorld ModelVideoTextMultimodalityBenchmark
🎯 What it does: Introduces the FreeAskWorld interactive closed-loop simulator and the Direction Inquiry Task, supporting high-level human-machine interaction and navigation;
FreeGaussian: Annotation-free Control of Articulated Objects via 3D Gaussian Splats with Flow Derivatives
Qizhi Chen (Zhejiang University), Bin Zhao (Shanghai AI Laboratory)
GenerationPose EstimationGaussian SplattingOptical FlowVideo
🎯 What it does: Proposes an annotation-free Gaussian splatting method called FreeGaussian, which automatically locates interactive objects and achieves controllable view synthesis through differential analysis of optical flow and camera motion.
FreeInpaint: Tuning-free Prompt Alignment and Visual Rationality Enhancement in Image Inpainting
Chao Gong (Fudan University), Tao Mei (HiDream.ai Inc.)
RestorationDiffusion modelImage
🎯 What it does: Propose FreeInpaint, a text-guided image inpainting method that requires no fine-tuning, directly enhancing restoration results during the inference stage by optimizing initial noise and intermediate latent variables.
FreeMem: Enhancing Consistency in Long Video Generation via Tuning-Free Memory
Jibin Peng (Tianjin University), Qing Guo (Shenzhen University)
GenerationTransformerDiffusion modelVideoBenchmark
🎯 What it does: Designed a parameter-free three-layer memory mechanism (noise, token, attention) to enhance consistency in long video generation.
FreLay: Frequency-aware Energy Function for Training-free Layout-to-Image Generation
Bonan Li (University of Chinese Academy of Sciences), Xinchao Wang (National University of Singapore)
GenerationDiffusion modelScore-based ModelImage
🎯 What it does: Proposed a training-free layout-to-image generation method called FreLay, achieving more precise spatial control and enhanced visual quality through a frequency-aware energy function
FreqCycle: A Multi-Scale Time-Frequency Analysis Method for Time Series Forecasting
Boya Zhang (Shanghai Jiao Tong University), Xing He (Shanghai Jiao Tong University)
Time Series
🎯 What it does: Propose the FreqCycle framework, combining low-frequency cycle extraction with mid-to-high frequency enhancement to achieve efficient time series forecasting.
FreqTAD: Multi-scale Frequency Encoding and Time-Frequency Attention for Anomaly Detection in Dynamic Graphs
Chao Li (Shandong University of Science and Technology), Qingtian Zeng (Shandong University of Science and Technology)
Anomaly DetectionRecurrent Neural NetworkGraph Neural NetworkTransformerContrastive LearningGraph
🎯 What it does: Proposed the FreqTAD model, which achieves dynamic graph anomaly detection by leveraging multi-scale frequency encoding and time-frequency attention.
Frequency Mining Empowered by Text Aggregation: A New Perspective on Document Image Tampering Detection
Ziqi Yi (South China University of Technology), Lianwen Jin (South China University of Technology)
SegmentationAnomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed TAFE-Net for document image tampering detection, aggregating multi-frequency features and text.
Frequency-Aligned Cross-Modal Learning with Top-K Wavelet Fusion and Dynamic Expert Routing for Enhanced Retinal Disease Diagnosis
Yuxin Lin (Harbin Institute of Technology), Wei Wang (Harbin Institute of Technology)
ClassificationConvolutional Neural NetworkMixture of ExpertsImageMultimodalityBiomedical Data
🎯 What it does: This paper proposes a multi-modal fusion framework for retinal disease diagnosis, dynamically routing and fusing features from color fundus photographs (CFP) and optical coherence tomography (OCT) images.
Frequency-Aware Vision-Language Multimodality Generalization Network for Remote Sensing Image Classification
Junjie Zhang (Xi'an University of Posts and Telecommunications), Jun Yu (Shaanxi Normal University)
ClassificationConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningMultimodality
🎯 What it does: Designed and implemented a frequency-domain aware vision-language multimodal generalization network (FVMGN) for cross-scenario, multimodal transfer in remote sensing image classification.
Frequency-Dependent Scheduled Schrödinger Bridge for Underwater Acoustic Signal Denoising
Pengsen Zhu, Yonggang Zhang (Harbin Engineering University)
RestorationDiffusion modelScore-based ModelPhysics RelatedOrdinary Differential EquationAudio
🎯 What it does: This paper proposes a frequency-dependent Schrodinger Bridge (FDSSB) framework for underwater acoustic signal denoising.
FRoD: Full-Rank Efficient Fine-Tuning with Rotational Degrees for Fast Convergence
Guoan Wan (Beihang University), Runhua Xu (Beihang University)
ClassificationComputational EfficiencyTransformerSupervised Fine-TuningImageText
🎯 What it does: Proposes a full-rank efficient fine-tuning method named FRoD, which leverages hierarchical joint decomposition to extract a shared base and introduces sparse learnable rotation perturbations on this base, achieving efficient adaptation for large pre-trained models.
From Attribution to Action: Jointly ALIGNing Predictions and Explanations
Dongsheng Hong (Fuzhou University), Xiangwen Liao (Fuzhou University)
Domain AdaptationExplainability and InterpretabilityImageBenchmark
🎯 What it does: Propose the ALIGN framework, which jointly trains a learnable masker and classifier to align model predictions with explanations, thereby enhancing interpretability and generalization capabilities.
From Blind Transfer to Wise Selection: Prototype-Driven Neighbor-Domain Adaptation for Fake News Detection
Wayne Lu (Independent Researcher), Yiheng Li (University of International Business and Economics)
ClassificationDomain AdaptationTransformerMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: This paper proposes the PANDA framework, which dynamically selects and fuses the most beneficial domain knowledge to address the negative transfer problem in multi-domain, multi-modal fake news detection.
From Chaos to Clarity: A Knowledge Graph-Driven Audit Dataset Generation Framework for LLM Unlearning
Weipeng Jiang (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextGraphBenchmark
🎯 What it does: Designed and implemented the LUCID framework, which automatically generates audit datasets using knowledge graphs to evaluate the machine forgetting effect of LLMs.
From Chaos to Cure: A Prefix Heuristics Guided Model-Agnostic Adaptive Detoxification Framework
Yuhu Shang (Beijing University of Posts and Telecommunications), Zhaofeng He (Beijing University of Posts and Telecommunications)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes the MAAD (Model-Agnostic Adaptive Detoxification) framework, achieving detoxification of large language models while maintaining generation quality through prefix heuristics, antitoxicity datasets, lightweight Detoxifiers, and dynamic truncation sampling.
From Dataset to Real-world: General 3D Object Detection via Generalized Cross-domain Few-shot Learning
Shuangzhi Li (University of Alberta), Xingyu Li (University of Alberta)
Object DetectionDomain AdaptationAutonomous DrivingContrastive LearningImagePoint CloudBenchmark
🎯 What it does: Proposed a cross-domain few-shot learning task and designed the GCFS framework for LiDAR 3D object detection.
From Decision Trees to Boolean Logic: A Fast and Unified SHAP Algorithm
Alexander Nadel (Technion), Ron Wettenstein (Technion)
Explainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: Proposed the WOODELF algorithm, which unifies the implementation of various feature importance metrics for decision tree ensemble models, including SHAP, Shapley interaction values, and Banzhaf values. It provides a pure Python implementation (using NumPy, SciPy, CuPy) and supports parallel computing on both CPU and GPU.
From Detection to Diagnosis: Advancing Hallucination Analysis with Automated Data Synthesis
Yanyi Liu (Northeastern University), Yingyou Wen (Northeastern University)
Data SynthesisAnomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: Propose the 'Hallucination Diagnosis' paradigm, construct an automated data synthesis pipeline HDG, and train a 4B-parameter HDM-4B-RL model for detection, localization, explanation, and correction;
From Diagnosis to Generalization: A Cognitive Approach to Data Selection for Educational LLMs
Yuxiang Guo (University of Science and Technology of China), Shijin Wang (IFLYTEK Research)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed the CASS framework, combining cognitive diagnosis, information selection, and hierarchical curriculum to efficiently select and fine-tune data subsets for educational LLMs.
From Dialogue to Destination: Geography-Aware Large Language Models with Multimodal Fusion for Conversational Recommendation
Yeming Li (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
Recommendation SystemTransformerLarge Language ModelMultimodality
🎯 What it does: Propose a geo-aware dialogue recommendation framework called GeoCRS, which collaborates with a frozen LLM and external trainable modules to jointly generate multimodal geo-guided signals.
From Discriminative to Generative: A Diffusion-Based Paradigm for Multi-Agent Collaborative Perception
Kexin Gong (Beijing University of Posts and Telecommunications), Jinglin Li (Beijing University of Posts and Telecommunications)
Autonomous DrivingDiffusion modelPoint Cloud
🎯 What it does: Proposes a two-stage generative supervised collaborative perception framework called DiGS-CP, which utilizes conditional diffusion models during the training phase to guide feature fusion, significantly enhancing perception performance while reducing communication overhead.
From Hypothesis to Premises: LLM-based Backward Logical Reasoning with Selective Symbolic Translation
Qingchuan Li, Tongxuan Liu (University Of Science And Technology Of China)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposes the Hypothesis-driven Backward Logical Reasoning (HBLR) framework, combining confidence-aware selective symbolic translation with hypothesis-driven backward reasoning to address redundancy and translation errors in LLM forward reasoning.
From IDs to Semantics: A Generative Framework for Cross-Domain Recommendation with Adaptive Semantic Tokenization
Peiyu Hu (Xi'an Jiaotong-Liverpool University), Jia Wang (Xi'an Jiaotong-Liverpool University)
Domain AdaptationRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed a cross-domain generative recommendation framework called GenCDR based on large language models, addressing the issues caused by traditional ID dependencies, such as item ID explosion and insufficient domain personalization.
From Imitation to Discrimination: Toward a Generalized Curriculum Advantage Mechanism Enhancing Cross-Domain Reasoning Tasks
Changpeng Yang (Xiaomi Corporation), Guoquan Zhang (Xiaomi Corporation)
Reinforcement LearningTextMultimodalityBenchmark
🎯 What it does: Proposed the CAPO (Curriculum Advantage Policy Optimization) framework, which constructs an adaptive two-phase reinforcement learning training process by first using positive advantage samples for imitation learning and then introducing negative advantage samples for discriminative learning.
From Intent to Execution: Multimodal Chain-of-Thought Reinforcement Learning for Precise CAD Code Generation
Ke Niu (Fudan University), Xiangyang Xue (Fudan University)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose CAD-RL, a multi-modal chain-of-thought reinforcement learning framework that automatically generates executable CADQuery code and outputs precise 3D models using natural language or structured inputs.
From Macro to Micro: Probing Dataset Diversity in Language Model Fine-Tuning
Haoyu Li (Beijing Institute of Technology), Kun Liu (Beijing Institute of Technology)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper systematically studies the impact of dataset diversity on model performance during the supervised fine-tuning phase of large language models, and classifies and experimentally verifies macro, meso, and micro diversity control strategies.
From Mathematical Reasoning to Code: Generalization of Process Reward Models in Test-Time Scaling
Zhengyu Chen (Meituan Inc), Jingang Wang (Meituan Inc)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Study the training, scaling, data diversity, and testing expansion of Process Reward Models (PRMs), evaluating their generalization in mathematical reasoning and code generation tasks.
From Parameter to Representation: A Closed-Form Approach for Controllable Model Merging
Jialin Wu (Rocket Force University of Engineering), Zhiyong Yu (Rocket Force University of Engineering)
ClassificationOptimizationComputational EfficiencyRepresentation LearningTransformerImage
🎯 What it does: This paper proposes the ReACT method, which achieves controllable model fusion by performing linear correction in the model's final representation space;
From Passive Perception to Active Memory: A Weakly Supervised Image Manipulation Localization Framework Driven by Coarse-Grained Annotations
Zhiqing Guo (Xinjiang University), Gaobo Yang
SegmentationAnomaly DetectionKnowledge DistillationTransformerPrompt EngineeringImage
🎯 What it does: Propose the BoxPromptIML framework, which utilizes coarse box prompts to generate pseudo masks via SAM and distills a lightweight student model, achieving weakly supervised image tampering localization.
From Pixels to Logic: A Perception-Reasoning Decomposition Framework for Open-World Referring Expression Comprehension
Lihong Huang (Shenzhen University), Yan Liu (Hong Kong Polytechnic University)
SegmentationDepth EstimationTransformerVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposes a training-free open-world gesture expression understanding framework (PRDF), which separates visual perception from language reasoning. It first generates rich textual scene descriptions using open-source foundation models, then employs a language model to perform logical reasoning for target localization.
From Points to Coalitions: Hierarchical Contrastive Shapley Values for Prioritizing Data Samples
Canran Xiao, Liwei Hou (National University Of Singapore)
Explainability and InterpretabilityComputational EfficiencyData-Centric LearningContrastive LearningImageTabular
🎯 What it does: Proposes Hierarchical Contrastive Data Valuation (HCDV), which geometrically preserves contrastive learning embeddings of data, constructs multi-level clustering, and assigns value to each sample using local Monte Carlo Shapley estimation.
From Pretrain to Pain: Adversarial Vulnerability of Video Foundation Models Without Task Knowledge
Hui Lu (Nanyang Technological University), Xudong Jiang (Nanyang Technological University)
Adversarial AttackTransformerContrastive LearningVideoMultimodality
🎯 What it does: Propose a transferable adversarial attack method (TVA) based on a video foundation model (VFM) that does not require downstream task knowledge, training data, model queries, or parameters. It directly generates adversarial perturbations on the temporal representations of VFM to attack various downstream video models and multimodal large language models.
From Sampling to Cognition: Modeling Internal Cognitive Confidence in Language Models for Robust Uncertainty Calibration
Hao Li (Harbin Institute of Technology), Ming Liu (Harbin Institute of Technology)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningText
🎯 What it does: Propose CogConf, an internal cognitive confidence metric based on semantic entropy and rejection rate, and build the COGALIGN framework, which uses reinforcement learning to align the linguistic confidence of LLMs with CogConf, thereby enhancing the model's self-assessment and reliability.
From Scene to Object: Enhancing Open-Vocabulary Object Detection via Foreground-Background Context Reasoning
Yanqi Li (Beihang University), Tao Ren (University of Chinese Academy of Sciences)
Object DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: Propose the BFDet framework, which leverages foreground-background context reasoning combined with LLM and VLM to generate high-quality pseudo labels, thereby enhancing open-vocabulary object detection.
From Semantics to Spectrum: A New Lens on Graph Augmentation Strategy
Xiangping Zheng (Harbin Engineering University), Zhiwen Yu (Beijing Jinghang Research Institute of Computing and Communication)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes a spectral-based graph contrastive learning framework named FA-GCL, which achieves graph augmentation by preserving low-frequency information and perturbing high-frequency information in the frequency domain.
From Sequential to Recursive: Enhancing Decision-Focused Learning with Bidirectional Feedback
Xinyu Wang (Hong Kong Polytechnic University), Wei Ma (Hong Kong Polytechnic University)
OptimizationTabularTime Series
🎯 What it does: Propose a recursive decision focusing learning framework (R-DFL), establishing a bidirectional feedback loop between prediction and optimization to achieve closed-loop decision-making.