NeurIPS 2025 Papers — Page 24
Conference on Neural Information Processing Systems · 5275 papers
Interactive Cross-modal Learning for Text-3D Scene Retrieval
Yanglin Feng (Sichuan University), Peng Hu (Sichuan University)
RetrievalDomain AdaptationTransformerLarge Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: An interactive text-3D scene retrieval method called IDeal is proposed, which enhances the alignment between text and 3D scenes through continuous question answering.
Intermediate Domain Alignment and Morphology Analogy for Patent-Product Image Retrieval
Haifan Gong (Chinese University of Hong Kong), Haofeng Li (Sun Yat-sen University)
RetrievalDomain AdaptationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A new task of cross-domain retrieval of patent images and product images is proposed—Patent Product Image Retrieval (PPIR), and a large-scale PPIRD dataset is constructed;
Interpretable and Parameter Efficient Graph Neural Additive Models with Random Fourier Features
Thummaluru Siddartha Reddy (Fujitsu Research of India), Mahesh Chandran (Fujitsu Research of India)
Explainability and InterpretabilityComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: A lightweight interpretable graph neural additive model G-NAMRFF is designed, which integrates graph structure and node features using random Fourier features and learnable FIR filters, achieving independent contributions of each feature and interpretability;
Interpretable Next-token Prediction via the Generalized Induction Head
Eunji Kim (Microsoft Research), Jianfeng Gao (Microsoft Research)
GenerationExplainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelTextBiomedical DataMagnetic Resonance ImagingRetrieval-Augmented Generation
🎯 What it does: This paper proposes an interpretable next-word prediction model GIM, which makes the induction head mechanism in large language models explicit as a retrieval-based method, and validates its effectiveness in text generation and fMRI prediction tasks.
Interpreting Arithmetic Reasoning in Large Language Models using Game-Theoretic Interactions
Leilei Wen (Tongji University), Wen Shen (Tongji University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Through interactive analysis using game theory, the internal mechanisms of large language models for arithmetic reasoning are decomposed into vocabulary-level interactions, quantifying different types and orders of interactions.
Interpreting Emergent Features in Deep Learning-based Side-channel Analysis
Sengim Karayalcin, Stjepan Picek (Radboud University)
Explainability and InterpretabilityConvolutional Neural NetworkTime SeriesBenchmark
🎯 What it does: The researchers conducted a post-hoc analysis of successful deep learning side-channel attack models using mechanistic interpretability methods, reverse-engineering and reconstructing the masked values in the cryptographic implementation, revealing which parts of the physical leakage the model utilized for predictions.
Interpreting vision transformers via residual replacement model
Jinyeong Kim (Yonsei University), Seong Jae Hwang (Yonsei University)
Explainability and InterpretabilityTransformerAuto EncoderImage
🎯 What it does: This paper studies the internal features and mechanisms of the Vision Transformer (ViT) and proposes a residual substitute model based on residual flow to explain the decision-making process of ViT and debug model biases.
Intervene-All-Paths: Unified Mitigation of LVLM Hallucinations across Alignment Formats
Jiaye Qian (Sun Yat-sen University), Sibei Yang (Sun Yat-sen University)
TransformerVision Language ModelImageText
🎯 What it does: A unified multi-path intervention framework called AllPath is proposed, which combines image-text and text-text attention paths to reduce hallucinations in large visual language models.
Intrinsic Benefits of Categorical Distributional Loss: Uncertainty-aware Regularized Exploration in Reinforcement Learning
Ke Sun (University of Alberta), Linglong Kong (University of Alberta)
Reinforcement LearningSequential
🎯 What it does: This paper reveals that the potential advantages of classification distributed loss come from an entropy regularization based on distribution matching through the decomposition of return density. It further explains the essential advantages of distributed reinforcement learning (especially classification distributed RL) compared to traditional RL.
Intrinsic Goals for Autonomous Agents: Model-Based Exploration in Virtual Zebrafish Predicts Ethological Behavior and Whole-Brain Dynamics
Reece Keller (Carnegie Mellon University), Aran Nayebi (Carnegie Mellon University)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningWorld ModelBiomedical Data
🎯 What it does: In an environment without external rewards, a self-driven exploration method based on model memory differences, called 3M-Progress, is proposed, allowing simulated zebrafish to exhibit active-passive behavior cycles similar to biological zebrafish and predict their whole-brain neural-astrocyte activity.
IntrinsiX: High-Quality PBR Generation using Image Priors
Peter Kocsis (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: This paper proposes a technique for directly generating physically based rendering (PBR) textures (albedo, roughness, metallic, normal) from text descriptions, utilizing diffusion models to generate high-quality, re-lightable, and editable raw images.
Inv-Entropy: A Fully Probabilistic Framework for Uncertainty Quantification in Language Models
Haoyi Song (University of Michigan), Raed Al Kontar (University of Michigan)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: A completely probability-based framework called Inv-Entropy is proposed to quantify the uncertainty of large language models, along with the GAAP perturbation algorithm and TSU evaluation metrics.
Inverse Methods for Missing Data Imputation
Hao Wang (Xiaohongshu Inc), Zhouchen Lin (Peking University)
OptimizationTabular
🎯 What it does: A dual-layer optimization-based missing value imputation method called Kernel Point Imputation (KPI) is proposed, which adaptively selects the model form for each feature in RKHS and uses complete features (oracle features) as supervisory signals for imputation.
Inverse Optimization Latent Variable Models for Learning Costs Applied to Route Problems
Alan Lahoud, Johannes A. Stork (Örebro University)
Anomaly DetectionOptimizationAuto EncoderGraphTabular
🎯 What it does: This paper proposes an Inverse Optimization Latent Variable Model (IO-LVM) that learns the cost latent variables of Constraint Optimization Problems (COP) through a variational autoencoder and generates constraint-satisfying paths or cycles during the decoding phase using a black-box solver.
Investigating and Mitigating Catastrophic Forgetting in Medical Knowledge Injection through Internal Knowledge Augmentation Learning
Yuxuan Zhou (Tsinghua University), Ji Wu (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: This paper studies the phenomenon of catastrophic forgetting in large language models during the process of medical knowledge injection and proposes an InternAL method that enhances injection using the model's internal knowledge.
Investigating Hallucinations of Time Series Foundation Models through Signal Subspace Analysis
Yufeng Zou (Northwestern University), Han Liu (Northwestern University)
Anomaly DetectionOptimizationRecurrent Neural NetworkTime SeriesSequential
🎯 What it does: This paper studies the hallucination problem in time series foundational models (TSFM) during zero-shot prediction. It first formalizes the definition of hallucination and proposes knowledge rules based on trends, frequencies, patterns, and ARMA for detection. Then, it identifies the signal subspace within the model through hidden state analysis and provides a signal strength measure called SSAS. Finally, it proposes an adaptive amplification intervention method, SSIM, which performs center-project-scale operations on hidden states during testing to improve prediction quality and reduce the hallucination rate.
InvFusion: Bridging Supervised and Zero-shot Diffusion for Inverse Problems
Noam Elata (Technion), Michael Elad
RestorationGenerationDiffusion modelImage
🎯 What it does: InvFusion is proposed, a diffusion model framework that simultaneously considers both training-based and zero-shot methods in inverse problems;
InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy
Vishnu Vinod (Cognitive and Robotics Artificial Intelligence Institute, Indian Institute of Technology Madras), Abhradeep Guha Thakurta
GenerationSafty and PrivacyLarge Language ModelText
🎯 What it does: A framework named InvisibleInk is proposed for long text generation of large models while maintaining strict differential privacy (DP) guarantees.
IOSTOM: Offline Imitation Learning from Observations via State Transition Occupancy Matching
Quang Anh Pham (Singapore Management University), Akshat Kumar (Singapore Management University)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningTabularBenchmark
🎯 What it does: In an offline environment, an implicit policy g(s'|s) is learned using only expert state trajectories and state-action data containing suboptimal behaviors, achieving observational learning by matching the joint distribution of state transitions.
IPAD: Inverse Prompt for AI Detection - A Robust and Interpretable LLM-Generated Text Detector
Zheng CHEN, Bo Li (Hong Kong University of Science and Technology)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposes the IPAD framework, which utilizes reverse prompting to reconstruct text generation prompts, and implements LLM text detection through a prompt-text consistency validator and a regeneration comparator.
IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals
Markus Gross (Technical University of Munich), Henri Meeß (Fraunhofer Institute IVI)
RestorationObject DetectionSegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: Proposes IPFormer, a visual 3D panoramic scene completion framework that utilizes context-adaptive instance proposals;
IPSI: Enhancing Structural Inference with Automatically Learned Structural Priors
Zhongben Gong (Shenzhen University), Mingyang Zhou (Shenzhen University)
Recurrent Neural NetworkGraph Neural NetworkAuto EncoderGraphTime SeriesPhysics Related
🎯 What it does: An iterative pre-training structure inference framework IPSI is proposed, which alternately updates a pre-trained structure estimator and a VAE-based joint inference module to improve the accuracy of structure inference in interactive dynamic systems.
Irrational Complex Rotations Empower Low-bit Optimizers
Zhen Tian (ByteDance), Ji-Rong Wen (Renmin University of China)
CompressionRecommendation SystemOptimizationText
🎯 What it does: Proposes π-Quant, a complex rotation algorithm that utilizes the properties of the irrational number π to compress the optimizer state to a low bit-width (as low as 3.32-bit) while maintaining full precision;
Is Grokking a Computational Glass Relaxation?
Xiaotian Zhang (City University of Hong Kong), Ge Zhang (City University of Hong Kong)
OptimizationTransformerSequentialPhysics Related
🎯 What it does: By viewing neural networks as physical systems, the Wang-Landau MD method is used to sample the Boltzmann entropy landscape of Transformers in modular arithmetic tasks, studying and explaining the grokking phenomenon.
Is Limited Participant Diversity Impeding EEG-based Machine Learning?
Philipp Bomatter (University of Edinburgh), Henry Gouk (University of Edinburgh)
Time SeriesBiomedical DataAlzheimer's Disease
🎯 What it does: A systematic quantitative study on the impact of sample partitioning and participant diversity in EEG machine learning, and an evaluation of the effectiveness of data augmentation and self-supervised pre-training under different data scales.
Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models
Jipeng Li (University of California), Yanning Shen (University of California)
Graph Neural NetworkDiffusion modelGraph
🎯 What it does: This paper explores whether explicit noise level conditioning is necessary in graph diffusion models and demonstrates that in large graphs, noise levels can be implicitly inferred through structural information, thus achieving unconditional models.
Is PRM Necessary? Problem-Solving RL Implicitly Induces PRM Capability in LLMs
Zhangyin Feng (Huawei Technologies), Zhirui Zhang (Huawei Technologies)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper systematically studies whether pure RL training is sufficient to enhance the capabilities of the Process Reward Model (PRM) in large language models, and experimentally verifies the co-evolution of RL and PRM functionalities.
Is the acquisition worth the cost? Surrogate losses for Consistent Two-stage Classifiers
Florence Regol (McGill University), Mark Coates (McGill University)
ClassificationOptimizationLarge Language ModelText
🎯 What it does: This study investigates a two-stage classification problem, jointly training a base classifier, an auxiliary classifier, and a decision module, and proposes a hinge-based surrogate loss to prove its consistency.
Is Your Diffusion Model Actually Denoising?
Daniel Pfrommer (Massachusetts Institute of Technology), Ali Jadbabaie (Massachusetts Institute of Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Investigate the deviation of conditional diffusion models and propose the Schedule Deviation metric to measure the difference between the generation process and the ideal denoising path;
Isotropic Noise in Stochastic and Quantum Convex Optimization
Annie Marsden (Google Deepmind), Chenyi Zhang (Stanford University)
Optimization
🎯 What it does: This paper studies the problem of minimizing d-dimensional Lipschitz convex functions using a stochastic gradient oracle, proposing a new algorithm that can improve the results by a factor of d in certain cases and provides new optimal complexity for sub-exponential noise.
It’s Hard to Be Normal: The Impact of Noise on Structure-agnostic Estimation
Jikai Jin (Stanford University), Vasilis Syrgkanis (Stanford University)
🎯 What it does: This paper studies the impact of treatment noise distribution on the performance of Structural Agnostic Estimation (SAE) in certain linear models. It proves that under Gaussian noise, Double Machine Learning (DML) has reached the lower bound, while under non-Gaussian noise, higher-order orthogonal moment functions can be constructed to achieve higher-order robustness, and proposes the ACE (Agnostic Cumulant-based Estimation) estimator.
ItDPDM: Information-Theoretic Discrete Poisson Diffusion Model
Sagnik Bhattacharya (Stanford University), Tsachy Weissman (Stanford University)
GenerationData SynthesisDiffusion modelImageTabular
🎯 What it does: A discrete diffusion model based on Poisson noise, ItDPDM, is proposed for generating non-negative discrete data.
Iterative Foundation Model Fine-Tuning on Multiple Rewards
Pouya M. Ghari (Biogen), Ye Wang (Biogen)
Reinforcement LearningText
🎯 What it does: An iterative multi-objective reinforcement learning algorithm, IterativeRS, is proposed for multi-objective fine-tuning of the base model, balancing various objectives through iterative merging of expert policies.
Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers
Zhengliang Shi (Shandong University), Zhaochun Ren (Leiden University)
GenerationRetrievalTransformerLarge Language ModelReinforcement LearningAgentic AITextRetrieval-Augmented Generation
🎯 What it does: This paper proposes the EXSEARCH framework, which enables large language models to become proactive search agents through a self-motivated iterative process of thinking-searching-recording.
Iterative Tool Usage Exploration for Multimodal Agents via Step-wise Preference Tuning
Pengxiang Li (Beijing Institute of Technology), Qing Li (Beijing Institute of Technology)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodality
🎯 What it does: The SPORT method is proposed, allowing multimodal agents to learn tool usage through self-exploration without the need for pre-collected data.
Jacobian-Based Interpretation of Nonlinear Neural Encoding Model
Xiaohui Gao (Northwestern Polytechnical University), Xintao Hu (Northwestern Polytechnical University)
Explainability and InterpretabilityImageMagnetic Resonance Imaging
🎯 What it does: This paper proposes and validates a nonlinear explanation metric JNE based on Jacobian dispersion to quantify the degree of nonlinearity in neural encoding models predicting BOLD signals, revealing hierarchical and stimulus-specific nonlinear distributions in the visual cortex.
JADE: Joint Alignment and Deep Embedding for Multi-Slice Spatial Transcriptomics
Yuanchuan Guo (Harvard University), Ying Ma (Brown University)
Representation LearningGraph Neural NetworkAuto EncoderContrastive LearningBiomedical Data
🎯 What it does: This paper proposes the JADE model, which can simultaneously perform spatial alignment and low-dimensional embedding learning for multi-slice spatial transcriptomic data.
JAFAR: Jack up Any Feature at Any Resolution
Paul Couairon (Sorbonne Université), Nicolas THOME
SegmentationDepth EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A lightweight, attention-based feature upsampling module called JAFAR has been developed, which can upsample low-resolution features from any base visual encoder to any high resolution, significantly improving downstream task performance.
JailBound: Jailbreaking Internal Safety Boundaries of Vision-Language Models
Jiaxin Song (Shanghai Jiao Tong University), Yingchun Wang (Shanghai Artificial Intelligence Laboratory)
Safty and PrivacyAdversarial AttackTransformerVision Language ModelMultimodality
🎯 What it does: This study proposes a framework for 'jailbreaking' attacks on the intrinsic safety decision boundaries of visual-language models—JailBound, which can simultaneously generate adversarial perturbations on visual and textual inputs, inducing the model to produce outputs that violate safety policies.
Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence
Octave Mariotti (University of Edinburgh), Hakan Bilen (University of Edinburgh)
SegmentationPose EstimationSupervised Fine-TuningContrastive LearningImageBenchmark
🎯 What it does: This paper studies the semantic correspondence problem and finds that supervised methods have poor generalization ability on unseen keypoints. To address this issue, the authors utilize monocular depth estimation to elevate 2D keypoints to 3D space, learning a category-level continuous 3D canonical manifold. They achieve an extension of sparse supervision through geometric consistency and, based on this, construct a new SPair-U evaluation benchmark to test the model's performance on unseen keypoints.
JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensemble Generation
Ameya Daigavane (Massachusetts Institute of Technology), Joseph Kleinhenz (Prescient Design)
GenerationProtein Structure PredictionGraph Neural NetworkScore-based ModelPoint CloudSequentialStochastic Differential Equation
🎯 What it does: The JAMUN model is constructed to generate conformational ensembles of peptide fragments by performing Langevin dynamics (walk) and denoising (jump) in a noise-added space.
Janus-Pro-R1: Advancing Collaborative Visual Comprehension and Generation via Reinforcement Learning
Kaihang Pan (Zhejiang University), Yueting Zhuang (Zhejiang University)
GenerationLarge Language ModelSupervised Fine-TuningReinforcement LearningImageTextMultimodalityChain-of-Thought
🎯 What it does: A two-stage training framework is proposed, enabling multimodal large language models (MLLM) to form a genuine chain of thought (CoT) between visual understanding and generation, thereby achieving iterative self-reflective image generation and unified image editing.
JanusDNA: A Powerful Bi-directional Hybrid DNA Foundation Model
Qihao Duan (Berlin Institute of Health), Benjamin Wild (Berlin Institute of Health)
Mixture of ExpertsBiomedical DataBenchmark
🎯 What it does: A bidirectional foundational model for DNA, JanusDNA, has been developed, capable of processing sequences of up to 1 Mbp at single nucleotide resolution and achieving efficient global modeling.
JarvisArt: Liberating Human Artistic Creativity via an Intelligent Photo Retouching Agent
Yunlong Lin (Xiamen University), Shuicheng YAN
Image TranslationRestorationGenerationOptimizationTransformerLarge Language ModelReinforcement LearningVision Language ModelImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Developed JarvisArt, a multimodal large language model-driven photo editing agent that can understand user intentions and automatically invoke over 200 Lightroom tools to complete both global and local edits.
Jasmine: Harnessing Diffusion Prior for Self-supervised Depth Estimation
JiYuan Wang, Yao Zhao (Beijing Jiaotong University)
Depth EstimationAutonomous DrivingRecurrent Neural NetworkDiffusion modelImage
🎯 What it does: A self-supervised monocular depth estimation framework called Jasmine based on Stable Diffusion is proposed, which enhances detail and cross-domain generalization by introducing Mixed Batch Image Reconstruction (MIR) and Scale-Shift GRU (SSG).
JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation
Kai Liu (Zhejiang University), Tat-Seng Chua (National University of Singapore)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelVideoTextMultimodalityAudio
🎯 What it does: Developed a unified multimodal large language model JavisGPT, capable of simultaneously understanding and generating synchronized audio-visual content (sounding video).
Jet-Nemotron: Efficient Language Model with Post Neural Architecture Search
Yuxian Gu (NVIDIA), Han Cai (NVIDIA)
OptimizationComputational EfficiencyKnowledge DistillationNeural Architecture SearchTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes Jet-Nemotron, a hybrid architecture language model that combines full attention and linear attention, and quickly transforms the pre-trained full attention model through PostNAS (Post Neural Architecture Search);
Johnson-Lindenstrauss Lemma Beyond Euclidean Geometry
Chengyuan Deng (Rutgers University), Cheng Xin (Rutgers University)
ImageGraph
🎯 What it does: This paper proposes two methods to extend Johnson-Lindenstrauss (JL) projection to non-Euclidean spaces, allowing for low-dimensional embedding solely from the similarity matrix of symmetric cavities;
Joint Design of Protein Surface and Backbone Using a Diffusion Bridge Model
Guanlue Li (University of Hamburg), Sören Laue (University of Hamburg)
GenerationDrug DiscoveryProtein Structure PredictionDiffusion modelPoint Cloud
🎯 What it does: An end-to-end protein surface and structure joint generation framework called PepBridge is designed, which directly drives the generation of ligand surfaces and complete scaffolds using receptor surface point clouds.
Joint Hierarchical Representation Learning of Samples and Features via Informed Tree-Wasserstein Distance
Ya-Wei Eileen Lin (Technion), Ronen Talmon (Technion)
ClassificationRepresentation LearningGraph Neural NetworkGraphBiomedical Data
🎯 What it does: An iterative framework based on Tree-Wasserstein distance is proposed, which jointly learns hierarchical representations of samples and features, and incorporates Haar wavelet-based filtering in each iteration to enhance the quality of the tree structure.
Joint Modeling of fMRI and EEG Imaging Using Ordinary Differential Equation-Based Hypergraph Neural Networks
YanZhang, Min Li (Central South University)
ClassificationRecognitionGraph Neural NetworkGenerative Adversarial NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: A hypergraph framework FE-NET based on Neural ODE is proposed for asynchronous fMRI-EEG joint modeling and behavior prediction.
Joint Relational Database Generation via Graph-Conditional Diffusion Models
Mohamed Amine Ketata (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
GenerationData SynthesisGraph Neural NetworkDiffusion modelTabular
🎯 What it does: A non-autoregressive relational database generation method based on a graph conditional diffusion model (GRDM) is proposed, which can jointly generate row attributes and primary-foreign key structures for multiple tables.
Joint Velocity-Growth Flow Matching for Single-Cell Dynamics Modeling
Dongyi Wang (Xi'an Jiaotong University), Jian Sun (Xi'an Jiaotong University)
Flow-based ModelBiomedical DataOrdinary Differential Equation
🎯 What it does: This paper proposes a Velocity-Growth Flow Matching (VGFM) framework that utilizes semi-relaxed optimal transport theory to jointly learn the velocity field of single-cell state transitions and the growth function of cell mass, thereby reconstructing the time evolution dynamics of single cells.
Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning
Hugues Van Assel (Genentech), Randall Balestriero (Meta AI)
Representation LearningData-Centric LearningConvolutional Neural NetworkTransformerContrastive LearningImageBiomedical Data
🎯 What it does: This study investigates the two paradigms of reconstruction and joint embedding in self-supervised learning, providing their closed-form solutions and analyzing the impact of data augmentation on the learning effectiveness of both.
Jury-and-Judge Chain-of-Thought for Uncovering Toxic Data in 3D Visual Grounding
Kaixiang Huang (Zhejiang University), Shengfeng He (Singapore Management University)
Anomaly DetectionTransformerLarge Language ModelPoint CloudChain-of-Thought
🎯 What it does: A Refer-Judge framework is constructed, utilizing multi-perspective and multi-role Chain-of-Thought to evaluate 3D visual localization data, identifying and filtering 'poisonous samples'.
Just One Layer Norm Guarantees Stable Extrapolation
Juliusz Ziomek (University of Oxford), Michael A Osborne
Protein Structure PredictionTabularBiomedical Data
🎯 What it does: This study investigates the impact of adding a single layer of LayerNorm on the out-of-distribution performance of models in infinitely wide networks, providing both theoretical proof and experimental validation.
K-DeCore: Facilitating Knowledge Transfer in Continual Structured Knowledge Reasoning via Knowledge Decoupling
Yongrui Chen (Southeast University), Tianxing Wu (Southeast University)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The K-DECORE framework is proposed, achieving continuous learning for heterogeneous structured knowledge reasoning tasks through knowledge decoupling, dual-view memory, and structure-guided pseudo-query synthesis.
KAIROS: Scalable Model-Agnostic Data Valuation
Jiongli Zhu (University of California San Diego), Babak Salimi (University of California San Diego)
Anomaly DetectionData-Centric LearningImageText
🎯 What it does: A scalable model-agnostic data evaluation framework called KAIROS is proposed, which uses the closed-form influence function of Maximum Mean Discrepancy (MMD) to measure the contribution of training samples to model performance.
KaRF: Weakly-Supervised Kolmogorov-Arnold Networks-based Radiance Fields for Local Color Editing
Wudi Chen (Jilin University), Ce Zhu
Image TranslationSegmentationNeural Radiance FieldImage
🎯 What it does: This paper proposes KaRF, a weakly supervised Kolmogorov-Arnold Network (KAN) driven two-stage radiance field for achieving high-quality local color editing.
KARMA: Leveraging Multi-Agent LLMs for Automated Knowledge Graph Enrichment
Yuxing Lu (Peking University), Jinzhuo Wang (Peking University)
TransformerLarge Language ModelAgentic AIPrompt EngineeringTextBiomedical Data
🎯 What it does: The KARMA framework is proposed, which utilizes multi-agent large language models (LLM) to automatically extract, verify, and integrate new knowledge from scientific literature, achieving the supplementation of knowledge graphs (KG).
KeeA*: Epistemic Exploratory A* Search via Knowledge Calibration
Dengwei Zhao (Shanghai Jiao Tong University), Lei Xu (Shanghai Jiao Tong University)
OptimizationDrug DiscoveryTabular
🎯 What it does: An improved A*-based search algorithm called KeeA* is proposed, which enhances search efficiency and solution quality by introducing 'cluster scouting' and 'path-aware' sampling to calibrate the selection of candidate nodes.
Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning
Yaxin Hou (Southeast University), Junhui Hou (City University of Hong Kong)
ClassificationContrastive LearningImage
🎯 What it does: A controllable pseudo-label generation framework (CPG) is proposed to address the challenge of unknown distribution of unlabeled data in long-tail semi-supervised learning; it dynamically filters controllable labels, iteratively constructs a label set with known distribution, and uses logit adjustment to obtain a Bayesian optimal classifier.
Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedy
Jie Ren (Michigan State University), Hui Liu (IBM T. J. Watson Research Center)
OptimizationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a stealthy attack (SA) method for fine-tuning unlearning in large language models and designs a Scope-aware Unlearning (SU) defense scheme.
Kernel conditional tests from learning-theoretic bounds
Pierre-François Massiani (RWTH Aachen University), Sebastian Trimpe (RWTH Aachen University)
Time Series
🎯 What it does: A conditional hypothesis testing framework based on learning theory confidence bounds is proposed, and point-level determinations of conditional expectations, functions, and two-sample tests are implemented on kernel ridge regression.
Kernel Density Steering: Inference-Time Scaling via Mode Seeking for Image Restoration
Yuyang Hu (Washington University in St. Louis), Mauricio Delbracio (Google)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes Kernel Density Steering (KDS), an image restoration framework that utilizes a set of N particles for local pattern seeking during the inference phase.
Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization
Antonio H. Ribeiro (Uppsala University), Francis Bach (PSL Research University / INRIA)
OptimizationComputational EfficiencyAdversarial AttackTabular
🎯 What it does: This paper proposes a kernel learning method for adversarial perturbations in the Reproducing Kernel Hilbert Space (RKHS) and provides an efficient iterative kernel ridge regression algorithm.
Kernel Regression in Structured Non-IID Settings: Theory and Implications for Denoising Score Learning
Dechen Zhang (University of Hong Kong), Difan Zou (University of Hong Kong)
RestorationOptimizationScore-based ModelTabular
🎯 What it does: This paper studies the generalization performance of Kernel Ridge Regression (KRR) under structured non-i.i.d. data (different noise observations of the same signal) and applies this theory to denoising score learning.
Kernel von Mises Formula of the Influence Function
Yaroslav Mukhin (Cornell University)
Optimization
🎯 What it does: By spectral decomposition, the influence function (IF) is expressed as a linear combination of path derivatives, proposing a low-rank approximate estimator based on kernel RKHS, and proving its consistency.
Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning
Yijin Ni (Georgia Institute of Technology), Xiaoming Huo (Georgia Institute of Technology)
Representation Learning
🎯 What it does: A kernel function-based equal probability criterion EO_k is proposed to quantitatively measure the accuracy-fairness trade-off in fair representation learning.
KeyDiff: Key Similarity-Based KV Cache Eviction for Long-Context LLM Inference in Resource-Constrained Environments
Junyoung Park (Qualcomm AI Research), Christopher Lott (Qualcomm AI Research)
RetrievalCompressionOptimizationComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: A training-free, key similarity-based KV cache eviction method called KEYDIFF is proposed, which can strictly control memory usage during long context LLM inference while maintaining high accuracy.
KGGen: Extracting Knowledge Graphs from Plain Text with Language Models
Belinda Mo (Stanford University), Sanmi Koyejo (Stanford University)
GenerationRetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: A method has been developed to automatically extract knowledge graphs (KGGen) from raw text using large language models (LLM), and a new evaluation benchmark MINE (MINE-1 and MINE-2) has been proposed to measure the quality of KG generation.
Kinaema: a recurrent sequence model for memory and pose in motion
Mert Bülent Sarıyıldız (NAVER LABS Europe), Christian Wolf (NAVER LABS Europe)
Pose EstimationRobotic IntelligenceRecurrent Neural NetworkTransformerReinforcement LearningSequential
🎯 What it does: A memory model called Kinaema based on recursive transformers is proposed to integrate visual observations and perform relative pose estimation in continuous robotic operations, which is then embedded into the DEBiT navigation agent to achieve the Mem-Nav task.
KINDLE: Knowledge-Guided Distillation for Prior-Free Gene Regulatory Network Inference
Rui Peng (Peking University), Jinzhuo Wang (Peking University)
Knowledge DistillationTransformerTime SeriesBiomedical Data
🎯 What it does: The KINDLE framework is proposed, utilizing knowledge distillation to achieve prior-free gene regulatory network inference.
Kinetics: Rethinking Test-Time Scaling Law
Ranajoy Sadhukhan (Carnegie Mellon University), Beidi Chen (Carnegie Mellon University)
OptimizationComputational EfficiencyTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: A new Kinetics Scaling Law is proposed, which explains that focusing on cost (attention) rather than the number of parameters is the dominant factor during inference, and redefines the resource allocation strategy between model size and generation length.
KL Penalty Control via Perturbation for Direct Preference Optimization
Sangkyu Lee (Yonsei University), Youngjae Yu (Seoul National University)
Recommendation SystemOptimizationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The ε-DPO method is proposed, which enhances the preference alignment effect of large language models by adaptively adjusting the KL penalty coefficient β at the instance level during the training process.
KL-Regularized RLHF with Multiple Reference Models: Exact Solutions and Sample Complexity
Gholamali Aminian (Alan Turing Institute), Youssef Mroueh (IBM Research)
Reinforcement Learning from Human FeedbackReinforcement LearningText
🎯 What it does: A regularization framework for RLHF with multiple reference models is proposed, providing closed-form solutions for reverse KL and forward KL, along with sample complexity analysis.
KLASS: KL-Guided Fast Inference in Masked Diffusion Models
Seo Hyun Kim (KAIST), Se-Young Yun (KAIST)
GenerationData SynthesisComputational EfficiencyDiffusion modelText
🎯 What it does: Proposed KL-Adaptive Stability Sampling (KLASS), a training-free, token-level KL divergence and confidence-based parallel decoding method.
Knee-Deep in C-RASP: A Transformer Depth Hierarchy
Andy Yang (University of Notre Dame), David Chiang (University of Notre Dame)
TransformerSequential
🎯 What it does: Investigate the impact of Transformer depth on expressive power, proving that fixed-precision Transformers are equivalent to C-RASP/TL[#] and establishing a strict depth hierarchy;
Know Thyself by Knowing Others: Learning Neuron Identity from Population Context
Vinam Arora (University of Pennsylvania), Eva L Dyer
ClassificationRepresentation LearningTransformerContrastive LearningBiomedical Data
🎯 What it does: This work proposes a self-supervised framework called NuCLR to learn neuron identity representations from neuronal population activity, supporting zero-shot classification.
Know What You Don't Know: Uncertainty Calibration of Process Reward Models
Young-Jin Park (Massachusetts Institute of Technology), Navid Azizan (Massachusetts Institute of Technology)
OptimizationComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: A process reward model (PRM) calibration method based on quantile regression is proposed, and the calibrated PRM is used for instance-adaptive inference time scaling (IAS), achieving dynamic allocation of computational resources based on instance demands in mathematical reasoning tasks.
Knowledge Distillation Detection for Open-weights Models
Qin Shi (Purdue University), Raymond A. Yeh (Purdue University)
Knowledge DistillationImageText
🎯 What it does: This paper proposes a knowledge distillation detection task, aiming to determine whether a student model originates from a certain teacher based solely on the student model weights and the teacher API.
Knowledge Distillation of Uncertainty using Deep Latent Factor Model
Sehyun Park (Seoul National University), Yongdai Kim (Seoul National University)
ClassificationKnowledge DistillationImageTabular
🎯 What it does: A Gaussian distribution distillation method based on the Deep Latent Factor model (DLF) is proposed to compress deep ensembles while preserving uncertainty.
Knowledge Graph Enhanced Generative Multi-modal Models for Class-Incremental Learning
Xusheng Cao (Nankai University), Ming-Ming Cheng (Nankai University)
ClassificationGenerationTransformerLarge Language ModelImageTextMultimodality
🎯 What it does: Proposes the KG-GMM method, which combines knowledge graphs with multimodal generative large models for incremental learning, significantly alleviating catastrophic forgetting.
Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better
Danny Driess, Sergey Levine
Computational EfficiencyRobotic IntelligenceTransformerVision-Language-Action ModelMultimodality
🎯 What it does: This paper proposes a new Knowledge Insulation training method to transform large-scale pre-trained Vision-Language Models (VLM) into real-time continuous action control Vision-Language-Action (VLA) models, balancing fast training, real-time inference, and better generalization capabilities.
Knowledge Starts with Practice: Knowledge-Aware Exercise Generative Recommendation with Adaptive Multi-Agent Cooperation
Yangtao Zhou (Xidian University), Qingshan Li (Xidian University)
Recommendation SystemTransformerLarge Language ModelGenerative Adversarial NetworkTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the Knowledge-Aware Exercise Generation Recommendation (KEGR) task and constructs a multi-agent collaborative framework based on LLM called ExeGen, which dynamically perceives students' knowledge states and generates personalized exercises.
Knowledge-based Visual Question Answer with Multimodal Processing, Retrieval and Filtering
Yuyang Hong (University of Chinese Academy of Sciences), Jieping Ye
RetrievalTransformerReinforcement LearningVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A three-stage Wiki-PRF framework is proposed, which first processes images using visual tools, then combines multimodal retrieval to acquire knowledge, and finally filters the retrieval results to generate answers.
KOALA++: Efficient Kalman-Based Optimization with Gradient-Covariance Products
Zixuan Xia (University of Bern), Paolo Favaro (University of Bern)
OptimizationImageText
🎯 What it does: KOALA++, a scalable optimization algorithm based on Kalman filtering, is proposed to explicitly model the structured gradient uncertainty in neural network training.
KORGym: A Dynamic Game Platform for LLM Reasoning Evaluation
Jiajun Shi (Beihang University), Ge Zhang (ByteDance Seed)
TransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: A gamified evaluation platform called KORGym is proposed, which includes over 50 text and visual games, supporting multi-round interactions and reinforcement learning scenarios, aimed at assessing the reasoning capabilities of large language models (LLMs) and visual language models (VLMs).
KScope: A Framework for Characterizing the Knowledge Status of Language Models
Yuxin Xiao (Massachusetts Institute of Technology), Marzyeh Ghassemi (Massachusetts Institute of Technology)
Large Language ModelTextBiomedical Data
🎯 What it does: Five types of knowledge states based on knowledge consistency and correctness are proposed, and a hierarchical verification framework KScope is constructed to provide a refined description and classification of the knowledge states of large language models in question-answering tasks.
KSP: Kolmogorov-Smirnov metric-based Post-Hoc Calibration for Survival Analysis
Jeongho Park (Yonsei University), Gwangsu Kim (Jeonbuk National University)
Biomedical Data
🎯 What it does: A post-processing calibration method based on the Kolmogorov–Smirnov (KS) statistic (KSP) is proposed to improve the predictive calibration of various survival models while maintaining discriminative performance.
KTAE: A Model-Free Algorithm to Key-Tokens Advantage Estimation in Mathematical Reasoning
Wei Sun (Institute of Automation Chinese Academy of Sciences), Jiajun Zhang (Institute of Automation Chinese Academy of Sciences)
Large Language ModelTextBenchmark
🎯 What it does: A model-free Keyword Advantage Estimation (KTAE) algorithm is proposed, which generates fine-grained token-level advantage signals by statistically associating correct and incorrect answers in the sampled rounds.
KungfuBot: Physics-Based Humanoid Whole-Body Control for Learning Highly-Dynamic Skills
Weiji Xie (China Telecom), Xuelong Li (China Telecom)
Robotic IntelligenceReinforcement LearningVideoPhysics Related
🎯 What it does: A motion processing pipeline and adaptive reward mechanism based on physical constraints have been constructed, and a control strategy for humanoid robots capable of executing high-dynamic kung fu and dance movements has been trained and deployed.
Kuramoto Orientation Diffusion Models
Yue Song (California Institute of Technology), Max Welling (University of Amsterdam)
GenerationData SynthesisDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: A periodic domain score-based generative model based on Kuramoto coupling dynamics is proposed, achieving directionally rich image generation through a synchronization-desynchronization process.
KVCOMM: Online Cross-context KV-cache Communication for Efficient LLM-based Multi-agent Systems
Hancheng Ye (Duke University), Yiran Chen (Duke University)
TransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Proposes the KVCOMM framework, which dynamically reuses the KV cache of multiple agent LLMs using an anchor pool, significantly reducing pre-filling latency;
KVFlow: Efficient Prefix Caching for Accelerating LLM-Based Multi-Agent Workflows
Zaifeng Pan (University of California San Diego), Yufei Ding (Amazon Web Services)
TransformerLarge Language ModelAgentic AITextFinance Related
🎯 What it does: Design and implement KVFlow for fine-grained management of workflow-aware prefix KV caching in multi-agent LLM workflows.
KVLink: Accelerating Large Language Models via Efficient KV Cache Reuse
Jingbo Yang (University of California Santa Barbara), Shiyu Chang (University of California Santa Barbara)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the KVLINK method, which precomputes the KV cache of documents in large language models and reuses it during inference to avoid redundant encoding.
KVzip: Query-Agnostic KV Cache Compression with Context Reconstruction
Jang-Hyun Kim (Seoul National University), Hyun Oh Song (Seoul National University)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes KVzip, a query-independent KV cache compression method that evaluates the importance of KV by utilizing LLM to reconstruct the context, thereby compressing the KV cache while maintaining model performance.
L-MTP: Leap Multi-Token Prediction Beyond Adjacent Context for Large Language Models
Xiaohao Liu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a Leap Multi-Token Prediction (L-MTP) method, which extends Multi-Token Prediction (MTP) to simultaneously enhance inference speed and model performance.
L$^2$M: Mutual Information Scaling Law for Long-Context Language Modeling
Zhuo Chen (NSF AI Institute for Artificial Intelligence and Fundamental Interactions), Marin Soljacic (NSF AI Institute for Artificial Intelligence and Fundamental Interactions)
TransformerLarge Language ModelText
🎯 What it does: A theoretical framework for long-context language modeling based on bipartite mutual information is proposed, and it is verified that the bipartite mutual information in natural language grows according to a power law with sequence length, providing a lower bound for the model's historical state dimension.
L2DGCN: Learnable Enhancement and Label Selection Dynamic Graph Convolutional Networks for Mitigating Degree Bias
jingxiao zhang, Xuan Li (China University of Mining and Technology)
ClassificationGraph Neural NetworkContrastive LearningGraphBenchmark
🎯 What it does: A teacher-student framework L2DGCN is proposed, which utilizes soft label propagation to generate pseudo-labels and enhances node classification under sparse labeling through a learnable graph structure enhancement and a pseudo-label selector.