ICLR 2025 Papers — Page 7
International Conference on Learning Representations · 3704 papers
CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution
Yunju Cho (Seoul National University), Jay-Yoon Lee (Seoul National University)
OptimizationTransformerTime Series
🎯 What it does: Proposes the CoMRes model, which utilizes multi-scale self-supervised learning and consistency to enhance long-term time series forecasting.
Concept Bottleneck Language Models For Protein Design
Aya Abdelsalam Ismail (Genentech), Nathan C. Frey (Genentech)
GenerationDrug DiscoveryTransformerLarge Language ModelBiomedical Data
🎯 What it does: This paper proposes the Concept Bottleneck Protein Language Model (CB-pLM), which embeds an interpretable concept bottleneck layer in a large-scale protein masked language model to achieve controllable, interpretable, and debuggable protein generation.
Concept Bottleneck Large Language Models
Chung-En Sun (University of California San Diego), Tsui-Wei Weng (University of California San Diego)
ClassificationGenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes Concept Bottleneck Large Language Models (CB-LLMs), which achieve self-explanatory text classification and generation models by incorporating an interpretable concept bottleneck layer into pre-trained LLMs.
Concept Pinpoint Eraser for Text-to-image Diffusion Models via Residual Attention Gate
Byung Hyun Lee (Seoul National University), Se Young Chun (Seoul National University)
GenerationData SynthesisAdversarial AttackTransformerDiffusion modelImageText
🎯 What it does: A concept elimination method (CPE) is proposed in text-to-image diffusion models, which can accurately remove target concepts while retaining diverse remaining concepts.
Concept-ROT: Poisoning Concepts in Large Language Models with Model Editing
Keltin Grimes (Carnegie Mellon University), Marissa Catherine Connor
Adversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the Concept-ROT method, which utilizes Rank-One Model Editing to inject concept-triggered Trojans in large language models, achieving precise, low-sample attacks on safety-tuned models.
ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning
Ruchika Chavhan (University of Edinburgh), Timothy Hospedales (Samsung AI Center)
GenerationData SynthesisAdversarial AttackConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A training-independent pruning method called ConceptPrune is proposed for concept editing in diffusion models, which directly removes concepts by identifying and pruning neurons that lead to undesirable concept generation.
ConcreTizer: Model Inversion Attack via Occupancy Classification and Dispersion Control for 3D Point Cloud Restoration
Youngseok Kim (Seoul National University), Saewoong Bahk (Seoul National University)
RestorationAutonomous DrivingAdversarial AttackPoint Cloud
🎯 What it does: This paper proposes a model inversion attack method called ConcreTizer for voxelized 3D point cloud features in autonomous driving scenarios, which can recover the original 3D point cloud scene from intermediate features.
CONDA: Adaptive Concept Bottleneck for Foundation Models Under Distribution Shifts
Jihye Choi (University of Wisconsin Madison), Somesh Jha (University of Wisconsin Madison)
ClassificationDomain AdaptationExplainability and InterpretabilityTransformerContrastive LearningImageBiomedical Data
🎯 What it does: This paper presents CONDA, an adaptive framework based on Concept Bottleneck (CBM) designed to enhance the robustness and accuracy of interpretable classifiers based on large foundation models (FM) in distribution shift environments during deployment (testing).
Conditional Diffusion Models are Minimax-Optimal and Manifold-Adaptive for Conditional Distribution Estimation
Rong Tang (Hong Kong University of Science and Technology), Yun Yang (University of Maryland)
Diffusion model
🎯 What it does: This paper studies the theoretical properties of conditional forward-backward diffusion models in distribution regression, providing error convergence rates in Euclidean space and low-dimensional manifold scenarios.
Conditional Diffusion with Ordinal Regression: Longitudinal Data Generation for Neurodegenerative Disease Studies
Hyuna Cho (POSTECH), Won Hwa Kim (University of North Carolina at Chapel Hill)
GenerationData SynthesisDiffusion modelTime SeriesBiomedical DataMagnetic Resonance ImagingComputed TomographyAlzheimer's Disease
🎯 What it does: This paper proposes a model called ConDOR, based on conditional diffusion and ordinal regression, for generating longitudinal biomarker sequences of neurodegenerative diseases under given ordinal conditions such as age and disease grading.
Conditional Testing based on Localized Conformal $p$-values
Xiaoyang Wu (Nankai University), Changliang Zou (Nankai University)
Anomaly DetectionTabular
🎯 What it does: This paper proposes a conditional testing method based on localized conformal p-values and applies it to conditional outlier detection, conditional label selection, and two-sample conditional distribution testing.
Confidence Elicitation: A New Attack Vector for Large Language Models
Brian Formento (National University of Singapore), See-Kiong Ng (National University of Singapore)
Adversarial AttackTransformerLarge Language ModelText
🎯 What it does: This work proposes to guide black-box word replacement attacks by requesting self-confidence from large language models, thereby improving the success rate of adversarial attacks while maintaining semantic similarity.
ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks
Qiang Liu (Technical University of Munich), Nils Thuerey (Technical University of Munich)
OptimizationComputational EfficiencyTabularTime SeriesPhysics Related
🎯 What it does: This paper proposes the ConFIG method and its momentum-accelerated variant M-ConFIG to address gradient conflicts between different loss terms in the training of Physics-Informed Neural Networks (PINNs). It provides convergence proofs and validates its effectiveness on various PDEs and the CelebA multi-task learning.
Conflict-Averse Gradient Aggregation for Constrained Multi-Objective Reinforcement Learning
Dohyeong Kim (Seoul National University), Songhwai Oh (Seoul National University)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A new multi-objective reinforcement learning algorithm CoMOGA is proposed, which can handle multi-objective tasks while satisfying safety constraints and converging to the constrained Pareto front.
Conformal Generative Modeling with Improved Sample Efficiency through Sequential Greedy Filtering
Klaus-Rudolf Kladny (Max Planck Institute for Intelligent Systems), Michael Muehlebach (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisTextBiomedical Data
🎯 What it does: A sequential conformal prediction method called SCOPE‑Gen is proposed for controlling the output set of generative models; it constructs the prediction set by first generating i.i.d. samples and then gradually applying greedy filters (quality filtering and diversity filtering), achieving statistical guarantees on overall admissibility.
Conformal Language Model Reasoning with Coherent Factuality
Maxon Rubin-Toles (University of Pennsylvania), Surbhi Goel (University of Pennsylvania)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: For reasoning tasks, a method is proposed to ensure 'Coherent Factuality' in the argument steps generated by language models, along with a filtering algorithm based on segmented conformal prediction.
Conformal Prediction Sets Can Cause Disparate Impact
Jesse C. Cresswell (Layer 6 AI), Mouloud Belbahri (Layer 6 AI)
ClassificationLarge Language ModelImageTextAudio
🎯 What it does: This study evaluates the impact of conformal prediction sets used in decision support on the fairness of different protected groups through human experiments, quantifying disparate impact.
Conformal Structured Prediction
Botong Zhang (University of Pennsylvania), Osbert Bastani (University of Pennsylvania)
ClassificationOptimizationImageSequential
🎯 What it does: A general conformal structured prediction framework is proposed, capable of constructing prediction sets with coverage guarantees in any structured label space (such as DAGs, trees, etc.);
Conformalized Interactive Imitation Learning: Handling Expert Shift and Intermittent Feedback
Michelle D Zhao, Andrea Bajcsy (Carnegie Mellon University)
OptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningTabularTime SeriesFinance Related
🎯 What it does: This paper proposes ConformalDAgger, a framework that combines online synthetic prediction and interactive imitation learning;
Conformalized Survival Analysis for General Right-Censored Data
Hen Davidov (Technion Israel Institute of Technology), Yaniv Romano (Technion Israel Institute of Technology)
Biomedical Data
🎯 What it does: A distribution-free finite sample low prediction bound (LPB) calibration method for general right-censored survival analysis is proposed, including focused calibration and fused calibration;
CONGO: Compressive Online Gradient Optimization
Jeremy Carleton (Texas A and M University), Aditya Akella (University of Texas at Austin)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a sublinear scheduling algorithm (CONGO) that utilizes compressed sensing techniques for gradient estimation in zero-order online convex optimization (OCO) problems with sparse gradients, achieving sample efficiency and dimension-independent performance.
ConMix: Contrastive Mixup at Representation Level for Long-tailed Deep Clustering
Zhixin Li (Southeast University), Yuheng Jia (Southeast University)
Representation LearningContrastive LearningImage
🎯 What it does: A contrastive learning level mixing method for long-tail deep clustering, ConMix, is proposed.
Connecting Federated ADMM to Bayes
Siddharth Swaroop (Harvard University), Finale Doshi-Velez (Harvard University)
OptimizationFederated LearningComputational EfficiencyImageTabular
🎯 What it does: This study explores the connection between ADMM and Variational Bayes (VB) in federated learning, and based on this, proposes three new algorithms: FedLap, FedLap-Cov, and FedLap-Func, which enhance model convergence speed and communication efficiency.
Connectome Mapping: Shape-Memory Network via Interpretation of Contextual Semantic Information
Kyungsu Lee (Jeonbuk National University), Jae Youn Hwang (Daegu Gyeongbuk Institute of Science and Technology)
SegmentationAutonomous DrivingConvolutional Neural NetworkTransformerImageVideo
🎯 What it does: A Shape Memory Network (SMN) is proposed, which achieves more accurate predictions in semantic segmentation tasks by dynamically adapting the structure to contextual semantic information.
Conservative Contextual Bandits: Beyond Linear Representations
Rohan Deb (University of Illinois), Arindam Banerjee (University of Illinois)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes the Conservative Contextual Bandits (CCBs) algorithm, which aims to ensure safety in sequential decision-making, requiring the agent's strategy to minimize regret while satisfying safety constraints, meaning its performance should not be worse than the baseline strategy (such as the company's existing strategy) by more than (1 + α) times.
Consistency Checks for Language Model Forecasters
Daniel Paleka (ETH Zurich), Florian Tramèr (ETH Zurich)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes a method to instantaneously evaluate language model predictors (LLM forecasters) using consistency checks (such as negation, synonyms, consequences, etc.), constructs an automated evaluation pipeline, and generates a long-term consistency benchmark.
Consistency Models Made Easy
Zhengyang Geng (Carnegie Mellon University), J Zico Kolter
GenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelImageStochastic Differential Equation
🎯 What it does: A two-stage training framework called Easy Consistency Tuning (ECT) is proposed to fine-tune pre-trained diffusion models into consistency models, achieving faster training and better generation quality.
Consistent Flow Distillation for Text-to-3D Generation
Runjie Yan (University of California San Diego), Xiaolong Wang (University of California San Diego)
GenerationData SynthesisKnowledge DistillationDiffusion modelNeural Radiance FieldTextMeshStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Utilizing the probabilistic flow gradient of a two-dimensional diffusion model for consistent flow distillation of 3D representations, achieving high-quality and diverse 3D generation from text.
Constraint-Conditioned Actor-Critic for Offline Safe Reinforcement Learning
Zijian Guo (Boston University), Wenchao Li (Boston University)
Safty and PrivacyReinforcement LearningTabular
🎯 What it does: The Constraint-Conditioned Actor-Critic (CCAC) method is proposed to learn adaptive and robust policies under different cost thresholds in offline safe reinforcement learning, modeling the state-action distribution through constraint conditions to address the safety-performance imbalance caused by out-of-distribution (OOD) state-action.
Constructing Confidence Intervals for Average Treatment Effects from Multiple Datasets
Yuxin Wang (Ludwig Maximilian University of Munich), Stefan Feuerriegel (Ludwig Maximilian University of Munich)
Biomedical DataElectronic Health Records
🎯 What it does: A method based on Predictive Enhanced Inference (PPI) is proposed to construct confidence intervals for the Average Treatment Effect (ATE) from two observational datasets (one small and unbiased, the other large and potentially subject to unobserved confounding).
Content-Style Learning from Unaligned Domains: Identifiability under Unknown Latent Dimensions
Sagar Shrestha (Oregon State University), Xiao Fu (Oregon State University)
GenerationData SynthesisDomain AdaptationTransformerGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a content-style learning framework based on cross-domain latent distribution matching (LDM), addressing the issue of how to identify content and style representations from mixed nonlinear models in unaligned multi-domain data without prior knowledge of the latent dimensions.
Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data
Michael Wornow (Stanford University), Nigam Shah
TransformerLarge Language ModelTabularBiomedical DataElectronic Health RecordsBenchmark
🎯 What it does: This study investigates the impact of different context lengths on clinical prediction tasks using electronic health records (EHR) and systematically evaluates the effects of three EHR-specific attributes (copy-forwarding, irregular time intervals, disease progression) on model performance.
Context Steering: Controllable Personalization at Inference Time
Jerry Zhi-Yang He (University of California Berkeley), Anca Dragan (University of California Berkeley)
GenerationRecommendation SystemTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes Context Steering (CoS), which controllably amplifies or weakens the influence of context on generation results during inference by comparing the probability differences of outputs from a large language model (LLM) with and without context, achieving adjustable personalized text generation.
Context-Alignment: Activating and Enhancing LLMs Capabilities in Time Series
Yuxiao Hu (Hong Kong Polytechnic University), Yuntian Chen (Ningbo Institute of Digital Twin)
ClassificationGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTime Series
🎯 What it does: Proposes the Context-Alignment concept, utilizing a Dual-Scale GNN to achieve structural and logical alignment between time series and language prompts, activating and enhancing the performance of large language models in temporal tasks.
Context-aware Dynamic Pruning for Speech Foundation Models
Masao Someki (Carnegie Mellon University), Shinji Watanabe (Amazon)
RecognitionComputational EfficiencyTransformerSupervised Fine-TuningAudio
🎯 What it does: A context-based dynamic pruning method is proposed, which dynamically prunes large speech foundation models during the inference phase based on speaker, language, and task information, significantly improving inference speed.
Context-Parametric Inversion: Why Instruction Finetuning May Not Actually Improve Context Reliance
Sachin Goyal (Carnegie Mellon University), Aditi Raghunathan (Carnegie Mellon University)
TransformerSupervised Fine-TuningText
🎯 What it does: This study investigates the phenomenon of context-dependent regression under knowledge conflict during instruction fine-tuning.
ContextGNN: Beyond Two-Tower Recommendation Systems
Yiwen Yuan (Kumo.AI), Matthias Fey (Kumo.AI)
Recommendation SystemGraph Neural NetworkTime Series
🎯 What it does: A Context-based Graph Neural Network that integrates pair-wise and two-tower representations is proposed for time series recommendation.
Contextual Document Embeddings
John Xavier Morris, Alexander M Rush
RetrievalRepresentation LearningTransformerContrastive LearningText
🎯 What it does: This paper proposes Contextual Document Embedding (CDE), which improves document representation for dense retrieval by incorporating information from neighboring documents in contrastive learning and encoder architecture.
Contextual Self-paced Learning for Weakly Supervised Spatio-Temporal Video Grounding
Akash Kumar (University of Central Florida), Yogesh S Rawat
Object DetectionObject TrackingTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper studies the weakly supervised spatiotemporal video localization (WSTVG) task, which aims to achieve object localization in space and time based solely on video-level textual descriptions.
Contextualizing biological perturbation experiments through language
Menghua Wu (Massachusetts Institute of Technology), Jan-Christian Huetter (Genentech)
TransformerLarge Language ModelBiomedical DataBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the PERTURBQA benchmark to evaluate the ability of language models in reasoning about biological perturbation experimental results, and constructs the SUMMER framework to combine knowledge graphs with experimental data for question-answer reasoning.
Continual Slow-and-Fast Adaptation of Latent Neural Dynamics (CoSFan): Meta-Learning What-How & When to Adapt
Ryan Missel (Rochester Institute of Technology), Linwei Wang (Rochester Institute of Technology)
Meta LearningAuto EncoderTime SeriesOrdinary Differential Equation
🎯 What it does: This paper proposes CoSFan, which combines fast 'what-how' inference through continual meta-learning with slow 'when' adaptation to achieve continual learning in high-dimensional time series prediction with unknown task boundaries and identities.
Continuity-Preserving Convolutional Autoencoders for Learning Continuous Latent Dynamical Models from Images
Aiqing Zhu (National University of Singapore), Qianxiao Li (National University of Singapore)
Convolutional Neural NetworkAuto EncoderImageOrdinary Differential Equation
🎯 What it does: This paper proposes Continuous Potential-Aware Convolutional Autoencoders (CpAEs), which ensure that the latent states change continuously over time while learning continuous latent dynamics from discrete image frames.
Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis
Weiwei Lin (Hong Kong Polytechnic University), Chenhang HE
GenerationData SynthesisTransformerAuto EncoderMultimodalityAudio
🎯 What it does: A self-regressive speech synthesis framework based on continuous Gaussian mixture VAE and GMM language model is proposed.
Continuous Diffusion for Mixed-Type Tabular Data
Markus Mueller (Erasmus University Rotterdam), Dennis Fok (Erasmus University Rotterdam)
GenerationData SynthesisDiffusion modelScore-based ModelTabular
🎯 What it does: This paper proposes CDTD, a continuous diffusion model for mixed tabular data, which uniformly uses Gaussian noise to diffuse continuous features and embedded categorical features, combining score matching and score interpolation.
Continuous Ensemble Weather Forecasting with Diffusion models
Martin Andrae (Linkoping University), Fredrik Lindsten (Linkoping University)
Diffusion modelScore-based ModelTime SeriesOrdinary Differential Equation
🎯 What it does: A continuous ensemble forecasting framework based on diffusion models is proposed, which directly models the distribution of arbitrary time delays in the noise space of the diffusion process and generates time-continuous trajectories using correlated noise; at the same time, it combines autoregressive continuous interpolation (ARCI) to achieve compatibility between long time delay forecasting and high temporal resolution.
Continuous Exposure Learning for Low-light Image Enhancement using Neural ODEs
Donggoo Jung (Hanyang University), Tae Hyun Kim (Hanyang University)
RestorationImageOrdinary Differential Equation
🎯 What it does: A continuous exposure learning method CLODE based on Neural Ordinary Differential Equations (NODE) is proposed, improving the convergence and effectiveness of traditional discrete iterative curve adjustment in unsupervised low-light image enhancement.
CONTRA: Conformal Prediction Region via Normalizing Flow Transformation
Zhenhan Fang (University of Iowa), Jian Huang (Hong Kong Polytechnic University)
Flow-based ModelTabular
🎯 What it does: A multidimensional confidence prediction method based on regularized flows, called CONTRA (and its extension ResCONTRA for any prediction model), is proposed to achieve coverage guarantees and generate compact, smooth prediction regions.
Contractive Dynamical Imitation Policies for Efficient Out-of-Sample Recovery
Amin Abyaneh (McGill University), Giancarlo Ferrari-Trecate (École Polytechnique Fédérale de Lausanne)
OptimizationRobotic IntelligenceReinforcement LearningTime SeriesSequentialOrdinary Differential Equation
🎯 What it does: A framework for imitation learning based on contract dynamical systems is proposed, utilizing Recursive Equilibrium Networks (REN) and reversible coupling layers to train policies using only state information, and providing an out-of-sample (OOS) error upper bound;
ContraDiff: Planning Towards High Return States via Contrastive Learning
Yixiang Shan (Jilin University), Liang Yin (Shanghai Jiao Tong University)
Robotic IntelligenceReinforcement LearningDiffusion modelContrastive LearningTabular
🎯 What it does: The ContraDiff method is proposed, which utilizes contrastive learning to treat low-reward trajectories from offline data as negative samples and high-reward trajectories as positive samples. By applying contrastive constraints to the states of generated trajectories, it enhances the performance of offline RL when high-reward samples are scarce.
Contrastive Learning from Synthetic Audio Doppelgängers
Manuel Cherep (Massachusetts Institute of Technology), Nikhil Singh (Dartmouth College)
ClassificationData SynthesisRepresentation LearningConvolutional Neural NetworkContrastive LearningAudio
🎯 What it does: By applying Gaussian noise perturbations to the parameters of a sound synthesizer, similar but not identical audio pairs (audio doppelgangers) are generated and used as positive samples for contrastive learning, thereby obtaining robust audio representations.
Control-oriented Clustering of Visual Latent Representation
Han Qi (Harvard University), Heng Yang (Harvard University)
Robotic IntelligenceReinforcement Learning from Human FeedbackConvolutional Neural NetworkRecurrent Neural NetworkDiffusion modelImage
🎯 What it does: In the behavior cloning-based visual control pipeline, we explore whether a clustering phenomenon similar to Neural Collapse in image classification appears in the visual latent space, and utilize this clustering pattern to pre-train the visual encoder, thereby enhancing test performance in both discrete and continuous control tasks.
ControlAR: Controllable Image Generation with Autoregressive Models
Zongming Li (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
GenerationData SynthesisTransformerSupervised Fine-TuningImage
🎯 What it does: A controllable autoregressive image generation framework called ControlAR is proposed, which embeds spatial control information into the autoregressive model through conditional decoding, achieving high-quality image generation at arbitrary resolutions.
Controllable Blur Data Augmentation Using 3D-Aware Motion Estimation
Insoo Kim (Samsung Electronics), Jinwoo Shin (KAIST)
RestorationData SynthesisConvolutional Neural NetworkOptical FlowImage
🎯 What it does: A 3D perception controllable blur synthesizer is proposed to generate diverse realistic blurred images to enhance the generalization ability of deblurring models.
Controllable Context Sensitivity and the Knob Behind It
Julian Minder (ETH Zürich), Ryan Cotterell (Cornell University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper designs a controllable context sensitivity task (CCS) and enables language models to provide different answers under the same context and query pair by fine-tuning or few-shot learning, according to the instruction to 'consider context' or 'ignore context'.
Controllable Generation via Locally Constrained Resampling
Kareem Ahmed (University of California), Guy Van den Broeck (University of California)
GenerationLarge Language ModelText
🎯 What it does: This paper proposes a method based on Locally Constrained Resampling (GEN-C) that can enforce given logical constraints during the generation of autoregressive language models, ensuring that the generated text adheres to the constraints while maintaining an approximation of the original model distribution.
Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements
Jingyu Zhang (Johns Hopkins University), Benjamin Van Durme (Johns Hopkins University)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: A framework called CoSA is proposed, which can dynamically control the safety behavior of large language models through secure configurations during inference, and implements a data-driven CoSAlign method.
Controllable Satellite-to-Street-View Synthesis with Precise Pose Alignment and Zero-Shot Environmental Control
Xianghui Ze (Nanjing University of Science and Technology), Yujiao Shi (ShanghaiTech University)
Image TranslationGenerationData SynthesisPose EstimationDiffusion modelContrastive LearningImage
🎯 What it does: Generate ground street view images from satellite images, achieving precise pose alignment and diverse environmental control.
Controllable Unlearning for Image-to-Image Generative Models via $\epsilon$-Constrained Optimization
XiaoHua Feng, Xiaolin Zheng (Ant Group)
GenerationOptimizationImage
🎯 What it does: A controllable image-to-image (I2I) generation model 'unlearning' framework is proposed, achieving a balance between the forgetting set and the retaining set through ε-constraint optimization.
Controlled LLM Decoding via Discrete Auto-regressive Biasing
Patrick Pynadath (Purdue University), Ruqi Zhang (Purdue University)
GenerationTransformerLarge Language ModelText
🎯 What it does: A controlled decoding algorithm DAB based on discrete gradient sampling is proposed to generate fluent text while satisfying external constraints.
Controlling Language and Diffusion Models by Transporting Activations
Pau Rodriguez, Xavier Suau (Apple)
GenerationData SynthesisOptimizationTransformerLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes the Activation Transport (ACT) framework, which achieves fine-grained controllability of large generative models (LLM and T2I) by optimally transporting activation values during inference.
Controlling Space and Time with Diffusion Models
Daniel Watson (Google DeepMind), David J. Fleet (Google DeepMind)
GenerationData SynthesisPose EstimationSuper ResolutionDiffusion modelImageVideo
🎯 What it does: We propose 4DiM, a diffusion model that can control space and time from a small number of input images or videos, achieving 4D novel view synthesis.
ConvCodeWorld: Benchmarking Conversational Code Generation in Reproducible Feedback Environments
Hojae Han (Snowflake AI Research), Yuxiong He (Snowflake AI Research)
GenerationAI Code AssistantTransformerLarge Language ModelTextBenchmark
🎯 What it does: A reproducible interactive code generation benchmark, CONVCODEWORLD, and its low-cost static version, CONVCODEBENCH, are proposed, and the performance of LLMs is evaluated under nine combinations of multi-turn feedback.
Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification
Hyunji Jung (POSTECH), Chulhee Yun (KAIST)
ClassificationOptimizationTabular
🎯 What it does: This study investigates continuous linear classification tasks, using Stochastic Gradient Descent (GD) to learn multiple tasks under a limited iteration budget, and proves its convergence to the joint maximum margin solution.
Convergence of Distributed Adaptive Optimization with Local Updates
Ziheng Cheng (University of California), Margalit Glasgow (Massachusetts Institute of Technology)
Optimization
🎯 What it does: This paper studies the use of adaptive optimization algorithms with local updates (intermittent communication) in a distributed environment, proving that Local SGDM and Local Adam can significantly reduce communication complexity and achieve better convergence rates in convex and weakly convex settings.
Convergence of Score-Based Discrete Diffusion Models: A Discrete-Time Analysis
Zikun Zhang (Fudan University), Quanquan Gu (University of California, Los Angeles)
Diffusion modelScore-based ModelImageText
🎯 What it does: A discrete-time sampling algorithm under the CTMC framework is proposed, and its convergence analysis is provided;
Convergent Privacy Loss of Noisy-SGD without Convexity and Smoothness
Eli Chien (Georgia Institute of Technology), Pan Li (Georgia Institute of Technology)
OptimizationSafty and Privacy
🎯 What it does: Theoretical research on the privacy analysis of the hidden state of Noisy-SGD (DP-SGD) is conducted, proving that a convergent Rényi DP upper bound can still be obtained under non-convexity and non-smoothness constraints.
Convex Formulations for Training Two-Layer ReLU Neural Networks
Karthik Prakhya (Umeå University), Alp Yurtsever (Imperial College London)
OptimizationTabularSequential
🎯 What it does: Transform the training problem of two-layer ReLU networks into a convex copositive program (completely positive program), and based on this, propose a solvable semidefinite relaxation and rounding method;
COPER: Correlation-based Permutations for Multi-View Clustering
Ran Eisenberg (Bar Ilan University), Ofir Lindenbaum (Bar Ilan University)
Representation LearningAuto EncoderContrastive LearningMultimodalityBenchmark
🎯 What it does: An end-to-end multi-view clustering framework called COPER is proposed, which utilizes self-supervised view-internal label permutation combined with CCA to learn shared representations and perform clustering directly.
Copyright-Protected Language Generation via Adaptive Model Fusion
Javier Abad (ETH Zurich), Fanny Yang (ETH Zurich)
GenerationOptimizationTransformerLarge Language ModelText
🎯 What it does: A post-processing model fusion method named CP-Fuse is proposed to reduce the risk of language models generating copyrighted content during the inference phase by combining two models trained on non-overlapping copyrighted data.
Coreset Selection via Reducible Loss in Continual Learning
Ruilin Tong (University of New South Wales), Dong Gong (University of New South Wales)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a core subset selection method based on Reducible Loss (ReL) called CSReL, specifically designed to enhance the quality of sample memory in experience replay (ER) based continuous learning (CL). Additionally, three extension schemes are provided, considering task interference, streaming data, and knowledge distillation (CSReL-CL, CSReL-RS, CSReL-RS-KD).
Coreset Spectral Clustering
Ben Jourdan (University of Edinburgh), He Sun (University of St Andrews)
Graph Neural NetworkGraph
🎯 What it does: Construct ε-coresets on sparse graphs and perform spectral clustering directly on the core graph, then map the clustering results back to the original graph to complete the labeling, forming a core spectral clustering algorithm.
CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking
Tarun Suresh (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
RetrievalAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: A high-quality contrastive learning dataset called CORNSTACK was constructed, and this dataset was used to train a code retriever and re-ranker, significantly improving the performance of code retrieval and functionality localization.
Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization
Audrey Huang (University of Illinois Urbana-Champaign), Dylan J Foster
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningText
🎯 What it does: An offline language model alignment algorithm χ PO is proposed, which incorporates χ²-regularization into the existing Direct Preference Optimization (DPO) objective to achieve a pessimistic treatment of uncertainty, thereby addressing the issue of over-optimization and providing theoretical guarantees on sample complexity.
Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking
Cassidy Laidlaw (University of California), Anca Dragan (University of California)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabularTime Series
🎯 What it does: A new definition of reward hacking is proposed, and theoretical and empirical protection against reward hacking is achieved through χ² occupancy measure regularization.
Correlating instruction-tuning (in multimodal models) with vision-language processing (in the brain)
SUBBA REDDY OOTA, Manish Gupta (Microsoft)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper utilizes instruction-tuned multimodal large language models (MLLMs) to extract instruction-specific text output embeddings in natural image viewing tasks and predicts human fMRI brain activity using a linear encoding model, studying the effects of different instructions, levels, and visual concepts on brain encoding.
Correlation and Navigation in the Vocabulary Key Representation Space of Language Models
Letian Peng (University of California), Jingbo Shang (University of California)
GenerationKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper studies the impact of irrelevant correlations in the key space of the vocabulary on next word prediction and proposes an in-context navigation (ICN) method that iteratively pushes the query vector away from explored keys by incorporating previously explored answers into the context, thereby alleviating this correlation.
CoTFormer: A Chain of Thought Driven Architecture with Budget-Adaptive Computation Cost at Inference
Amirkeivan Mohtashami (École Polytechnique Fédérale de Lausanne), Martin Jaggi (École Polytechnique Fédérale de Lausanne)
TransformerTextChain-of-Thought
🎯 What it does: A new Transformer architecture called CoTFormer is proposed, along with a set of adaptive training methods based on token-level variable depth;
Counterfactual Concept Bottleneck Models
Gabriele Dominici (Università della Svizzera italiana), Marc Langheinrich (Università della Svizzera italiana)
Explainability and InterpretabilityAdversarial AttackAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A joint Generative Adversarial Concept Bottleneck Model (CF-CBM) is proposed, addressing three core issues: prediction, scenario simulation, and adversarial explanation.
Counterfactual Generative Modeling with Variational Causal Inference
Yulun Wu (University of California), Claudia Iriondo (Genentech)
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkImageBiomedical Data
🎯 What it does: A variational Bayesian causal inference framework is proposed for counterfactual generation of high-dimensional outcomes (such as gene expression and images), along with a robust marginal estimation method.
Counterfactual Realizability
Arvind Raghavan (Columbia University), Elias Bareinboim (Columbia University)
Reinforcement LearningPhysics Related
🎯 What it does: This study investigates which layer 3 (counterfactual) distributions in causal models can be directly sampled through physical experiments, proposing a definition of 'realizability' and providing a decision algorithm.
CPSample: Classifier Protected Sampling for Guarding Training Data During Diffusion
Joshua Kazdan (Stanford University), Stefano Ermon (UCLA)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: A CPSample method is proposed that incorporates classifier guidance during the sampling phase of the diffusion model, preventing the model from replicating training data during inference while maintaining image quality.
CR-CTC: Consistency regularization on CTC for improved speech recognition
Zengwei Yao (Xiaomi Corporation), Daniel Povey (Xiaomi Corporation)
RecognitionKnowledge DistillationRepresentation LearningTransformerContrastive LearningAudio
🎯 What it does: Proposes Consistency-Regularized CTC (CR-CTC), which utilizes the CTC distribution of two different augmented views of the same speech for consistency regularization to enhance the recognition performance of pure CTC.
CR2PQ: Continuous Relative Rotary Positional Query for Dense Visual Representation Learning
Shaofeng Zhang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Object DetectionSegmentationKnowledge DistillationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Continuous Relative Rotation Position Query (CR2PQ) framework to eliminate the need for pixel/patch correspondence in dense visual representation learning and achieve cross-view alignment through a relative coordinate system and continuous RoPE.
CraftRTL: High-quality Synthetic Data Generation for Verilog Code Models with Correct-by-Construction Non-Textual Representations and Targeted Code Repair
Mingjie Liu (NVIDIA Corporation), Haoxing Ren (NVIDIA Corporation)
Data SynthesisAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A high-quality synthesis data generation framework for Verilog code generation has been developed, which includes correctly constructed non-text representations (Karnaugh maps, FSMs, waveforms) and automated target code repair data, followed by fine-tuning on Starcoder2-15B.
CREAM: Consistency Regularized Self-Rewarding Language Models
Zhaoyang Wang (University of North Carolina at Chapel Hill), Huaxiu Yao (University of North Carolina at Chapel Hill)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper proposes a self-rewarding language model based on consistency regularization (CREAM), which reduces reward bias by utilizing reward consistency between different iterations during the self-rewarding process, thereby enhancing the alignment performance of a small 7B LLM.
Credal Wrapper of Model Averaging for Uncertainty Estimation in Classification
Kaizheng Wang (KU Leuven), Hans Hallez (KU Leuven)
ClassificationConvolutional Neural NetworkTransformerImage
🎯 What it does: A framework named Credal Wrapper is proposed, which constructs probability intervals by extracting upper and lower probability bounds from a limited single distribution, thereby generating credible sets and mapping them to cross probabilities, enhancing uncertainty estimation in classification tasks of BNN and DE.
Credit-based self organizing maps: training deep topographic networks with minimal performance degradation
Amirozhan Dehghani, Pouya Bashivan (McGill University)
OptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Introducing task credit-based self-organizing maps (CB-SOM) in deep convolutional networks allows the filters of all layers to form a topological structure while maintaining low task performance loss.
CREIMBO: Cross-Regional Ensemble Interactions in Multi-view Brain Observations
Noga Mudrik (Johns Hopkins University), Adam Shabti Charles
Biomedical DataMagnetic Resonance Imaging
🎯 What it does: The CREIMBO model is proposed to jointly learn the dynamics of cross-regional neural networks from asynchronous multi-session neural recordings, and to identify session-specific and shared neural ensembles and their time-varying interactions with the support of a global subcircuit dictionary.
CREMA: Generalizable and Efficient Video-Language Reasoning via Multimodal Modular Fusion
Shoubin Yu (University of North Carolina at Chapel Hill), Mohit Bansal (University of North Carolina at Chapel Hill)
RecognitionGenerationComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoMultimodalityPoint CloudAudio
🎯 What it does: The CREMA framework is proposed, which implements multimodal video language reasoning between the frozen Q-Former and LLM using pluggable lightweight modules (LoRA + queries). It can seamlessly incorporate various perceptual modalities such as video, optical flow, 3D point clouds, audio, heat maps, and touch, significantly enhancing reasoning performance.
Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion
Marco Mistretta (University of Florence), Andrew D. Bagdanov (University of Florence)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This study investigates the intra-modal mismatch problem present in single-modal tasks (image-image retrieval, text-text retrieval) using the CLIP pre-trained model. It transforms single-modal tasks into cross-modal tasks through single-feature-level modality inversion (OTI and OVI) to leverage the cross-modal alignment advantages of CLIP and improve retrieval performance.
Cross-Attention Head Position Patterns Can Align with Human Visual Concepts in Text-to-Image Generative Models
Jungwon Park (Seoul National University), Wonjong Rhee (Seoul National University)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes a Head Relevant Vector (HRV) method based on cross-attention heads to map human-specified visual concepts to the attention heads of stable diffusion models, enhancing and adjusting concepts in image generation, editing, and multi-concept generation tasks.
Cross-Domain Off-Policy Evaluation and Learning for Contextual Bandits
Yuta Natsubori (Hakuhodo DY Holdings), Yuta Saito (Cornell University)
Recommendation SystemReinforcement LearningTabular
🎯 What it does: This study investigates the problem of cross-domain offline policy evaluation and learning in the presence of deterministic logging strategies, new actions, and limited data.
Cross-Domain Offline Policy Adaptation with Optimal Transport and Dataset Constraint
Jiafei Lyu (Tsinghua University), Xiu Li (Tsinghua University)
Domain AdaptationReinforcement LearningTabular
🎯 What it does: Explores cross-domain offline reinforcement learning in the context of limited target domain data, using source domain data to assist in target domain policy learning.
Cross-Embodiment Dexterous Grasping with Reinforcement Learning
Haoqi Yuan (Peking University), Zongqing Lu (Peking University)
Robotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: A robot hand grasping strategy that can cross different fingertip structures (CrossDex) is proposed and implemented. It learns a unified, visual grasping strategy through reinforcement learning, enabling successful grasping on various robot hands.
Cross-Entropy Is All You Need To Invert the Data Generating Process
Patrik Reizinger (Max Planck Institute for Intelligent Systems), David Klindt (Cold Spring Harbor Laboratory)
ClassificationData SynthesisContrastive LearningImage
🎯 What it does: The study investigates whether models trained with cross-entropy classification can reverse the data generation process and provides a theory of identifiability.
Cross-Modal Safety Mechanism Transfer in Large Vision-Language Models
Shicheng Xu (Chinese Academy of Sciences), Xueqi Cheng (Chinese Academy of Sciences)
Safty and PrivacyTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents a new perspective called cross-modal safety mechanism transfer, aimed at addressing the vulnerabilities of large visual language models (LVLMs) when handling toxic visual inputs. By comparing the safety mechanisms of text and visuals, it was found that existing methods are unable to effectively transfer the safety mechanisms of text to visuals.
CrossMPT: Cross-attention Message-passing Transformer for Error Correcting Codes
Seong-Joon Park (POSTECH), Jong-Seon No (Seoul National University)
Transformer
🎯 What it does: A new Transformer architecture called CrossMPT is proposed for decoding error correction codes (ECC); this architecture updates magnitude and syndrome embeddings through cross-attention, forming an iterative update process similar to information transmission.
CryoFM: A Flow-based Foundation Model for Cryo-EM Densities
Yi Zhou (ByteDance Research), Quanquan Gu (ByteDance Research)
RestorationGenerationTransformerFlow-based ModelImageBiomedical DataOrdinary Differential Equation
🎯 What it does: This paper proposes CRYOFM, a foundational model based on flow matching, designed to learn the prior distribution of high-quality cryo-EM density maps, and achieves high-quality reconstruction without fine-tuning through flow posterior sampling in various inverse problem tasks.
CryoGEN: Generative Energy-based Models for Cryogenic Electron Tomography Reconstruction
Yunfei Teng (Beijing Academy of Artificial Intelligence), Qiwei Ye (Beijing Academy of Artificial Intelligence)
RestorationGenerationGenerative Adversarial NetworkImageComputed Tomography
🎯 What it does: We propose CryoGEN, a generative method based on an energy model, to address the anisotropic resolution problem caused by missing wedges in cryo-ET, achieving faster and more stable training.
cryoSPHERE: Single-Particle HEterogeneous REconstruction from cryo EM
Gabriel Ducrocq (Linkoping University), Fredrik Lindsten (Linkoping University)
Protein Structure PredictionAuto EncoderImage
🎯 What it does: A cryoSPHERE method based on variational autoencoders is proposed, which utilizes benchmark structures predicted by AlphaFold to learn how to segment protein chains into several segments and predict rigid transformations for each segment in each cryo-EM image, thereby recovering the continuous conformation of the protein at the single-particle level.