NeurIPS 2025 Papers — Page 9
Conference on Neural Information Processing Systems · 5275 papers
Conditional Diffusion Anomaly Modeling on Graphs
Chunyu Wei (Renmin University of China), Yunhai Wang (Renmin University of China)
Anomaly DetectionGraph Neural NetworkDiffusion modelGraphFinance Related
🎯 What it does: A graph anomaly detection method based on a conditional diffusion model, CGADM, is proposed, transforming anomaly detection into generative modeling.
Conditional Distribution Compression via the Kernel Conditional Mean Embedding
Dominic Broadbent (University of Bristol), Tom Lovett (University of Oxford)
CompressionOptimizationTabular
🎯 What it does: A conditional distribution compression method for labeled data is proposed, introducing the Average Maximum Conditional Mean Difference (AMCMD) metric. Based on this, two linear-time compression algorithms are proposed: Average Conditional Kernel Herding (ACKH) and Average Conditional Kernel Inducing Points (ACKIP), along with an extension for joint distribution compression (JKH, JKIP).
Conditional Forecasts and Proper Scoring Rules for Reliable and Accurate Performative Predictions
Philip Boeken (University of Amsterdam), Joris M. Mooij (University of Amsterdam)
🎯 What it does: The research investigates how to achieve reliable and accurate performative predictions by conditioning predictions and using appropriate scoring rules in situations where the predictions may affect the predicted outcomes.
Conditional Gradient Methods with Standard LMO for Stochastic Simple Bilevel Optimization
Khanh-Hung Giang-Tran (Cornell University), Nam Ho-Nguyen (University of Sydney)
OptimizationHyperparameter SearchTabular
🎯 What it does: An iterative regularization projection-free conditional gradient algorithm is proposed to solve a single-layer simple two-layer optimization problem with randomness using a linear optimization oracle.
Conditional Panoramic Image Generation via Masked Autoregressive Modeling
Chaoyang Wang (Peking University), Yunhai Tong (Peking University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: A unified mask autoregressive model PAR is designed for text-to-panoramic image, panoramic extension, and editing tasks.
Conditional Representation Learning for Customized Tasks
Honglin Liu (Sichuan University), Xi Peng (Sichuan University)
ClassificationRetrievalRepresentation LearningTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: This paper studies a Conditional Representation Learning method (CRL) that utilizes large language models to generate descriptive vocabulary to construct a textual basis. It then projects image features onto this basis through a visual-language model, thereby generating conditional image representations based on user-specified semantic criteria (such as color, shape, texture, etc.) for customized tasks like classification, clustering, and retrieval.
Conditioning Matters: Training Diffusion Policies is Faster Than You Think
Zibin Dong (Tianjin University), Jianye HAO
OptimizationRobotic IntelligenceTransformerDiffusion modelAuto EncoderMultimodality
🎯 What it does: This paper addresses the issue of loss collapse in diffusion strategy training caused by the difficulty in distinguishing conditions, proposing to enhance training efficiency and performance by associating the source distribution with conditions (Cocos).
Confidence-Aware With Prototype Alignment for Partial Multi-label Learning
Weijun Lv (Guangdong University of Technology), Sixian Chan (Zhejiang University of Technology)
ClassificationSupervised Fine-Tuning
🎯 What it does: This paper proposes a part multi-label learning method based on mutually aligned prototypes, CAPML, which aligns unsupervised clustering with weakly supervised label prototypes and utilizes a confidence indicator for label denoising and classifier training.
Conflict-Aware Knowledge Editing in the Wild: Semantic-Augmented Graph Representation for Unstructured Text
Zhange Zhang (Beihang University), Xianglong Liu
Graph Neural NetworkTransformerLarge Language ModelTextBenchmark
🎯 What it does: The CAKE framework is proposed for knowledge editing of wild unstructured text, addressing representation ambiguity and editing conflicts.
Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Models
William Overman (Stanford Graduate School of Business), Mohsen Bayati (Stanford Graduate School of Business)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A post-routing framework called Conformal Arbitrage is proposed, which combines the scores of black-box language models with Conformal Risk Control (CRC) to achieve a trade-off between two objectives (such as cost and accuracy, or help and safety) while satisfying a preset risk budget.
Conformal Inference under High-Dimensional Covariate Shifts via Likelihood-Ratio Regularization
Sunay Joshi (University of Pennsylvania), Hamed Hassani (University of Pennsylvania)
Tabular
🎯 What it does: A conformal prediction method is proposed that does not require explicit estimation of the likelihood ratio under high-dimensional co-migration;
Conformal Information Pursuit for Interactively Guiding Large Language Models
Kwan Ho Ryan Chan (University of Pennsylvania), Rene Vidal
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes an interactive question-answering framework called Conformal Information Pursuit (C-IP), which measures uncertainty by using the size of the prediction set in large language models (LLMs) to select the most informative queries at each step, thereby reducing the number of queries and improving prediction accuracy.
Conformal Linguistic Calibration: Trading-off between Factuality and Specificity
Zhengping Jiang (Johns Hopkins University), Benjamin Van Durme (Johns Hopkins University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the Conformal Linguistic Calibration (CLC) framework, which allows language models to express uncertainty by generating statements that are not sufficiently precise, while maintaining the integrity of the answers and controlling the probability of factual errors.
Conformal Mixed-Integer Constraint Learning with Feasibility Guarantees
Daniel Ovalle (Carnegie Mellon University), Mateo Dulce Rubio (New York University)
OptimizationComputational EfficiencyTabular
🎯 What it does: A Conformal Mixed-Integer Constraint Learning (C-MICL) framework is proposed to embed data-driven constraints in optimization problems and provide probabilistic feasibility guarantees.
Conformal Online Learning of Deep Koopman Linear Embeddings
Ben Gao (Université Jean Monnet Saint-Etienne), Olivier Alata (Université Jean Monnet Saint-Etienne)
TabularTime Series
🎯 What it does: An online learning framework named COLoKe is proposed, which adaptively updates deep Koopman embeddings using a conformal mechanism.
Conformal Prediction Beyond the Horizon: Distribution-Free Inference for Policy Evaluation
Feichen Gan (East China Normal University), Yukun Liu (East China Normal University)
Reinforcement LearningSequential
🎯 What it does: A distribution-independent prediction interval framework for infinite-horizon Markov decision processes is proposed to evaluate the reward distribution of a given policy.
Conformal Prediction Beyond the Seen: A Missing Mass Perspective for Uncertainty Quantification in Generative Models
Sima Noorani (University of Pennsylvania), Hamed Hassani (University of Pennsylvania)
GenerationData SynthesisTransformerLarge Language ModelText
🎯 What it does: A conformal prediction framework CPQ is proposed from the perspective of missing mass to quantify uncertainty in situations where only generative models can be queried.
Conformal Prediction for Causal Effects of Continuous Treatments
Maresa Schröder, Stefan Feuerriegel (Ludwig Maximilian University of Munich)
Biomedical DataElectronic Health Records
🎯 What it does: A method for finite sample credible prediction intervals of potential outcomes for continuous treatment variables is proposed, which can provide valid confidence intervals in the presence of known or unknown propensity scores, addressing the coverage failure issue of traditional methods under continuous treatment and unknown propensity scores.
Conformal Prediction for Ensembles: Improving Efficiency via Score-Based Aggregation
Yash Patel, Ambuj Tewari
OptimizationComputational EfficiencyScore-based ModelTabular
🎯 What it does: A conformal score aggregation (CSA) framework based on multidimensional scores and quantile envelopes is proposed to generate efficient prediction intervals without distributional assumptions in ensemble models.
Conformal Prediction for Time-series Forecasting with Change Points
Sophia Huiwen Sun (University of California San Diego), Rose Yu (University of California San Diego)
Anomaly DetectionOptimizationTime Series
🎯 What it does: The CPTC algorithm is proposed, which combines state prediction from state space models with online shape prediction to generate reliable prediction intervals on time series with change points.
Conformal Prediction in The Loop: A Feedback-Based Uncertainty Model for Trajectory Optimization
Han Wang (Shanghai Jiao Tong University), Chao Ning (Shanghai Jiao Tong University)
Autonomous DrivingOptimizationRecurrent Neural NetworkReinforcement LearningTime SeriesSequential
🎯 What it does: A feedback-based conformal prediction (Fb-CP) framework is proposed for trajectory optimization in uncertain environments, capable of real-time adjustment of the prediction region and risk allocation.
Conformal Prediction under Lévy-Prokhorov Distribution Shifts: Robustness to Local and Global Perturbations
Liviu Aolaritei (University of California Berkeley), Youssef Marzouk (Massachusetts Institute of Technology)
ClassificationDomain AdaptationImage
🎯 What it does: A robust consistency prediction method for modeling distribution shifts using Lévy-Prokhorov (LP) fuzzy sets is proposed, which provides theoretical coverage guarantees for both local and global perturbations.
Conformal Risk Training: End-to-End Optimization of Conformal Risk Control
Christopher Yeh (California Institute of Technology), Yisong Yue (California Institute of Technology)
SegmentationOptimizationImageBiomedical Data
🎯 What it does: A distribution-independent, finite-sample risk control mechanism for optimizing coherent risk equivalence (OCE) (including CVaR) is proposed, which is embedded into the model training process, forming a 'synthetic risk training' method.
Confounding Robust Deep Reinforcement Learning: A Causal Approach
Mingxuan Li (Columbia University), Elias Bareinboim (Columbia University)
Convolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImageSequential
🎯 What it does: This paper proposes a deep Q-learning algorithm called Causal-DQN that can learn robust policies in offline reinforcement learning environments with unobserved confounding factors.
ConfTuner: Training Large Language Models to Express Their Confidence Verbally
Yibo Li (National University of Singapore), Bryan Hooi (National University of Singapore)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The ConfTuner framework is proposed, which fine-tunes the tokenized Brier loss on LLMs, enabling the model to express confidence that aligns with actual reliability.
Confusion-Driven Self-Supervised Progressively Weighted Ensemble Learning for Non-Exemplar Class Incremental Learning
Kai Hu (Xiangtan University), Xieping Gao (Hunan Normal University)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: The CLOVER framework is proposed to address the issue of representation overlap between old and new classes in non-example class incremental learning (NECIL);
Connecting Jensen–Shannon and Kullback–Leibler Divergences: A New Bound for Representation Learning
Reuben Dorent, William M Wells
OptimizationRepresentation LearningTabularSequential
🎯 What it does: A new lower bound of KLD for JSD is proposed, proving that maximizing JSD is equivalent to optimizing the lower bound of mutual information, and the relationship between cross-entropy loss and JSD is provided.
Connecting Neural Models Latent Geometries with Relative Geodesic Representations
Hanlin Yu, Marco Fumero (Institute of Science and Technology Austria)
GenerationRetrievalRepresentation LearningAuto EncoderImageMultimodality
🎯 What it does: This study investigates the parameterization of different neural networks learning the same latent manifold on similar data and proposes a method to construct relative geodesic representations using pulled-back Riemannian metrics to achieve unsupervised communication between latent spaces.
Connectome-Based Modelling Reveals Orientation Maps in the Drosophila Optic Lobe
Jia-Nuo Liew (Tsinghua University), Xiaolin Hu (Tsinghua University)
Spiking Neural NetworkGraph
🎯 What it does: Based on the complete connectome of the fruit fly brain, the LIF neuron model was used to simulate retinal input, and by stimulating striped images in different directions, neuronal activation in the optical lobe was observed, leading to the discovery and mapping of direction selectivity maps and columnar structures.
Consensus-Robust Transfer Attacks via Parameter and Representation Perturbations
Shixin Li (Huazhong University of Science and Technology), Bin Benjamin Zhu (Microsoft Corporation)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: A transfer attack based on consistent robustness, CORTA, is proposed, which enhances the transferability of black-box attacks using parameter perturbation and representation mixing.
Conservative classifiers do consistently well with improving agents: characterizing statistical and online learning
Dravyansh Sharma (Northwestern University), Alec Sun (University of Chicago)
ClassificationOptimization
🎯 What it does: This paper studies supervised learning in scenarios of strategic improvements, providing PAC learning and online error bounds under various conditions (arbitrary improvement functions, Euclidean ball improvement sets, noise, online learning).
Consistency Conditions for Differentiable Surrogate Losses
Drona Khurana (University of Colorado Boulder), Rafael Frongillo (University of Colorado Boulder)
Optimization
🎯 What it does: This paper proposes and studies the calibration conditions for differentiable convex surrogate losses, proving the equivalence of indirect inducement (IE) and calibration in one dimension, and providing a stronger strong indirect inducement (strong IE) in multi-dimensional cases, proving its sufficiency for calibration and necessity in strongly convex cases. It also constructs a one-dimensional differentiable surrogate that achieves calibration for any rankable objective and applies it to ordinal regression.
Consistency of Physics-Informed Neural Networks for Second-Order Elliptic Equations
Yuqian Cheng (Tsinghua University), Qian Lin (Tsinghua University)
OptimizationPhysics Related
🎯 What it does: This study investigates the consistency issue of Physics-Informed Neural Networks (PINN) under second-order elliptic equations and provides a unified framework for training gradient flows and kernel methods.
Consistency of the $k_n$-nearest neighbor rule under adaptive sampling
Robi Bhattacharjee (University of Tübingen), Sanjoy Dasgupta (University of California San Diego)
Classification
🎯 What it does: This paper studies the long-term consistency problem of the k-nearest neighbors (k-NN) algorithm under noisy labels within the framework of adaptive sampling online learning.
Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning
Kongcheng Zhang (Zhejiang University), Dacheng Tao (Nanyang Technological University)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: A self-rewarding reinforcement learning framework called CoVo is proposed, which constructs internal rewards by utilizing the consistency and volatility of intermediate states in LLM reasoning trajectories, enhancing reasoning capabilities without external supervision.
Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models
Michael Plainer (Freie Universität Berlin), Frank Noe (Microsoft Research)
OptimizationDrug DiscoveryGraph Neural NetworkTransformerMixture of ExpertsDiffusion modelScore-based ModelBiomedical Data
🎯 What it does: This study trained an energy-based diffusion model and achieved consistency between sampling and simulation by incorporating Fokker-Planck regularization. It also proposed a Mixture of Experts method for time periods and constructed a transferable dipeptide Boltzmann simulator, applying this method to coarse-grained biomolecules such as dipeptides, Chignolin, and BBA.
Consistent Story Generation: Unlocking the Potential of Zigzag Sampling
Mingxiao Li (KU Leuven), Marie-Francine Moens (KU Leuven)
GenerationTransformerDiffusion modelImageTextMultimodality
🎯 What it does: A training-agnostic zigzag sampling strategy is proposed, which enhances subject consistency in visual story generation by injecting visual semantics during the zig step, using empty prompts during the zag step, and employing only text prompts during the generation step.
Consistent Supervised-Unsupervised Alignment for Generalized Category Discovery
Jizhou Han (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
ClassificationRepresentation LearningImage
🎯 What it does: A Generalized Category Discovery (GCD) framework NC-GCD based on fixed Equiangular Tight Frame (ETF) prototypes is proposed, unifying supervised and unsupervised feature alignment, and stabilizing clustering labels through a Semantic Consistency Matcher (SCM), significantly improving classification performance for both known and new categories.
Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning
Marwa Abdulhai (University of California Berkeley), Natasha Jaques (University of Washington)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This study investigates how to maintain a consistent persona for large language models (LLMs) in multi-turn dialogues and improve their consistency through reinforcement learning.
Constant Bit-size Transformers Are Turing Complete
Qian Li (Shenzhen International Center For Industrial And Applied Mathematics), Yuyi Wang (CRRC Zhuzhou Institute)
Transformer
🎯 What it does: This paper proves that a Transformer with any fixed number of bits can simulate any Turing machine as long as the context window is sufficiently long, thus demonstrating Turing completeness.
Constrained Best Arm Identification
Tyron Darnell Lardy (Centrum Wiskunde & Informatica and Leiden University), Wouter M Koolen
OptimizationTabular
🎯 What it does: The paper presents the Cost-Constrained Best Arm Identification (CBAI) problem, provides an information-theoretic lower bound, designs a general algorithm based on Track-and-Stop and GLR, and achieves and proves asymptotic optimality across three types of arm distributions (fixed covariance Gaussian, unknown covariance Gaussian, and non-parametric two-dimensional distributions);
Constrained Diffusers for Safe Planning and Control
Jichen Zhang (University of Oxford), Jack Umenberger (University of Oxford)
OptimizationRobotic IntelligenceDiffusion modelScore-based ModelSequential
🎯 What it does: Based on pre-trained diffusion models, a distribution-level constraint is introduced during the reverse diffusion process to achieve safe planning and control; online execution is completed using discrete control barrier functions and inverse dynamics models.
Constrained Discrete Diffusion
Michael Cardei (University of Virginia), Ferdinando Fioretto (University of Virginia)
GenerationOptimizationDrug DiscoveryLarge Language ModelDiffusion modelTextSequential
🎯 What it does: This paper proposes Constrained Discrete Diffusion (CDD), a generative framework that combines differential discrete diffusion models with differentiable constrained optimization to directly enforce global sequence constraints during the generation process.
Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models
Taha Entesari (Johns Hopkins University), Mahyar Fazlyab (Johns Hopkins University)
OptimizationTransformerLarge Language ModelText
🎯 What it does: This paper proposes a framework that views the 'forgetting' problem of large language models as a constrained optimization problem, and achieves uniformity of the forgetting set through a new 'logit-margin flattening' loss; it employs a scalable primal-dual algorithm to achieve efficient forgetting while maintaining the performance of the retained set.
Constrained Feedback Learning for Non-Stationary Multi-Armed Bandits
Shaoang Li (Stony Brook University), Jian Li (Stony Brook University)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: A new non-stationary multi-armed bandit (NSMAB) model is proposed—CONFEE-NSMAB, in which the reward feedback is constrained by a query budget; and the first algorithm without prior knowledge, capable of achieving approximately optimal dynamic returns under this constraint, HYQUE, is designed.
Constrained Linear Thompson Sampling
Aditya Gangrade (Boston University), Venkatesh Saligrama (Boston University)
OptimizationSafty and PrivacyComputational EfficiencyReinforcement LearningTabular
🎯 What it does: A safe linear bandit algorithm based on Thompson sampling, COLTS, along with its hard constraint version S-COLTS and soft constraint version R-COLTS, is proposed. It achieves approximately optimal cumulative rewards and constraint risks while ensuring safety, significantly reducing the computational complexity per round.
Constrained Optimization From a Control Perspective via Feedback Linearization
Runyu Zhang (Massachusetts Institute of Technology), Na Li (Harvard University)
OptimizationFederated LearningTabular
🎯 What it does: A continuous-time constrained optimization framework based on feedback linearization (FL) is proposed and analyzed, covering both equality and inequality constraints, and a momentum-accelerated FL algorithm is designed based on this framework.
Constrained Posterior Sampling: Time Series Generation with Hard Constraints
Sai Shankar Narasimhan (University of Texas at Austin), Sandeep P. Chinchali
GenerationData SynthesisOptimizationDiffusion modelTime SeriesSequentialFinance Related
🎯 What it does: A training-free, diffusion model-based constrained posterior sampling method is proposed for generating time series data that satisfies hard constraints.
Constrained Sampling for Language Models Should Be Easy: An MCMC Perspective
Emmanuel Anaya Gonzalez (University of California San Diego), Loris D'Antoni (University of California San Diego)
GenerationOptimizationLarge Language ModelReinforcement LearningText
🎯 What it does: A constraint sampling framework based on MCMC is proposed, allowing language models to efficiently and feasibly sample from constrained distributions while satisfying context-free grammar constraints.
Constructing an Optimal Behavior Basis for the Option Keyboard
Lucas N. Alegre (Federal University of Rio Grande do Sul), Bruno Castro da Silva
OptimizationRobotic IntelligenceReinforcement LearningAgentic AI
🎯 What it does: A method called OKB is proposed for efficiently constructing the optimal behavior basis of the Option Keyboard, ensuring zero-shot optimality in any linear task.
Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp Generation
Yiyao Ma (Chinese University of Hong Kong), Qi Dou (Chinese University of Hong Kong)
GenerationRobotic IntelligenceDiffusion modelPoint Cloud
🎯 What it does: A grasp transfer framework based on conditional diffusion models is proposed, which migrates contact maps, part maps, and direction maps to new objects through shape templates, achieving task-oriented dexterous grasping.
Context-Aware Hierarchical Learning: A Two-Step Paradigm towards Safer LLMs
Tengyun Ma (Harbin Institute of Technology), Zhuotao Tian (Harbin Institute of Technology)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This paper studies prompt injection attacks in the context of tool invocation, proposing a novel Tool-Completion Attack (TCA), constructing a corresponding benchmark dataset, and designing a Context-Aware Hierarchical Learning (CAHL) mechanism to enhance the security of large language models.
Context-Aware Regularization with Markovian Integration for Attention-Based Nucleotide Analysis
Mohammadsaleh Refahi, Gail Rosen (Drexel University)
TransformerSupervised Fine-TuningBiomedical Data
🎯 What it does: Designed and trained the CARMANIA model, achieving genome sequence modeling of up to 160kbp using sliding window attention and transfer matrix (TM) loss;
ContextAgent: Context-Aware Proactive LLM Agents with Open-world Sensory Perceptions
Bufang Yang (Chinese University of Hong Kong), Zhenyu Yan (Chinese University of Hong Kong)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityBenchmarkChain-of-ThoughtAudio
🎯 What it does: This paper proposes ContextAgent, a LLM agent that utilizes multimodal perception from wearable devices for context awareness and actively invokes tools, along with the design of a corresponding evaluation benchmark, ContextAgentBench.
ConTextTab: A Semantics-Aware Tabular In-Context Learner
Marco Spinaci (SAP France), Sam Thelin (SAP SE)
TransformerTabular
🎯 What it does: A semantic-aware In-Context learning model called ConTextTab is proposed, which is based on a table-native architecture. It utilizes real-world table data for pre-training and employs multi-modal embeddings such as column names, text, and dates during prediction, balancing scalability and semantic understanding.
Contextual Dynamic Pricing with Heterogeneous Buyers
Thodoris Lykouris (Massachusetts Institute of Technology), Julian Zimmert (Google)
Optimization
🎯 What it does: This paper studies the dynamic pricing problem with observable context in the presence of heterogeneous buyer groups, proposing and analyzing an algorithm based on optimistic posterior sampling, proving that it achieves approximately optimal regret loss in both time and context dimensions.
Contextual Integrity in LLMs via Reasoning and Reinforcement Learning
Guangchen Lan (Purdue University), Robert Sim (Microsoft)
Safty and PrivacyTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextChain-of-Thought
🎯 What it does: By explicitly performing Context Integrity (CI) reasoning when large language models complete user tasks and further optimizing the model's reasoning behavior using reinforcement learning, significant reductions in inappropriate information leakage are achieved.
Contextual Online Pricing with (Biased) Offline Data
Yixuan Zhang (University of Wisconsin Madison), Qiaomin Xie (University of Wisconsin Madison)
OptimizationReinforcement LearningTabular
🎯 What it does: The study investigates how to perform contextual online pricing in the presence of biased offline data, proposing three algorithms (CO3, GCO3, RCO3) to achieve adaptive exploration and exploitation.
Contextual Thompson Sampling via Generation of Missing Data
Kelly W. Zhang (Imperial College London), Daniel Russo (Columbia Business School)
Recommendation SystemRecurrent Neural NetworkTransformerReinforcement LearningTabularSequential
🎯 What it does: A generative Thompson Sampling (TS-Gen) is proposed, which fills in missing latent outcomes through offline learning of a generative model and generates complete data online to fit the 'oracle' strategy for contextual bandit decision-making.
Contextual Tokenization for Graph Inverted Indices
Pritish Chakraborty, Abir De
RetrievalGraph Neural NetworkGraph
🎯 What it does: The CORGII framework is proposed for efficiently retrieving graphs containing a given query subgraph from large graph corpora. The core idea is to convert the contextual embeddings of graph nodes into discrete binary tokens, and then use inverted indexing for candidate retrieval, followed by re-ranking using an exact subgraph matching model on the candidate set.
Contimask: Explaining Irregular Time Series via Perturbations in Continuous Time
Max Moebus (ETH Zurich), Christian Holz (ETH Zurich)
Anomaly DetectionExplainability and InterpretabilityNeural Architecture SearchRecurrent Neural NetworkTabularTime SeriesElectronic Health Records
🎯 What it does: The Contimask framework is proposed for post-hoc instance-level explanations of irregular time series models, primarily by perturbing the input using learnable masks in continuous time and observing changes in model output;
Continual Gaussian Mixture Distribution Modeling for Class Incremental Semantic Segmentation
Guilin Zhu (Huazhong University of Science and Technology), Changxin Gao (Huazhong University of Science and Technology)
SegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: The CoGaMiD method is proposed, which utilizes Gaussian Mixture Model to perform multimodal modeling of the feature distribution of learned classes and dynamically updates it during the incremental learning process to achieve class-incremental semantic segmentation.
Continual Knowledge Adaptation for Reinforcement Learning
Jinwu Hu (South China University of Technology), Mingkui Tan (South China University of Technology)
Reinforcement Learning
🎯 What it does: The CKA-RL method is proposed, which maintains a task-specific knowledge vector pool and dynamically utilizes historical knowledge for adaptation, achieving knowledge accumulation and transfer in continuous reinforcement learning while reducing catastrophic forgetting.
Continual Model Merging without Data: Dual Projections for Balancing Stability and Plasticity
Enneng Yang (Shenzhen Campus of Sun Yat-sen University), Jie Zhang (Nanyang Technological University)
OptimizationTransformerMixture of ExpertsImageText
🎯 What it does: A data-free continuous model merging method (DOP) is proposed, achieving model fusion in a scenario of gradually receiving expert models while balancing stability and plasticity.
Continual Multimodal Contrastive Learning
Xiaohao Liu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
ClassificationRetrievalRepresentation LearningContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes a formal framework for Continuous Multimodal Contrastive Learning (CMCL), defining two main objectives: stability and plasticity. It achieves incremental learning of new modality data at the gradient level based on the Dual-Sided Null Space (DNS) method while retaining knowledge of old modalities.
Continual Optimization with Symmetry Teleportation for Multi-Task Learning
Zhipeng Zhou (Nanyang Technological University), Chunyan Miao (Nanyang Technological University)
OptimizationSupervised Fine-TuningImage
🎯 What it does: The research proposes a continuous optimization framework COST based on symmetric moment transfer and LoRA to address gradient conflicts and task imbalance in multi-task learning.
Continual Release Moment Estimation with Differential Privacy
Nikita Kalinin, Christoph H. Lampert (Institute of Science and Technology Austria)
OptimizationSafty and PrivacyGaussian SplattingTabular
🎯 What it does: Proposes the Joint Moment Estimation (JME) algorithm, achieving joint estimation of first and second moments under continuous privacy protection, significantly reducing noise;
Continuity and Isolation Lead to Doubts or Dilemmas in Large Language Models
Hector Pasten (Pontifical Catholic University of Chile), Alexander Kozachinskiy (CENIA)
GenerationData SynthesisTransformerLarge Language ModelTextSequential
🎯 What it does: Analyzes the phenomena of continuity and isolation in Transformers, proving that the decoder-only CPE Transformer has theoretical limitations when learning multiple sequences, and validates its effects through various LLM experiments.
Continuous Concepts Removal in Text-to-image Diffusion Models
Tingxu Han (Nanjing University), Zhenting Wang (Rutgers University)
GenerationOptimizationKnowledge DistillationTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: A continuous concept removal framework (CCRT) is designed, which utilizes knowledge distillation and a genetic algorithm + fuzzy strategy to dynamically generate calibration prompts, capable of gradually removing multiple concepts while maintaining the alignment between text and images.
Continuous Diffusion Model for Language Modeling
Jaehyeong Jo (Korea Advanced Institute of Science and Technology), Sung Ju Hwang (DeepAuto.ai)
GenerationData SynthesisTransformerDiffusion modelImageTextBiomedical Data
🎯 What it does: A continuous diffusion model RDLM based on Riemannian statistical manifolds is proposed, which can map discrete tokens onto the orthogonal sphere and design a continuous diffusion process on that sphere to achieve parallel generation of language, images, and biological sequences.
Continuous Domain Generalization
Zekun Cai (Jilin University), Liang Zhao (Emory University)
Domain AdaptationAdversarial AttackMeta LearningNeural Radiance FieldTabularSequential
🎯 What it does: This paper proposes the Continuous Domain Generalization (CDG) task and achieves continuous transfer of model parameters in a multi-dimensional continuous domain space through the Neural Lie transport operator, addressing the issue of continuous evolution between domains.
Continuous Q-Score Matching: Diffusion Guided Reinforcement Learning for Continuous-Time Control
Chengxiu HUA, Yushun Tang (Huawei Technologies Co., Ltd.)
Reinforcement LearningScore-based ModelTime SeriesStochastic Differential Equation
🎯 What it does: A continuous-time reinforcement learning framework is proposed, utilizing the score function of diffusion policies and the continuous-time Q-function for policy improvement;
Continuous Simplicial Neural Networks
Aref Einizade (Telecom Paris Institute Polytechnique de Paris), Jhony H. Giraldo (Telecom Paris Institute Polytechnique de Paris)
Graph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: A continuous complex-valued neural network (COSIMO) is proposed, utilizing partial differential equations to achieve information propagation on complex numbers;
Continuous Soft Actor-Critic: An Off-Policy Learning Method Robust to Time Discretization
Huimin Han (Zhongtai Securities Institute for Financial Studies Shandong University), Shaolin Ji (Zhongtai Securities Institute for Financial Studies Shandong University)
Reinforcement LearningStochastic Differential Equation
🎯 What it does: This paper proposes the Continuous Soft Actor-Critic (CSAC) and its multi-agent extension (CMASAC) for offline reinforcement learning in continuous-time stochastic environments, addressing the sensitivity issue of time discretization.
Continuous Subspace Optimization for Continual Learning
Quan Cheng (Nanjing University), Lijun Zhang (Nanjing University)
OptimizationTransformerImage
🎯 What it does: A Continuous Subspace Optimization (CoSO) framework is proposed, which dynamically updates the pre-trained model using low-rank subspaces to avoid catastrophic forgetting.
Continuous Thought Machines
Luke Nicholas Darlow (Sakana AI), Llion Jones (Sakana AI)
Recurrent Neural NetworkImageSequential
🎯 What it does: Proposes the Continuous Thought Machine (CTM) — a neural network model that unfolds step by step on an internal time axis, achieving cognition through neuron-level dynamics and neural synchronization.
Continuous-time Riemannian SGD and SVRG Flows on Wasserstein Probabilistic Space
Mingyang Yi (Renmin University of China), Bohan Wang (Alibaba Group)
OptimizationTabularStochastic Differential Equation
🎯 What it does: This paper proposes a continuous Riemannian stochastic gradient descent (SGD) and stochastic variance reduced gradient (SVRG) flow on Wasserstein space to minimize KL divergence.
Contrastive Consolidation of Top-Down Modulations Achieves Sparsely Supervised Continual Learning
Viet Anh Khoa Tran (Peter Grünberg Institute Forschungszentrum Jülich and RWTH Aachen), Willem A.M. Wybo
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A biologically inspired continuous learning framework TMCL is proposed, utilizing task-specific upsampling under sparse supervision (top-down modulations) and solidifying it into a feedforward network to achieve class-incremental continuous learning and transfer learning.
Contrastive Learning with Data Misalignment: Feature Purity, Training Dynamics and Theoretical Generalization Guarantees
Jiawei Sun (Rensselaer Polytechnic Institute), Meng Wang (Rensselaer Polytechnic Institute)
ClassificationRepresentation LearningRecurrent Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: This paper presents a theoretical analysis of bimodal contrastive learning in the presence of misalignment (noise or missing information) in image-text pairs, revealing the critical role of feature purity in model generalization, and demonstrating that text recaptioning and filtering can significantly enhance feature purity and zero-shot classification performance.
Contrastive Representations for Temporal Reasoning
Alicja Ziarko (University of Warsaw), Piotr Miłoś (University of Warsaw)
Representation LearningContrastive LearningTime Series
🎯 What it does: This paper proposes the CRTR method, which generates representations that ignore context and focus on temporal structure in time series through contrastive learning, in order to enhance the performance of combinatorial reasoning tasks.
Contrastive Self-Supervised Learning As Neural Manifold Packing
Guanming Zhang (New York University), Stefano Martiniani (New York University)
ClassificationObject DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes the CLAMP framework, which reinterprets contrastive self-supervised learning as a neural manifold packing problem, and constructs a unidirectional loss function through short-range repulsive potential, dynamically optimizing the size and position of the enhancement sub-manifold for each image.
Contribution of task-irrelevant stimuli to drift of neural representations
Farhad Pashakhanloo (Center for Brain Science Harvard University)
Auto EncoderImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: In an online learning setting, we analyze how task-irrelevant stimuli lead to long-term drift in neural representations, with theoretical and simulation validation across various networks and learning rules.
ControlFusion: A Controllable Image Fusion Network with Language-Vision Degradation Prompts
Linfeng Tang (Wuhan University), Jiayi Ma (Macau University of Science and Technology)
RestorationObject DetectionTransformerPrompt EngineeringImageMultimodality
🎯 What it does: A controllable image fusion network called ControlFusion is proposed, which utilizes language-visual prompts to achieve adaptive denoising and fusion for composite degradation.
Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration
Seungyeon Choi (Yonsei University), Sanghyun Park (Yonsei University)
Drug DiscoveryScore-based ModelFlow-based ModelBiomedical Data
🎯 What it does: A generative framework named CBYG has been constructed, which can maintain high binding affinity while further controlling the synthetic feasibility and selectivity of molecules, thus aligning more closely with the actual needs of drug development.
Controllable Human-centric Keyframe Interpolation with Generative Prior
Zujin Guo (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
GenerationData SynthesisPose EstimationDiffusion modelVideoBenchmark
🎯 What it does: This paper proposes the PoseFuse3D-KI framework, achieving controllable human keyframe interpolation.
Controlled Visual Hallucination via Thalamus-Driven Decoupling Network for Domain Adaptation of Black-Box Predictors
Yuwu Lu (South China Normal University), Chunzhi Liu (South China Normal University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A domain adaptation method for black-box predictors, CVH-TDN, is proposed, which utilizes thalamus-driven illusion generation, alignment, and calibration to control illusions in the feature space, thereby enhancing target domain performance.
Controlling the Flow: Stability and Convergence for Stochastic Gradient Descent with Decaying Regularization
Sebastian Kassing (Technical University of Berlin), Leif Döring (University of Mannheim)
RestorationOptimizationImageComputed TomographyOrdinary Differential Equation
🎯 What it does: This paper studies the use of time-decaying Tikhonov regularization in the stochastic gradient descent (reg-SGD) algorithm on convex smooth objective functions. It proves that reg-SGD can almost surely and in the L2 sense converge to the minimum norm solution of the original problem without the assumption of boundedness of the iterative sequence. It also provides the convergence rates for polynomial step sizes and regularization parameters, and validates the theoretical predictions through experiments in image reconstruction and ODE inverse problems.
Controlling The Spread of Epidemics on Networks with Differential Privacy
Dung Nguyen (Haverford College), Jiayi Wu (University of Maryland)
OptimizationSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: A vaccination strategy based on edge differential privacy is designed to solve the optimal scheme for deleting nodes in a graph to reduce the maximum degree and spectral radius;
Controlling Thinking Speed in Reasoning Models
Zhengkai Lin (Zhejiang University), Jieping Ye (Alibaba Cloud)
Large Language ModelPrompt EngineeringText
🎯 What it does: A method is proposed for dynamically controlling the thinking speed of large reasoning models (LRM) during inference, utilizing the transition vectors of the model's internal representations to switch between different thinking modes (fast intuitive and slow deep) and dynamically adjusting the reasoning speed through real-time difficulty estimation.
Convergence of Clipped SGD on Convex $(L_0,L_1)$-Smooth Functions
Ofir Gaash (Tel Aviv University), Yair Carmon (Tel Aviv University)
OptimizationTabular
🎯 What it does: This paper studies the convergence of stochastic gradient descent (SGD) with gradient clipping under convex (L, L0,1)-smooth functions.
Convergence of the Gradient Flow for Shallow ReLU Networks on Weakly Interacting Data
Léo Dana (INRIA), Francis Bach (INRIA)
OptimizationTabular
🎯 What it does: Analyzed the convergence of a ReLU network with one hidden layer on weakly correlated data, proving that in high-dimensional space, with a network width of log(n), global convergence can be achieved with high probability.
Convergence Rates for Gradient Descent on the Edge of Stability for Overparametrised Least Squares
Lachlan Ewen MacDonald (University of Pennsylvania), Rene Vidal
Optimization
🎯 What it does: In the over-parameterized least squares problem, the convergence behavior of large step gradient descent is studied, and the convergence rates under three learning rate intervals (subcritical, critical, and supercritical) are provided.
Convergence Rates of Constrained Expected Improvement
Haowei Wang (National University of Singapore), Cosmin G. Petra (Lawrence Livermore National Laboratory)
OptimizationTabular
🎯 What it does: This paper studies the convergence rate of the Constrained Expected Improvement (CEI) algorithm, establishing for the first time an upper bound on its simple regret and analyzing its convergence under both frequentist and Bayesian assumptions.
Convergence Theorems for Entropy-Regularized and Distributional Reinforcement Learning
Yash Jhaveri (Rutgers University Newark), David Meger (Mila Quebec AI Institute)
Reinforcement LearningTabular
🎯 What it does: This paper studies the convergence properties of entropy-regularized reinforcement learning (ERL) as the temperature approaches zero, proposing the temperature decoupling gambit. It proves that this strategy can converge to an interpretable and diverse optimal policy as the temperature vanishes, and further extends this idea to distributed reinforcement learning (DRL), providing a convergent iterative method for returning distribution estimates.
Convergent Functions, Divergent Forms
Hyeonseong Jeon (University of Washington), Ranjay Krishna (University of Washington)
Robotic IntelligenceTransformerReinforcement Learning
🎯 What it does: A co-design framework named LOKI is proposed, which discovers diverse and high-performance robotic morphologies through shared multi-design control strategies and dynamic local search.
Convex Approximation of Two-Layer ReLU Networks for Hidden State Differential Privacy
Rob Romijnders (University of Amsterdam), Antti Koskela (Nokia Bell Labs)
OptimizationSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: The paper proposes a convex approximation method that transforms the training problem of a two-layer ReLU network into a strongly convex optimization problem, achieving feasible private training under a hidden state differential privacy model.
Convex Potential Mirror Langevin Algorithm for Efficient Sampling of Energy-Based Models
Yang Zitao, Jun Li (Fudan University)
GenerationData SynthesisComputational EfficiencyFlow-based ModelImageStochastic Differential Equation
🎯 What it does: A Convex Potential Mirror Langevin Algorithm (CPMLA) is proposed for efficient sampling of energy-based models (EBM).
Convolution Goes Higher-Order: A Biologically Inspired Mechanism Empowers Image Classification
Simone Azeglio (Institut de la Vision & Laboratoire des Systèmes Perceptifs INSERM, CNRS, Sorbonne University & École Normale Supérieure - PSL), Ulisse Ferrari (Institut de la Vision)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A learnable high-order convolution (HoConv) is proposed, which captures multiplicative interactions between pixels through Volterra expansion to enhance the feature representation capability of convolutional neural networks.
COOPERA: Continual Open-Ended Human-Robot Assistance
Chenyang Ma (University of Oxford), Niki Trigoni (University of Oxford)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: The COOPERA framework is proposed to enable robots to learn and adapt to individual human characteristics, habits, and temporal contexts during prolonged open-ended tasks, providing personalized assistance.
Cooperative Bargaining Games Without Utilities: Mediated Solutions from Direction Oracles
Kushagra Gupta (University of Texas at Austin), David Fridovich-Keil (University of Texas at Austin)
OptimizationTabularTime SeriesFinance Related
🎯 What it does: The study proposes and analyzes a new negotiation algorithm DiBS that only uses a direction oracle for cooperative negotiation problems where proxy utility values cannot be obtained, and only the optimal direction for each proxy can be obtained.