ICML 2025 Papers — Page 5
International Conference on Machine Learning · 3257 papers
Breaking the Curse of Multiagency in Robust Multi-Agent Reinforcement Learning
Laixi Shi (Johns Hopkins University), Adam Wierman (California Institute of Technology)
Reinforcement Learning
🎯 What it does: This paper proposes a class of robust Markov games based on behavioral economics, termed 'simulated uncertainty sets', and provides a proof of existence for its optimal solution concept. Subsequently, a sample-efficient algorithm, Robust-Q-FTRL, is designed to learn approximate robust coarse equilibria under generative models.
Breaking the Quadratic Barrier: Robust Cardinality Sketches for Adaptive Queries
Edith Cohen (Google Research), Uri Stemmer (School of Computer Science, Tel Aviv University)
🎯 What it does: This paper addresses the robustness issue of cardinality estimation sketches in adaptive query environments by designing two new robust estimators (Basic Robust Estimator and Tracking Robust Estimator), and proves that they can support exponential adaptive queries when the participation count r of each element in the query is approximately O(k^2).
BRIDGE: Bootstrapping Text to Control Time-Series Generation via Multi-Agent Iterative Optimization and Diffusion Modeling
Hao Li (Microsoft Research), Jiang Bian (Nanjing University)
GenerationOptimizationLarge Language ModelAgentic AIDiffusion modelTextTime SeriesFinance Related
🎯 What it does: A framework for text-controlled time series generation (BRIDGE) that combines multi-agent iterative optimization with diffusion models is proposed, and a large-scale text-time series paired dataset is constructed.
Bridging Fairness and Efficiency in Conformal Inference: A Surrogate-Assisted Group-Clustered Approach
Chenyin Gao (Harvard University), Larry Han (Northeastern University)
Biomedical Data
🎯 What it does: A SAGCCI framework that combines auxiliary variable assistance and group clustering is proposed to construct prediction intervals that are both fair and efficient.
Bridging Layout and RTL: Knowledge Distillation based Timing Prediction
Mingjun Wang (Institute of Computing Technology Chinese Academy of Sciences), Huawei Li (Institute of Computing Technology Chinese Academy of Sciences)
Knowledge DistillationGraph Neural NetworkGraph
🎯 What it does: The RTLDistil framework is proposed, which transfers layout-level physical features to RTL-level models through cross-stage knowledge distillation, achieving high-accuracy timing prediction.
Bridging Protein Sequences and Microscopy Images with Unified Diffusion Models
Dihan Zheng (University of California San Francisco), Bo Huang (University of California San Francisco)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImageBiomedical Data
🎯 What it does: A unified diffusion model CELL-Diff is proposed to achieve bidirectional generation of protein sequences and fluorescence microscopy images.
Bring Reason to Vision: Understanding Perception and Reasoning through Model Merging
Shiqi Chen (City University of Hong Kong), Junxian He (Hong Kong University of Science and Technology)
RecognitionGenerationOptimizationTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: This paper achieves an enhancement in the multimodal reasoning ability of VLM by performing a weighted average combination of the language module of LLM focused on mathematical reasoning and VLM, without the need for additional training.
BRiTE: Bootstrapping Reinforced Thinking Process to Enhance Language Model Reasoning
Han Zhong (Peking University), Zhaoran Wang (Northwestern University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: A unified probabilistic framework is proposed, based on which the BRiTE algorithm is designed to automatically generate high-quality thinking processes (CoT) using reinforcement learning, thereby enhancing the reasoning ability of large language models.
Broadband Ground Motion Synthesis by Diffusion Model with Minimal Condition
Jaeheun Jung (Korea University), Donghun Lee (Korea University)
GenerationData SynthesisDiffusion modelAuto EncoderTime Series
🎯 What it does: An end-to-end diffusion model named HEGGS is proposed to generate high-fidelity broadband seismic waveforms by minimizing seismic metadata.
BSemiFL: Semi-supervised Federated Learning via a Bayesian Approach
Haozhao Wang (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
Federated LearningImage
🎯 What it does: This paper proposes a semi-supervised federated learning framework called BSemiFL based on Bayesian methods, which aims to relabel local data in the absence of labeled data on the client side by combining weighted voting of the global model and local model to enhance the quality of pseudo-labels.
BSLoRA: Enhancing the Parameter Efficiency of LoRA with Intra-Layer and Inter-Layer Sharing
Yuhua Zhou (Zhejiang University), Aimin PAN
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes BSLoRA, which significantly reduces the number of trainable parameters by splitting LoRA into three sub-LoRAs: local, intra-layer shared, and inter-layer shared, combined with shape transformation techniques.
BSO: Binary Spiking Online Optimization Algorithm
Yu Liang (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
OptimizationSpiking Neural NetworkImageAudio
🎯 What it does: This paper proposes an online training algorithm for Binary Spiking Neural Networks (BSNN) - BSO and its time-aware variant T-BSO, significantly reducing training memory requirements.
Byzantine-Resilient Federated Alternating Gradient Descent and Minimization for Partly-Decoupled Low Rank Matrix Learning
Ankit Pratap Singh (Iowa State University), Namrata Vaswani (Iowa State University)
Recommendation SystemOptimizationFederated LearningTabular
🎯 What it does: This paper proposes a robust Byzantine-tolerant federated low-rank matrix learning algorithm that can efficiently recover low-rank matrices in the presence of node attacks.
C-3PO: Compact Plug-and-Play Proxy Optimization to Achieve Human-like Retrieval-Augmented Generation
Guoxin Chen (Renmin University of China), Kai Fan (Alibaba)
RetrievalOptimizationTransformerReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: C-3PO is proposed, a proxy center framework based on a lightweight multi-agent system, achieving unmodified alignment between the retriever and large language models (LLM), and simulating the interactive process of human search behavior.
C2IQL: Constraint-Conditioned Implicit Q-learning for Safe Offline Reinforcement Learning
Zifan LIU, Jun Zhang (Hong Kong University of Science and Technology)
Safty and PrivacyRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: An algorithm called C2IQL based on constrained implicit Q-learning is proposed in offline safe reinforcement learning to simultaneously maximize rewards while satisfying cost constraints, addressing the OOD problem and discount constraint errors of traditional methods.
Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing
Kaifeng Gao (Zhejiang University), Long Chen (Hong Kong University of Science and Technology)
GenerationComputational EfficiencyTransformerDiffusion modelVideoText
🎯 What it does: A self-regressive video diffusion model named Ca2-VDM is proposed, which significantly improves efficiency in generating long videos by utilizing causal generation and cache sharing.
CABS: Conflict-Aware and Balanced Sparsification for Enhancing Model Merging
zongzhen yang, Xiang Gao (Beihang University)
OptimizationTransformerLarge Language ModelText
🎯 What it does: A model fusion method based on Conflict-Aware and Balanced Sparsification (CABS) is proposed, significantly improving the performance of multi-task models.
Cache Me If You Must: Adaptive Key-Value Quantization for Large Language Models
Alina Shutova (HSE University), Dan Alistarh (ISTA)
CompressionTransformerLarge Language ModelText
🎯 What it does: An adaptive quantization method AQUA-KV is proposed for the Key-Value cache of large language models, which predicts and quantizes the residuals by utilizing the linear correlation between keys and values in adjacent and same layers, significantly compressing the KV cache while maintaining inference accuracy.
CACTI: Leveraging Copy Masking and Contextual Information to Improve Tabular Data Imputation
Aditya Gorla (University of California Los Angeles), Sriram Sankararaman (University of California Los Angeles)
TransformerAuto EncoderTabular
🎯 What it does: A Transformer-based autoencoder model CACTI is proposed to fill in missing values in tabular data.
CAD-Editor: A Locate-then-Infill Framework with Automated Training Data Synthesis for Text-Based CAD Editing
Yu Yuan (University of Science and Technology of China), Jiang Bian (Microsoft Research Asia)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: This paper proposes and implements the first text instruction-based CAD editing framework, CAD-Editor, which can automatically modify existing CAD models based on user natural language descriptions.
CaDA: Cross-Problem Routing Solver with Constraint-Aware Dual-Attention
Han Li (Southern University of Science and Technology), Zhenkun Wang (Southern University of Science and Technology)
OptimizationTransformerReinforcement LearningPrompt EngineeringTabular
🎯 What it does: A cross-problem constraint-aware dual attention model (CaDA) is proposed for unified learning of various vehicle routing problems (VRP).
Calibrated Language Models and How to Find Them with Label Smoothing
Jerry Huang (Universite de Montreal), QIUHAO Zeng
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper studies the confidence calibration problem of large language models after instruction fine-tuning and demonstrates that label smoothing can alleviate overconfidence in most cases.
Calibrated Physics-Informed Uncertainty Quantification
Vignesh Gopakumar (University College London), Marc Peter Deisenroth (University College London)
OptimizationComputational EfficiencyTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper proposes a Confidence Prediction (CP-PRE) framework based on physical residuals for label-free, model-agnostic uncertainty quantification (UQ) of neural PDE solvers, providing statistical coverage guarantees.
Calibrated Value-Aware Model Learning with Probabilistic Environment Models
Claas A Voelcker, Amir-massoud Farahmand (Polytechnique Montreal)
Reinforcement Learning
🎯 What it does: This paper conducts a theoretical analysis of the Value-Aware Model Learning (VAML) loss (including MuZero loss) and finds that it is uncalibrated when using a stochastic environment model. It proposes a CVAML loss that is corrected for variance; subsequently, its performance is validated in the finite state Garnet MDP and the Humanoid and Dog tasks of the DeepMind Control Suite.
Calibrating Video Watch-time Predictions with Credible Prototype Alignment
Chao Cui (Kuaishou Inc), Hechang Chen (Jilin University)
Recommendation SystemAuto EncoderVideoMultimodality
🎯 What it does: A two-stage ProWTP framework is proposed to predict video viewing duration; the first stage uses Hierarchical Vector Quantization Variational Autoencoder (HVQ-VAE) to transform the multimodal continuous distribution of viewing proportions into high-dimensional discrete prototypes; the second stage aligns instance representations with prototype representations through Semi-Relaxed Unbalanced Optimal Transport (SUOT), incorporating allocation loss and compactness loss to refine clustering, ultimately achieving more accurate predictions using weighted prototype vectors in a regression model.
CALM: Consensus-Aware Localized Merging for Multi-Task Learning
Kunda Yan (Tsinghua University), Changshui Zhang (Tsinghua University)
ClassificationOptimizationTransformerSupervised Fine-TuningImageText
🎯 What it does: A new multi-task learning model fusion method called CALM is proposed, which can merge multiple fine-tuned models into a unified model without retraining, while maintaining the performance of each task.
Can Biologically Plausible Temporal Credit Assignment Rules Match BPTT for Neural Similarity? E-prop as an Example
Yuhan Helena Liu (Princeton University), Christopher J Cueva
Recurrent Neural NetworkTime SeriesSequential
🎯 What it does: Compared the biological feasible time credit assignment rule e-prop with traditional BPTT in terms of similarity to neural recordings after task learning.
Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence
Yuankai Luo (Beihang University), Xiao-Ming Wu (Hong Kong Polytechnic University)
Graph Neural NetworkGraphBenchmark
🎯 What it does: This study investigates the performance of classic GNNs in graph-level tasks and proposes a unified GNN+ framework to enhance their capabilities.
Can Compressed LLMs Truly Act? An Empirical Evaluation of Agentic Capabilities in LLM Compression
Peijie Dong (Hong Kong University of Science and Technology), Bo Li (Hong Kong University of Science and Technology)
CompressionOptimizationTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: A benchmark for evaluating the capabilities of compressed LLMs in agent tasks, called ACBench, is proposed, and a systematic assessment of the impact of quantization and pruning on workflow generation, tool usage, long context understanding, and real applications is conducted.
Can DBNNs Robust to Environmental Noise for Resource-constrained Scenarios?
Wendong Zheng (Shanxi University), Wenjian Wang (Shanxi University)
ClassificationCompressionImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a robust training framework based on L1,∞ norm constraints to enhance the inference robustness of Deep Binary Neural Networks (DBNN) under environmental noise, and provides an analysis of the corresponding noise disturbance upper bound.
Can Diffusion Models Learn Hidden Inter-Feature Rules Behind Images?
Yujin Han (University of Hong Kong), Difan Zou (University of Hong Kong)
GenerationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: This paper systematically evaluates the ability of diffusion models (DM) to learn implicit rules between image features (such as light and shadow, mirroring, size-texture, etc.) through the design of four types of adjustable fineness synthetic tasks and theoretical analysis.
Can Large Language Models Understand Intermediate Representations in Compilers?
Hailong Jiang (Kent State University), Qiang Guan (Kent State University)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Systematically evaluate the understanding ability of six major LLMs on compiler intermediate representation (IR) through four types of tasks (CFG reconstruction, IR decompilation, code summarization, execution reasoning).
Can MLLMs Reason in Multimodality? EMMA: An Enhanced MultiModal ReAsoning Benchmark
Yunzhuo Hao (University of Electronic Science and Technology of China), Yu Cheng (Chinese University of Hong Kong)
TransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmarkPhysics RelatedChain-of-Thought
🎯 What it does: This paper proposes the EMMA multimodal reasoning benchmark to evaluate the performance of large language models on tasks that require reasoning using both images and text.
Can RLHF be More Efficient with Imperfect Reward Models? A Policy Coverage Perspective
Jiawei Huang (ETH Zurich), Niao He (ETH Zurich)
OptimizationReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper studies how to utilize existing reward models, which may not be of perfect quality, in RLHF (Reinforcement Learning from Human Feedback) to improve the sample efficiency of online learning.
Can Transformers Learn Full Bayesian Inference in Context?
Arik Reuter (Ludwig Maximilian University of Munich), David Rügamer
TransformerDiffusion modelFlow-based ModelTabularOrdinary Differential Equation
🎯 What it does: A context learning framework based on transformers is proposed, utilizing flow matching and continuous normalizing flows to achieve complete Bayesian inference for models such as GLM, FA, and GMM.
Can Transformers Reason Logically? A Study in SAT Solving
Leyan Pan (Georgia Institute of Technology), Wenke Lee (Georgia Institute of Technology)
TransformerChain-of-Thought
🎯 What it does: This paper proves and implements a Chain-of-Thought (CoT) architecture using only a Decoder-only Transformer to determine the 3-SAT problem under non-uniform models. It converts the theoretical construction into an executable PyTorch model using the PARAT tool and subsequently evaluates it on three types of datasets: random, marginalized, and skewed.
Can We Predict Performance of Large Models across Vision-Language Tasks?
Qinyu Zhao (Australian National University), Stephen Gould (Australian National University)
Recommendation SystemTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: Construct a performance matrix for large visual language models (LVLM) across multiple tasks, and predict unassessed model-task scores through probabilistic matrix factorization, proposing an uncertainty-based active evaluation strategy.
CAN: Leveraging Clients As Navigators for Generative Replay in Federated Continual Learning
Xuankun Rong (Wuhan University), Mang Ye (Wuhan University)
GenerationFederated LearningImage
🎯 What it does: In federated incremental learning, generative replay is utilized with clients acting as navigators to alleviate forgetting.
Cannot See the Forest for the Trees: Invoking Heuristics and Biases to Elicit Irrational Choices of LLMs
Haoming Yang (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A two-stage iterative jailbreak framework ICRT based on cognitive heuristics is proposed, inducing LLMs to make irrational decisions and generate harmful content.
Canonical Rank Adaptation: An Efficient Fine-Tuning Strategy for Vision Transformers
Lokesh Veeramacheneni (University of Bonn), Juergen Gall (University of Bonn)
ClassificationRecognitionOptimizationTransformerSupervised Fine-TuningImage
🎯 What it does: A new parameter-efficient fine-tuning method called Canonical Rank Adaptation (CaRA) is proposed for fine-tuning a small number of parameters on Vision Transformers.
Cape: Context-Aware Prompt Perturbation Mechanism with Differential Privacy
Haoqi Wu (TikTok), Qiang Yan
GenerationSafty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A prompt perturbation mechanism called Cape is proposed, based on local differential privacy, to hide user sensitive information during inference in large language models.
Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation
Jan Pauls (University of Münster), Fabian Gieseke (Zuse Institute Berlin)
SegmentationDepth EstimationConvolutional Neural NetworkImageTime Series
🎯 What it does: Using time series images from Sentinel-1 and Sentinel-2, combined with sparse elevation labels from GEDI LiDAR, a 3D U-Net model was trained to generate time series canopy height maps at a 10 m resolution for the European continent from 2019 to 2022.
CASE-Bench: Context-Aware SafEty Benchmark for Large Language Models
Guangzhi Sun (University of Cambridge), Jose Such (Universitat Politècnica de València)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper presents CASE-Bench, a benchmark for LLM that incorporates context into security assessments. It utilizes CI theory to generate 900 query-context pairs from 450 harmful queries and collects non-binary security ratings from over 2000 high-quality annotators.
CAT Merging: A Training-Free Approach for Resolving Conflicts in Model Merging
Wenju Sun (Beijing Jiaotong University), Boyang Li (Nanyang Technological University)
OptimizationTransformerImageTextMultimodality
🎯 What it does: This paper proposes a training-free model merging method called Conflict-Aware Task Merging (CAT Merging), which achieves the integration of multi-task models by removing high-conflict components from the task vectors.
CAT: Contrastive Adversarial Training for Evaluating the Robustness of Protective Perturbations in Latent Diffusion Models
Sen Peng (City University of Hong Kong), Xiaohua Jia (City University of Hong Kong)
GenerationData SynthesisAdversarial AttackDiffusion modelContrastive LearningImage
🎯 What it does: This paper studies the robustness of using protective perturbations in latent diffusion models to prevent unauthorized model customization and proposes a Contrastive Adversarial Training (CAT) method based on lightweight adapters, which reduces the effect of protective perturbations by reshaping latent representations.
Catch Your Emotion: Sharpening Emotion Perception in Multimodal Large Language Models
Yiyang Fang (Wuhan University), Mang Ye (Wuhan University)
ClassificationRecognitionTransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper studies a framework called SEPM for emotion recognition in multimodal large language models that is free to train and enhances the inference phase.
Catching Two Birds with One Stone: Reward Shaping with Dual Random Networks for Balancing Exploration and Exploitation
Haozhe Ma (National University of Singapore), Tze-Yun Leong (National University of Singapore)
Reinforcement LearningSequential
🎯 What it does: Proposes the Dual Random Networks Distillation (DuRND) framework, which uses both positive and negative random network modules to simultaneously generate exploration rewards and contribution rewards, balancing exploration and exploitation in reinforcement learning (RL);
Categorical Distributional Reinforcement Learning with Kullback-Leibler Divergence: Convergence and Asymptotics
Tyler Kastner (University of Toronto), Amir-massoud Farahmand (Polytechnique Montreal)
Reinforcement LearningTabular
🎯 What it does: This paper studies the problem of distributed reinforcement learning using classification parameterization and KL divergence loss, proposing a preprocessed version of the algorithm and proving its convergence.
Categorical Schrödinger Bridge Matching
Grigoriy Ksenofontov (Skolkovo Institute of Science and Technology), Alexander Korotin (Skolkovo Institute of Science and Technology)
GenerationData SynthesisGenerative Adversarial NetworkImageText
🎯 What it does: This paper proposes a Schrödinger bridge matching algorithm for discrete spaces (CSBM), theoretically proving that the discrete time Iterative Markov Fitting (D-IMF) can converge to the SB solution under discrete Markov reference processes, and provides a feasible algorithm implementation.
CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration
Haoyun Jiang (Shanghai Jiao Tong University), Jiangchao Yao (Shanghai Jiao Tong University)
OptimizationComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: A hybrid KV caching method called CateKV is proposed, based on attention head sequence consistency recognition, to accelerate long-context LLM inference.
Catoni Contextual Bandits are Robust to Heavy-tailed Rewards
Chenlu Ye (University of Illinois Urbana-Champaign), Tong Zhang (University of Illinois Urbana-Champaign)
OptimizationReinforcement Learning
🎯 What it does: This study proposes a context multi-armed bandit algorithm based on the Catoni mean estimator, which maintains low regret even when rewards have heavy-tailed distributions or large value ranges.
Causal Abstraction Inference under Lossy Representations
Kevin Muyuan Xia (Columbia University), Elias Bareinboim (Columbia University)
Generative Adversarial NetworkImage
🎯 What it does: This paper proposes a projection abstraction framework that allows for causal abstraction and inference under distorted representations, and provides a construction algorithm.
Causal Abstraction Learning based on the Semantic Embedding Principle
Gabriele D'Acunto (Sapienza University), Paolo Di Lorenzo (Sapienza University)
Biomedical DataMagnetic Resonance Imaging
🎯 What it does: A causal abstraction learning framework based on the principle of semantic embedding is proposed to address the problem of linear causal abstraction learning in the absence of observable aligned data.
Causal Attribution Analysis for Continuous Outcomes
Shanshan Luo (Beijing Technology and Business University), Zhi Geng (Beijing Technology and Business University)
Tabular
🎯 What it does: A causal attribution framework and identifiable posterior causal estimators for continuous outcomes are proposed.
Causal Discovery from Conditionally Stationary Time Series
Carles Balsells-Rodas (Imperial College London), Yingzhen Li (Imperial College London)
Graph Neural NetworkAuto EncoderTime Series
🎯 What it does: A causal discovery method for conditionally stationary time series, SCDI, is proposed, which can infer state-dependent causal structures from observational data.
Causal Effect Identification in lvLiNGAM from Higher-Order Cumulants
Daniele Tramontano (Technical University of Munich), Mathias Drton (Munich Center for Machine Learning)
Tabular
🎯 What it does: This paper utilizes higher-order cumulants in the latent variable linear non-Gaussian model (lvLiNGAM) to identify and estimate causal effects, particularly addressing the challenges of single proxy variables and underdetermined instrumental variables.
Causal Invariance-aware Augmentation for Brain Graph Contrastive Learning
Minqi Yu (Beijing University of Technology), Junzhong Ji (Beijing University of Technology)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraphBiomedical DataAlzheimer's Disease
🎯 What it does: A CIA-GCL method based on causal invariance adaptive augmentation is proposed for multi-site brain network disease diagnosis in brain graph contrastive learning.
Causal Logistic Bandits with Counterfactual Fairness Constraints
Jiajun Chen (Iowa State University), Christopher John Quinn (Iowa State University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabular
🎯 What it does: This study investigates the multi-armed bandit problem with causal logistic regression rewards, incorporating constraints based on counterfactual fairness.
Causal-PIK: Causality-based Physical Reasoning with a Physics-Informed Kernel
Carlota Parés Morlans (Stanford University), Jeannette Bohg (Stanford University)
OptimizationRobotic IntelligenceReinforcement LearningTabularBenchmarkPhysics Related
🎯 What it does: Proposes the Causal-PIK method, which utilizes Bayesian optimization combined with Physics-Informed Kernel to quickly search for optimal actions in single-intervention physical reasoning tasks, thereby solving problems with very few attempts.
Causality Inspired Federated Learning for OOD Generalization
Jiayuan Zhang (Beihang University), Xinghao Wu
Domain AdaptationFederated LearningContrastive LearningImage
🎯 What it does: This paper proposes the FedUni framework, which simultaneously extracts all possible causal features in federated learning and filters out unnecessary features through an irrelevant causal feature compressor, thereby enhancing the model's generalization ability under unknown distributions.
Causality-Aware Contrastive Learning for Robust Multivariate Time-Series Anomaly Detection
HyunGi Kim (Seoul National University), Sungroh Yoon (Seoul National University)
Anomaly DetectionRecurrent Neural NetworkTransformerContrastive LearningTime Series
🎯 What it does: A multivariate time series anomaly detection framework called CAROTS is proposed, which utilizes causal-aware contrastive learning to train a contrastive encoder with causal preservation and perturbation-enhanced samples, combined with prediction error for anomaly scoring.
Cavia: Camera-controllable Multi-view Video Diffusion with View-Integrated Attention
Dejia Xu (University of Texas at Austin), Hao Tang (Apple)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: The Cavia framework is designed to achieve multi-view video generation, allowing precise control of camera trajectories while maintaining object motion consistency.
CEGA: A Cost-Effective Approach for Graph-Based Model Extraction and Acquisition
Zebin Wang (Harvard University), Yushun Dong (Florida State University)
Graph Neural NetworkGraph
🎯 What it does: A framework for efficient model extraction and acquisition of graph neural networks under limited query budgets (CEGA) is proposed, which iteratively selects the most informative nodes for querying and trains an intermediate model to gradually approximate the target model.
CellFlux: Simulating Cellular Morphology Changes via Flow Matching
Yuhui Zhang (Stanford University), Serena Yeung-Levy (Stanford University)
GenerationData SynthesisFlow-based ModelImageBiomedical Data
🎯 What it does: The CellFlux model is constructed using flow matching technology to simulate morphological changes caused by chemical and genetic interventions at the cellular level.
Censor Dependent Variational Inference
Chuanhui Liu (Purdue University), Xiao Wang (Purdue University)
TabularBiomedical Data
🎯 What it does: This paper proposes a latent variable survival model method based on censor-dependent variational inference (CDVI), aimed at addressing the bias issue of traditional VI under right-censored data.
CERTAIN: Context Uncertainty-aware One-Shot Adaptation for Context-based Offline Meta Reinforcement Learning
Hongtu Zhou (Tongji University), Changjun Jiang (Tongji University)
Meta LearningReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes the CERTAIN method, which identifies uncertain (ambiguous or OOD) contexts through heteroscedastic uncertainty estimation in one-shot adaptation, and trains an uncertainty-aware context collection strategy, thereby achieving more robust task inference and higher adaptation performance.
Certifiably Robust Model Evaluation in Federated Learning under Meta-Distributional Shifts
Amir Najafi (Sharif University of Technology), Farzan Farnia (Chinese University of Hong Kong)
Federated LearningImage
🎯 What it does: This paper studies a verifiable robust evaluation method for model performance on the target network (which may have meta-distribution shifts) by utilizing the query information from clients in the source network in federated learning.
Certification for Differentially Private Prediction in Gradient-Based Training
Matthew Robert Wicker (Imperial College London), Calvin Tsay (Imperial College London)
ClassificationSafty and PrivacyTextBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a private prediction framework based on parameter space reachability analysis, utilizing convex relaxation and bound propagation to compute specific upper bounds on prediction sensitivity for datasets, thereby improving the estimation of differential privacy.
Certified Unlearning for Neural Networks
Anastasia Koloskova (Stanford University), Sanmi Koyejo (Stanford University)
Safty and PrivacyComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: To address the forgetting problem of machine learning models, an approximate no-learning method is proposed that can remove the influence of training samples upon request for deletion.
CFP-Gen: Combinatorial Functional Protein Generation via Diffusion Language Models
Junbo Yin (King Abdullah University of Science and Technology), Xin Gao (King Abdullah University of Science and Technology)
GenerationDrug DiscoveryTransformerDiffusion modelMultimodalityBiomedical Data
🎯 What it does: CFP-GEN is proposed, a multimodal functional protein generation model based on discrete diffusion, capable of simultaneously satisfying multiple constraints such as functional annotation, sequence, and structure to generate multifunctional proteins.
CFPT: Empowering Time Series Forecasting through Cross-Frequency Interaction and Periodic-Aware Timestamp Modeling
Feifei Kou (Beijing University of Posts and Telecommunications), Junping Du (Beijing University of Posts and Telecommunications)
OptimizationConvolutional Neural NetworkTransformerTime Series
🎯 What it does: The CFPT framework is proposed, which achieves cross-frequency interaction and periodic-aware timestamp modeling through a dual-branch structure to enhance long-term sequence prediction performance.
Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning
Wanyun Xie (École Polytechnique Fédérale de Lausanne), Volkan Cevher (École Polytechnique Fédérale de Lausanne)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The CHAMELEON framework is proposed, which utilizes Kernel Ridge Leverage Scores (KRLS) to assess the importance of various domains in the training data of language models, and calculates domain weights accordingly. This supports adaptive data mixing during both pre-training and fine-tuning phases, while allowing for quick adaptation to new domains without retraining the proxy model.
Channel Normalization for Time Series Channel Identification
Seunghan Lee (KRAFTON), Kibok Lee (Yonsei University)
Time Series
🎯 What it does: Proposes Channel Normalization (CN) and its adaptive (ACN) and prototype (PCN) variants to enhance channel identifiability (CID) in multi-channel time series models.
Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction
Yi He (University College London), Yukun Hu (University College London)
TransformerTime SeriesSequentialPhysics Related
🎯 What it does: A scalable Transformer framework specifically designed for large-scale ergodic chaotic systems is proposed, capable of autoregressively generating long time series trajectories while maintaining statistical invariance.
CHATS: Combining Human-Aligned Optimization and Test-Time Sampling for Text-to-Image Generation
Minghao Fu (Nanjing University), Kaifu Zhang (Alibaba Group)
GenerationOptimizationDiffusion modelFlow-based ModelImage
🎯 What it does: The CHATS framework is proposed, unifying human preference alignment with test-time sampling through the training of two models that learn the preferred and dispreferred distributions, and achieving collaboration between the two models during sampling using a proxy prompt.
Chip Placement with Diffusion Models
Vint Lee (University of California Berkeley), John Wawrzynek (University of California Berkeley)
OptimizationGraph Neural NetworkReinforcement LearningDiffusion modelGraph
🎯 What it does: This paper proposes the use of diffusion models for macro placement, achieving high-quality layouts for zero-shot samples on new circuits, replacing traditional reinforcement learning methods.
Circumventing Backdoor Space via Weight Symmetry
Jie Peng (Harbin Institute of Technology), Haoyu He (Monash University)
Adversarial AttackData-Centric LearningContrastive LearningImage
🎯 What it does: A two-stage symmetric connection (TSC) method is proposed for unsupervised backdoor cleaning of deep networks with implanted backdoors in both supervised and self-supervised learning.
CLARIFY: Contrastive Preference Reinforcement Learning for Untangling Ambiguous Queries
Ni Mu (Tsinghua University), Qing-Shan Jia (Tsinghua University)
TransformerReinforcement LearningContrastive LearningTabular
🎯 What it does: The CLARIFY method is proposed, which effectively eliminates ambiguous queries and improves labeling efficiency in offline preference reinforcement learning through contrastive learning of trajectory embeddings.
Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off
Yuecheng Li (Sun Yat-sen University), Chuan Chen (Sun Yat-sen University)
Federated LearningSafty and PrivacyImage
🎯 What it does: A new federated learning framework called FedCEO is proposed, aimed at balancing model utility and user privacy by enabling client collaboration.
CLIMB: Data Foundations for Large Scale Multimodal Clinical Foundation Models
Wei Dai (Massachusetts Institute of Technology), Paul Pu Liang (Massachusetts Institute of Technology)
ClassificationAnomaly DetectionData-Centric LearningGraph Neural NetworkTransformerSupervised Fine-TuningReinforcement LearningMultimodalityTime SeriesBiomedical DataUltrasoundElectronic Health RecordsElectrocardiogramBenchmark
🎯 What it does: This paper constructs the CLIMB benchmark, summarizing 44 public datasets covering 15 clinical modalities (2D, 3D, 1D, graphical, mixed) with a total of 4.51M samples and 19TB of data, and provides a unified interface and code.
Clipped SGD Algorithms for Performative Prediction: Tight Bounds for Stochastic Bias and Remedies
Qiang LI, Hoi To Wai
OptimizationSafty and PrivacyTabularFinance RelatedStochastic Differential Equation
🎯 What it does: This paper studies the convergence of the clipped stochastic gradient descent (SGD) algorithm under decision-dependent data distributions, particularly in the context of privacy-preserving optimization algorithms and expressive data interactions.
Clipping Improves Adam-Norm and AdaGrad-Norm when the Noise Is Heavy-Tailed
Savelii Chezhegov (Moscow Institute of Physics and Technology), Eduard Gorbunov (Mohamed bin Zayed University of Artificial Intelligence)
OptimizationTransformerSupervised Fine-TuningText
🎯 What it does: This paper studies the high-probability convergence performance of adaptive optimizers such as Adam and AdaGrad under heavy-tailed noise, and proves that not clipping the gradients can lead to a deterioration of high-probability complexity.
Clone-Robust AI Alignment
Ariel D. Procaccia (Harvard University), Shirley Zhang (Harvard University)
Reinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: A new reinforcement learning algorithm called weighted maximum likelihood estimation (weighted MLE) is proposed to improve the robustness of human feedback reinforcement learning (RLHF) algorithms when facing approximate clones.
Closed-form Solutions: A New Perspective on Solving Differential Equations
Shu Wei (Institute of Semiconductors, Chinese Academy of Sciences), Yan Pang (University of Chinese Academy of Sciences)
OptimizationRecurrent Neural NetworkReinforcement LearningTabularPhysics RelatedOrdinary Differential Equation
🎯 What it does: By generating symbolic expression skeletons through reinforcement learning and incorporating physical constraints, a closed-form analytical solution for differential equations is obtained.
Closed-Loop Long-Horizon Robotic Planning via Equilibrium Sequence Modeling
Jinghan Li (Peking University), Yadong MU
OptimizationRobotic IntelligenceLarge Language ModelWorld ModelSequential
🎯 What it does: A self-improving large language model (LLM) planner based on a deep equilibrium model is proposed, capable of achieving self-iterative optimization planning in closed-loop long-sequence robotic tasks.
CLOVER: Cross-Layer Orthogonal Vectors Pruning
Fanxu Meng (Peking University), Muhan Zhang (Peking University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A cross-layer orthogonal vector pruning method named CLOVER is proposed, which utilizes SVD to perform low-rank decomposition on the Query-Key and Value-Output matrices of each attention head to obtain orthogonal bases and singular values.
Clustering Items through Bandit Feedback: Finding the Right Feature out of Many
Maximilian Graf (University of Potsdam), Nicolas Verzelen (INRAE)
OptimizationTabular
🎯 What it does: An adaptive two-group clustering method called BanditClustering is proposed, which uses bandit feedback to gradually select features that can distinguish the two classes and complete the clustering accordingly.
Clustering Properties of Self-Supervised Learning
Xi Weng (Beihang University), Lei Huang (Beihang University)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes a positive feedback framework called ReSA based on self-supervised learning representations of clustering information, which guides contrastive learning by extracting high-quality clustering assignments at the encoding layer.
Clustering via Self-Supervised Diffusion
Roy Uziel (Ben Gurion University of the Negev), Ari Pakman (Ben Gurion University of the Negev)
Representation LearningTransformerDiffusion modelContrastive LearningImage
🎯 What it does: A self-supervised clustering framework called CLUDI is proposed using diffusion models.
CMoS: Rethinking Time Series Prediction Through the Lens of Chunk-wise Spatial Correlations
Haotian Si (Computer Network Information Center, Chinese Academy of Sciences), Gaogang Xie (Computer Network Information Center, Chinese Academy of Sciences)
Mixture of ExpertsTime Series
🎯 What it does: This paper proposes a super lightweight multivariate time series forecasting model CMoS, which directly models the spatial correlations between different time segment blocks, replacing traditional shape embedding methods.
CoastalBench: A Decade-Long High-Resolution Dataset to Emulate Complex Coastal Processes
Zelin Xu (University of Florida), Zhe Jiang (University of Florida)
TransformerTime Series
🎯 What it does: This paper presents CoastalBench, a high-resolution ocean simulation dataset covering the Port of Carloton area in South Florida, with a spatiotemporal resolution of 30 minutes, <100 meters, spanning ten years, and including ocean variables, meteorological and river forcing, as well as static physical features; it also designs a model based on Vision Transformer that can perform conditional predictions for different forecasting time slots.
CoCoA-Mix: Confusion-and-Confidence-Aware Mixture Model for Context Optimization
Dasol Hong (Korea Advanced Institute of Science and Technology), Hyun Myung (Korea Advanced Institute of Science and Technology)
OptimizationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: The CoCoA-Mix framework is proposed, which enhances the specialization and generalization of VLM prompt tuning by combining mixed models with confusion-aware loss and confidence-aware weights.
Code-Generated Graph Representations Using Multiple LLM Agents for Material Properties Prediction
Jiao Huang (Jilin University), Bo Yang (Jilin University)
Graph Neural NetworkLarge Language ModelGraphPhysics Related
🎯 What it does: This paper proposes a code generation framework called Rep-CodeGen, based on multi-large language model agents, for automatically generating graph representations that satisfy material structure symmetry constraints.
CodeIO: Condensing Reasoning Patterns via Code Input-Output Prediction
Junlong Li (DeepSeek-AI), Junxian He (HKUST)
AI Code AssistantSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Transform the code into executable functions and set the prediction task as a given function and query, predicting the natural language reasoning chain (CoT) for inputs or outputs.
CodeSteer: Symbolic-Augmented Language Models via Code/Text Guidance
Yongchao Chen (Harvard University), Chuchu Fan (Massachusetts Institute of Technology)
OptimizationAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed and implemented the CodeSteer framework, which intelligently switches between text reasoning and code generation in multi-turn interactions through a small model guiding a large language model, significantly improving symbolic computation performance.
CodeSync: Synchronizing Large Language Models with Dynamic Code Evolution at Scale
Chenlong Wang (Huazhong University of Science and Technology), Dongping Chen (Huazhong University of Science and Technology)
AI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: The CODESYNC data engine and CODESYNCBENCH benchmark are proposed to evaluate the adaptability of LLMs to the evolution of real-time APIs.
CoDy: Counterfactual Explainers for Dynamic Graphs
Zhan Qu (TU Dresden), Michael Färber
Explainability and InterpretabilityComputational EfficiencyGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Designed and implemented CoDy—a counterfactual explanation method for continuous-time dynamic graph neural networks (TGNN) that can automatically search for and provide minimized event subgraphs to explain the reasons behind model predictions.
COExpander: Adaptive Solution Expansion for Combinatorial Optimization
Jiale Ma (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationGraph Neural NetworkDiffusion modelGraphTabular
🎯 What it does: The Adaptive Expansion (AE) framework is proposed, and the COExpander solver is implemented, combining the advantages of global prediction and local construction to solve combinatorial optimization problems by adaptively expanding decision variables with variable step sizes.
CogMath: Assessing LLMs' Authentic Mathematical Ability from a Human Cognitive Perspective
Jiayu Liu (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
Large Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: The CogMath framework is proposed to systematically evaluate the mathematical abilities of LLMs from the perspective of human cognition, covering three stages and nine dimensions: problem understanding, solving, and summarizing.
COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning
Chamika Sudusinghe (University of Illinois Urbana-Champaign), Charith Mendis (University of Illinois Urbana-Champaign)
Auto Encoder
🎯 What it does: The COGNATE framework is proposed, utilizing a low-cost data pre-training model on the CPU, followed by few-shot fine-tuning on emerging sparse accelerators (such as SPADE, A100) to achieve automatic configuration and acceleration of sparse tensor programs.