NeurIPS 2025 Papers — Page 3
Conference on Neural Information Processing Systems · 5275 papers
AdaptGrad: Adaptive Sampling to Reduce Noise
Linjiang Zhou (Wuhan University), XIAOCHUAN SHI
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an adaptive noise sampling-based gradient smoothing method called AdaptGrad, aimed at reducing noise in gradient interpretation methods and improving visualization quality.
Adapting to Stochastic and Adversarial Losses in Episodic MDPs with Aggregate Bandit Feedback
Shinji Ito (University of Tokyo and RIKEN), Taira Tsuchiya (University of Tokyo and RIKEN)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes the Best of Both Worlds (BOBW) algorithm with feedback in finite-horizon discrete MDPs, achieving logarithmic expected regret in stochastic environments and square-root expected regret in adversarial environments.
Adaptive 3D Reconstruction via Diffusion Priors and Forward Curvature-Matching Likelihood Updates
Seunghyeok Shin (Inha University), Hongki Lim (Inha University)
GenerationData SynthesisOptimizationDiffusion modelImagePoint Cloud
🎯 What it does: A diffusion backward sampling framework based on Forward Curvature Matching (FCM) is proposed to recover high-quality colored point clouds from single or multi-view images, depth maps, and other arbitrary measurements.
Adaptive Algorithms with Sharp Convergence Rates for Stochastic Hierarchical Optimization
Xiaochuan Gong (George Mason University), Mingrui Liu (George Mason University)
OptimizationHyperparameter SearchText
🎯 What it does: An adaptive algorithm suitable for stochastic hierarchical optimization (including non-convex-strongly concave minimization and non-convex-strongly convex bilevel optimization) is proposed;
Adaptive and Multi-scale Affinity Alignment for Hierarchical Contrastive Learning
Jiawei Huang (University of Science and Technology of China), Hu Ding (University of Science and Technology of China)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical DataUltrasound
🎯 What it does: An Adaptive Multi-Scale Affinity Alignment (AMA-alignment) framework is proposed, which aligns semantic and representation affinity graphs at different scales using local contrastive loss and distributionally robust optimization, significantly enhancing the preservation of hierarchical structures and performance on downstream tasks.
Adaptive Batch-Wise Sample Scheduling for Direct Preference Optimization
Zixuan Huang (Beihang University), deqing wang
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The 'sample scheduling' problem is proposed, and the SamS algorithm is designed to dynamically and adaptively select samples from each batch during Direct Preference Optimization (DPO) training to enhance the human preference alignment effect of large language models (LLMs).
Adaptive Cannistraci-Hebb Network Automata Modelling of Complex Networks for Path-based Link Prediction
Jialin Zhao (Tsinghua University), Carlo Vittorio Cannistraci (University of Padova)
Graph Neural NetworkGraphTime Series
🎯 What it does: The Cannistraci-Hebb Adaptive (CHA) network automaton is proposed for topological link prediction without relying on node attributes, and it can automatically select the best CH rules and path lengths.
Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking
Pengxiang Li (Polytechnic University), Xiaowei Gao (Peking University)
GenerationDiffusion modelText
🎯 What it does: Proposes Adaptive Classifier-Free Guidance (A-CFG), which improves the unsupervised guidance of iterative masked diffusion language models by dynamically identifying low-confidence positions in the model and temporarily re-masking them.
Adaptive Context Length Optimization with Low-Frequency Truncation for Multi-Agent Reinforcement Learning
Wenchang Duan (Shanghai Jiao Tong University), Yi Shi (Shanghai Jiao Tong University)
OptimizationReinforcement LearningTime SeriesSequential
🎯 What it does: A multi-agent reinforcement learning framework based on a central agent that adaptively optimizes context length and combines Fourier low-frequency truncation to extract global temporal trends (ACL-LFT) is proposed.
Adaptive Data Analysis for Growing Data
Neil G Marchant, Benjamin I. P. Rubinstein (University of Melbourne)
OptimizationSafty and Privacy
🎯 What it does: This study investigates the generalization guarantees of adaptive data analysis on datasets that grow over time, proposing the first generalization bounds for dynamic data.
Adaptive Data-Borrowing for Improving Treatment Effect Estimation using External Controls
Qinwei Yang (Beijing Technology and Business University), Peng Wu (Beijing Technology and Business University)
Tabular
🎯 What it does: Proposes an external control sample adaptive borrowing method based on influence functions to improve the efficiency of RCT treatment effect estimation.
Adaptive Defense against Harmful Fine-Tuning for Large Language Models via Bayesian Data Scheduler
Zixuan Hu (Nanyang Technological University), Dacheng Tao (Nanyang Technological University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study investigates adaptive defenses against harmful fine-tuning in large language models, proposing the Bayesian Data Scheduler (BDS) framework, which can safely weight training data without the need for attack simulations, thereby enhancing the model's safety and performance.
Adaptive Discretization for Consistency Models
Jiayu Bai (Huazhong University of Science and Technology), Zenan Ling (Huazhong University of Science and Technology)
GenerationOptimizationDiffusion modelImage
🎯 What it does: This paper proposes an Adaptive Discretization Method (ADCMs) for training Consistency Models (CMs), which dynamically adjusts the discretization step size over time to balance the model's trainability and stability.
Adaptive Distraction: Probing LLM Contextual Robustness with Automated Tree Search
Yanbo Wang (Mohamed bin Zayed University of Artificial Intelligence), Xiangliang Zhang (University of Notre Dame)
OptimizationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: An adaptive interference generation framework based on tree search is proposed, which automatically generates semantically coherent and task-independent contextual interference to test the contextual robustness of LLMs.
Adaptive Divergence Regularized Policy Optimization for Fine-tuning Generative Models
Jiajun Fan (University of Illinois Urbana-Champaign), Ge Liu (University of Illinois Urbana-Champaign)
GenerationOptimizationReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningTextMultimodalityAudio
🎯 What it does: An adaptive divergence regularization strategy (ADRPO) has been designed and validated for reinforcement learning fine-tuning of generative models (text-image, LLM, multimodal audio) to automatically balance exploration and exploitation.
Adaptive Fission: Post-training Encoding for Low-latency Spike Neural Networks
Yizhou Jiang (Tsinghua University), Ying Fang (Fujian Normal University)
ClassificationGenerationSpiking Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposes Adaptive Fission, a post-training multi-neuron group encoding method that can convert ANNs to SNNs while significantly reducing latency and energy consumption.
Adaptive Frontier Exploration on Graphs with Applications to Network-Based Disease Testing
Davin Choo (Harvard University), Cheryl Johnson (World Health Organization)
OptimizationGraph Neural NetworkReinforcement LearningGraphBiomedical Data
🎯 What it does: An adaptive decision framework AFEG suitable for frontier exploration on graphs is proposed, and a Gittins index-based strategy GITTINS is designed, proving its optimality on tree-structured graphs.
Adaptive Gradient Masking for Balancing ID and MLLM-based Representations in Recommendation
Yidong Wu (Imperial College London), Jiechao Gao (Stanford University)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningImageVideoTextMultimodality
🎯 What it does: This paper proposes a two-stage framework that first fine-tunes a multimodal large language model (MLLM) to generate unified multimodal representations of users and products, aligning them with ID embeddings afterward. It then uses Adaptive Gradient Masking (AGM) to dynamically balance the gradient updates of the ID branch and the MLLM branch, addressing the issue of inconsistent convergence speeds during joint training.
Adaptive Inference-Time Scaling via Cyclic Diffusion Search
Gyubin Lee (Korea Advanced Institute of Science and Technology), Sungjin Ahn (New York University)
GenerationData SynthesisComputational EfficiencyDiffusion modelImageText
🎯 What it does: Proposes the Adaptive Bi-directional Cyclic Diffusion (ABCD) framework, which achieves adaptive computation allocation during inference through cyclic forward/backward diffusion, automatic exploration-exploitation balance, and adaptive termination, thereby enhancing multi-task generation quality.
Adaptive Kernel Design for Bayesian Optimization Is a Piece of CAKE with LLMs
Richard Cornelius Suwandi (Chinese University of Hong Kong), Sergios Theodoridis (University of Athens)
OptimizationHyperparameter SearchLarge Language ModelPrompt EngineeringTabular
🎯 What it does: Proposes a method for adaptive Gaussian process kernel evolution based on large language models (CAKE), which generates and optimizes kernel structures in real-time during the Bayesian optimization process.
Adaptive Latent-Space Constraints in Personalized Federated Learning
Sana Ayromlou (Vector Institute), D. B. Emerson (Vector Institute)
Federated LearningImage
🎯 What it does: In federated learning, adaptive MMD metrics (MK-MMD and MMD-D) are introduced into personalized FL frameworks such as Ditto and MR-MTL to constrain the potential space distribution differences between local models and global models, thereby enhancing model performance.
Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning
Lei Wang (University of Florida), Jie Xu (University of Florida)
Federated LearningTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This work addresses the low-rank adaptation (LoRA) fine-tuning problem of large language models in federated learning and proposes the FedLEASE framework, which can automatically allocate and select expert modules.
Adaptive Neighborhood-Constrained Q Learning for Offline Reinforcement Learning
Yixiu Mao (Tsinghua University), Xiangyang Ji (Tsinghua University)
Reinforcement LearningTabular
🎯 What it does: An offline reinforcement learning method based on adaptive neighborhood constraints, ANQ, is proposed, utilizing bi-level optimization to implement Q-learning and avoid OOD errors.
Adaptive Prediction-Powered AutoEval with Reliability and Efficiency Guarantees
Sangwoo Park (King's College London), Osvaldo Simeone (King's College London)
Large Language ModelPrompt EngineeringText
🎯 What it does: A self-assessment method R-AutoEval+ is proposed, which can provide reliable performance estimates with limited samples and automatically reduce reliance on synthetic data when needed, ensuring improved sample efficiency without compromising reliability.
Adaptive Preference Arithmetic: A Personalized Agent with Adaptive Preference Arithmetic for Dynamic Preference Modeling
Hongyi Nie (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
Recommendation SystemTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper proposes the AdaPA-Agent framework for dynamically modeling user preference intensity in LLM agents, combining alignment estimation and preference arithmetic to achieve personalized generation.
Adaptive Quantization in Generative Flow Networks for Probabilistic Sequential Prediction
Nadhir Hassen (University of Adelaide), Johan W. Verjans
GenerationRecurrent Neural NetworkTransformerReinforcement LearningFlow-based ModelTime SeriesSequentialBiomedical DataElectronic Health Records
🎯 What it does: A novel time series prediction framework, Temporal GFN, is designed based on Generative Flow Networks (GFN) to achieve probabilistic predictions of continuous values.
Adaptive Re-calibration Learning for Balanced Multimodal Intention Recognition
Qu Yang (Wuhan University), Mang Ye (Wuhan University)
RecognitionTransformerSupervised Fine-TuningMultimodality
🎯 What it does: An adaptive recalibration learning framework (ARL) for multimodal intent recognition is proposed, which alleviates the modality imbalance problem by dynamically recalibrating modality importance.
Adaptive Riemannian ADMM for Nonsmooth Optimization: Optimal Complexity without Smoothing
Kangkang Deng (National University of Defense Technology), Hongxia Wang (National University of Defense Technology)
Optimization
🎯 What it does: An adaptive Riemannian ADMM (ARADMM) is proposed for minimizing the sum of smooth functions and non-smooth convex regularization terms on compact Riemannian submanifolds, achieving convergence without the need for smoothing on the non-smooth terms for the first time.
Adaptive Sigmoid Clipping for Balancing the Direction–Magnitude Mismatch Trade-off in Differentially Private Learning
Faeze Moradi Kalarde (University of Toronto), Min Dong (Ontario Tech University)
OptimizationSafty and PrivacyImageText
🎯 What it does: This paper proposes an adaptive Sigmoid clipping method called AdaSig, which balances the bias in gradient direction and magnitude during differential privacy training.
Adaptive Stochastic Coefficients for Accelerating Diffusion Sampling
Ruoyu Wang (Westlake University), Chi Zhang (Westlake University)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposes AdaSDE, a single-step SDE sampler that achieves adaptive noise coefficient learning at low step counts;
Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks
Korneel Van den Berghe (Delft University of Technology), Guido De Croon
Robotic IntelligenceSpiking Neural NetworkReinforcement LearningSequential
🎯 What it does: An adaptive agent gradient scheduling method suitable for sequential reinforcement learning is proposed, and a low-level quadrotor controller training based on spiking neural networks is achieved in conjunction with a jump-guided policy.
Adaptive Time Encoding for Irregular Multivariate Time-Series Classification
Sangho Lee (Gyeongsang National University), Hyungrok Do (Dongguk University)
ClassificationRecurrent Neural NetworkTime SeriesSequentialBiomedical DataElectronic Health Records
🎯 What it does: For irregular multivariate time series classification, ATENet is proposed to perform adaptive time encoding through learnable reference time points and construct fixed-length representations, followed by predictions using a simple classifier.
Adaptive Variance Inflation in Thompson Sampling: Efficiency, Safety, Robustness, and Beyond
Feng Zhu (Massachusetts Institute of Technology), David Simchi-Levi (Massachusetts Institute of Technology)
Reinforcement LearningTabular
🎯 What it does: An adaptive variance inflation TS-VI algorithm is proposed in Gaussian Thompson Sampling, aiming to improve efficiency, safety, and robustness simultaneously;
Adaptively Coordinating with Novel Partners via Learned Latent Strategies
Benjamin Li (Carnegie Mellon University), Simon Stepputtis (Virginia Tech)
Robotic IntelligenceReinforcement LearningAuto EncoderSequential
🎯 What it does: This paper presents TALENTS, which can achieve immediate adaptive collaboration by learning the potential strategy space when unknown partners appear.
AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking
Xiangqi Wang (University of Notre Dame), Xiangliang Zhang (University of Notre Dame)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Designed and trained a plugin named AdaReasoner to automatically adjust the reasoning configurations (prompting method, temperature, number of steps) of LLMs for different tasks.
AdaSPEC: Selective Knowledge Distillation for Efficient Speculative Decoders
Yuezhou Hu (University of California), Tuo Zhao (Georgia Institute of Technology)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes AdaSPEC, a selective knowledge distillation method under the Speculative Decoding framework, aimed at training small draft models and enhancing their alignment with the target model.
AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners
Woosung Koh (KAIST AI), Se-Young Yun (KAIST AI)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The AdaSTaR algorithm is proposed, incorporating adaptive sampling into the STaR self-training framework to improve efficiency and performance.
AdaTS: Learning Adaptive Time Series Representations via Dynamic Soft Contrasts
Denizhan Kara (University of Illinois Urbana-Champaign), Tarek F. Abdelzaher (University of Illinois Urbana-Champaign)
ClassificationRepresentation LearningContrastive LearningTime SeriesSequential
🎯 What it does: This paper proposes AdaTS, a pluggable self-supervised temporal contrastive learning module specifically designed to model the physical correlations and non-stationarity of temporal data.
AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding
Zhucun Xue (Zhejiang University), Dacheng Tao (Nanyang Technological University)
RetrievalComputational EfficiencyLarge Language ModelVideoTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the AdaVideoRAG framework, which utilizes query intent classification to dynamically select different retrieval strategies (no retrieval, naive retrieval, graph retrieval) for efficient deep understanding of multimodal question answering in long videos.
Additive Models Explained: A Computational Complexity Approach
Shahaf Bassan (Hebrew University of Jerusalem), Guy Katz (Hebrew University of Jerusalem)
Explainability and Interpretability
🎯 What it does: This study systematically analyzes the complexity of explanation generation in Generalized Additive Models (GAM) under different input domains, component models, and types of explanations, revealing various feasible and infeasible computational boundaries.
Addressing Mark Imbalance in Integration-free Marked Temporal Point Processes
Sishun Liu (RMIT University), Xiuzhen Zhang (RMIT University)
Recurrent Neural NetworkTime SeriesSequential
🎯 What it does: This study investigates the impact of label imbalance on the prediction of marked temporal point processes (MTPP), proposes a method for adjusting label probabilities based on thresholds, and develops an integral-free neural MTPP (IFNMTPP) that implements a strategy of predicting labels first and then predicting time.
ADG: Ambient Diffusion-Guided Dataset Recovery for Corruption-Robust Offline Reinforcement Learning
Zeyuan Liu (Tsinghua University), Xiu Li (Tsinghua University)
Reinforcement LearningDiffusion modelTabular
🎯 What it does: This study proposes the Ambient Diffusion-Guided Dataset Recovery (ADG) method, which utilizes diffusion models to detect, recover, and denoise trajectories with noise or damage in offline reinforcement learning, thereby enhancing the robustness of policy learning.
Adjacent Words, Divergent Intents: Jailbreaking Large Language Models via Task Concurrency
Yukun Jiang (CISPA Helmholtz Center for Information Security), Yang Zhang (CISPA Helmholtz Center for Information Security)
Adversarial AttackTransformerLarge Language ModelTextBenchmark
🎯 What it does: A word-level task concurrency-based LLM jailbreak framework called JAIL-CON is proposed, demonstrating that LLMs can still answer tasks under concurrent interactions while reducing the identification rate of security protections.
Adjoint Schrödinger Bridge Sampler
Guan-Horng Liu (Meta), Ricky T. Q. Chen (Meta)
GenerationOptimizationDrug DiscoveryGraph Neural NetworkDiffusion modelGraphPhysics RelatedStochastic Differential Equation
🎯 What it does: A novel Diffusion Sampler called the Adjoint Schrödinger Bridge Sampler (ASBS) is proposed for sampling Boltzmann distributions given only the energy function.
Adjusted Count Quantification Learning on Graphs
Clemens Damke (LMU Munich), Eyke Hüllermeier
ClassificationDomain AdaptationGraph Neural NetworkGraph
🎯 What it does: Two quantization learning methods for graph-structured data are proposed—Structural Importance Sampling (SIS) and Neighborhood-Aware ACC (NACC), and experiments are conducted under different distribution shifts.
Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models
Hyeonggeun Han (Seoul National University), Jungwoo Lee (Seoul National University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes to alleviate the memorization problem in text-to-image diffusion models by adjusting the initial noise samples and presents two mitigation strategies during inference (batch-wise and per-sample).
ADMN: A Layer-Wise Adaptive Multimodal Network for Dynamic Input Noise and Compute Resources
Jason Wu (University of California Los Angeles), Mani Srivastava (University of California Los Angeles)
ClassificationRecognitionComputational EfficiencyTransformerAuto EncoderVideoMultimodalityAudio
🎯 What it does: Proposes an ADMN network that dynamically allocates the number of computation layers based on the quality of multimodal input for each sample;
AdmTree: Compressing Lengthy Context with Adaptive Semantic Trees
Yangning Li (Tsinghua University), Philip S. Yu (University of Illinois Chicago)
RetrievalCompressionTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The AdmTree framework is proposed, achieving adaptive hierarchical context compression that significantly reduces input length while maintaining high semantic fidelity.
ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: A pre-training framework for industrial anomaly detection, ADPretrain, is proposed, utilizing large-scale industrial data RealIAD and residual features for angle and norm contrastive learning to obtain more discriminative pre-trained representations.
Adv-SSL: Adversarial Self-Supervised Representation Learning with Theoretical Guarantees
Chenguang Duan (Wuhan University), Jerry Zhijian Yang (Wuhan University)
Representation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: A new adversarial self-supervised representation learning framework, Adv-SSL, is proposed, which utilizes unbiased min-max optimization to eliminate estimation bias in traditional covariance regularization, thereby enhancing representation transfer performance.
Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization
Cong Wang (Nanjing University), Qing Gu (Nanjing University)
GenerationData SynthesisDiffusion modelAuto EncoderVideoText
🎯 What it does: Developed the SignViP framework to generate identity-preserving sign language videos from text;
Advancing Compositional Awareness in CLIP with Efficient Fine-Tuning
Amit Peleg (University of Tübingen), Matthias Hein (University of Tübingen)
RetrievalRepresentation LearningTransformerContrastive LearningImageText
🎯 What it does: The CLIC method is proposed to efficiently fine-tune CLIP to enhance its compositional reasoning ability.
Advancing Expert Specialization for Better MoE
Hongcan Guo (Beijing University of Posts and Telecommunications), Xudong Jiang (Nanyang Technological University)
Mixture of ExpertsText
🎯 What it does: A training method is proposed to enhance the specialization of Mixture-of-Experts (MoE) experts and the diversity of routing through orthogonality and variance loss.
Advancing Interpretability of CLIP Representations with Concept Surrogate Model
Nhat Hoang-Xuan (University of Florida), My T. Thai (University of Florida)
RetrievalExplainability and InterpretabilityKnowledge DistillationContrastive LearningImageTextMultimodality
🎯 What it does: By constructing a conceptual substitution model, we explain the CLIP image representations and provide the contribution of each concept to the representation.
Advancing Machine-Generated Text Detection from an Easy to Hard Supervision Perspective
Chenwang Wu (Hong Kong Baptist University), Defu Lian (University of Science and Technology of China)
ClassificationOptimizationLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A framework for easy-to-difficult supervised enhancement is proposed, which improves the performance of machine-generated text detection by constructing long-text supervisors.
Advancing Wasserstein Convergence Analysis of Score-Based Models: Insights from Discretization and Second-Order Acceleration
Yifeng Yu (Tsinghua University), Lu Yu (City University of Hong Kong)
Diffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: This paper conducts a quantitative analysis of the Wasserstein distance convergence of fractional-based diffusion models, comparing discretization schemes such as Euler, exponential integrators, and stochastic midpoint, and proposes a second-order accelerated sampler utilizing Hessian information to achieve faster convergence rates.
AdvEDM: Fine-grained Adversarial Attack against VLM-based Embodied Agents
Yichen Wang (Huazhong University of Science and Technology), Leo Yu Zhang (Griffith University)
Autonomous DrivingAdversarial AttackRobotic IntelligenceTransformerVision Language ModelImageMultimodality
🎯 What it does: A fine-grained adversarial attack framework ADVEDM is proposed for embodied decision-making systems based on visual language models (VLM), along with two variants: ADVEDM-R (removing target object semantics) and ADVEDM-A (adding target object semantics).
Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment
Xiaojun Jia (Nanyang Technological University), Yang Liu (Nanyang Technological University)
Adversarial AttackTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: A transferable attack method for closed-source multimodal large language models, called FOA-Attack, is proposed, which generates target attack samples through optimal alignment of global and local features.
Adversarial Diffusion for Robust Reinforcement Learning
Daniele Foffano (Royal Institute of Technology), Alexandre Proutiere (Royal Institute of Technology)
Reinforcement LearningDiffusion modelSequential
🎯 What it does: A robust reinforcement learning framework based on adversarial diffusion, AD-RRL, is proposed, which utilizes diffusion models to generate worst-case trajectories and enhances the robustness of the policy against model errors during training.
Adversarial generalization of unfolding (model-based) networks
Vicky Kouni (Paris Dauphine PSL Research University)
RestorationOptimizationAdversarial AttackImage
🎯 What it does: The study investigates the adversarial generalization performance of the deconvolution network (ADMM-DAD) under the l2 norm constrained FGSM adversarial attack.
Adversarial Graph Fusion for Incomplete Multi-view Semi-supervised Learning with Tensorial Imputation
Zhangqi Jiang (National University of Defense Technology), Xinyan Liang (Shanxi University)
ClassificationOptimizationGraph Neural NetworkGenerative Adversarial NetworkGraph
🎯 What it does: An algorithm AGF-TI for incomplete multi-view semi-supervised learning is proposed to address the sub-cluster problem (SCP) caused by missing views, achieving more robust label propagation through graph fusion and missing information recovery.
Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning
Jiyuan Shi (Institute of Artificial Intelligence), Xuelong Li (Institute of Artificial Intelligence)
Robotic IntelligenceTransformerReinforcement LearningTextMultimodality
🎯 What it does: This paper proposes the ALMI (Adversarial Locomotion and Motion Imitation) framework, which utilizes adversarial learning of the upper and lower limbs to achieve coordinated control of the entire body, and implements robust gait and precise motion tracking on the Unitree H1-2 robot.
Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text
Yize CHENG, Soheil Feizi (University of Maryland)
GenerationAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a training-free, general adversarial paraphrasing framework that utilizes instruction-tuned LLMs to select the most 'human-like' token at each step of generation based on the scores from AI text detectors, effectively circumventing various AI text detectors.
Adversarial Robustness of Nonparametric Regression
Parsa Moradi (University of Minnesota), Mohammad Ali Maddah-Ali (University of Minnesota)
OptimizationAdversarial Attack
🎯 What it does: This paper studies the adversarial robustness of nonparametric regression under adversarial data contamination, particularly exploring the robustness characteristics of nonparametric regression when the adversary can arbitrarily corrupt the input data.
Adversary Aware Optimization for Robust Defense
Daniel Wesego (University of Illinois Chicago), Pedram Rooshenas (University of Illinois Chicago)
OptimizationAdversarial AttackDiffusion modelScore-based ModelImage
🎯 What it does: A diffusion model-based adversarial purification framework called AAOpt is proposed, which utilizes a pre-trained diffusion prior and a learned adversarial perturbation score network for MAP optimization purification during testing.
AdvPrefix: An Objective for Nuanced LLM Jailbreaks
Sicheng Zhu (University of Maryland), Ivan Evtimov (FAIR Meta)
OptimizationAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A new prefix enforcement target called 'AdvPrefix' is proposed to achieve more fine-grained LLM jailbreaking, enhancing attack effectiveness through an automated prefix generation and filtering process.
AegisGuard: RL-Guided Adapter Tuning for TEE-Based Efficient & Secure On-Device Inference
CHE WANG, Wei Yang Bryan Lim (Nanyang Technological University)
Safty and PrivacyComputational EfficiencyTransformerSupervised Fine-TuningReinforcement LearningText
🎯 What it does: The AegisGuard framework is proposed to achieve collaborative LoRA tuning and deployment between TEE and GPU, enabling efficient and secure large model inference on the device side.
AF-UMC: An Alignment-Free Fusion Framework for Unaligned Multi-View Clustering
Bohang Sun (Beijing University of Technology), Gengyu Lyu (Beijing University of Technology)
Auto EncoderContrastive LearningImageVideo
🎯 What it does: A non-aligned multi-view clustering framework AF-UMC is proposed, which directly extracts consistent representations from each view and globally fuses them;
Affine-Invariant Global Non-Asymptotic Convergence Analysis of BFGS under Self-Concordance
Qiujiang Jin (University of Texas at Austin), Aryan Mokhtari (University of Texas at Austin)
Optimization
🎯 What it does: This paper studies the global non-asymptotic linear and superlinear convergence of BFGS under strictly convex and strongly self-concordant functions, providing explicit convergence rates.
AffordBot: 3D Fine-grained Embodied Reasoning via Multimodal Large Language Models
Xinyi Wang (University of Science and Technology of China), Na Zhao (Singapore University of Technology and Design)
Object DetectionSegmentationRobotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelMultimodalityPoint CloudChain-of-Thought
🎯 What it does: This paper proposes the Fine-grained 3D Embodied Reasoning task and constructs the AffordBot framework, utilizing multimodal large language models to accomplish instruction-driven object localization and motion prediction in point cloud scenes.
Afterburner: Reinforcement Learning Facilitates Self-Improving Code Efficiency Optimization
Mingzhe Du (Nanyang Technological University), See-Kiong Ng (National University of Singapore)
OptimizationAI Code AssistantReinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: An iterative optimization framework (Afterburner) is proposed, which continuously generates, evaluates, and provides feedback on code during inference using LLMs to improve code efficiency.
AgentAuditor: Human-level Safety and Security Evaluation for LLM Agents
Hanjun Luo (New York University), Hanan Salam (New York University)
Safty and PrivacyTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the AgentAuditor framework and ASSEBench benchmark for evaluating the safety and security of LLM agents;
AgentBreeder: Mitigating the AI Safety Risks of Multi-Agent Scaffolds via Self-Improvement
J Rosser (University of Oxford), Jakob Nicolaus Foerster
OptimizationSafty and PrivacyLarge Language ModelAgentic AIText
🎯 What it does: AGENTBREEDER is proposed, utilizing multi-objective evolutionary search to generate multi-agent scaffolds and evaluate their capabilities and safety.
Agentic Plan Caching: Test-Time Memory for Fast and Cost-Efficient LLM Agents
Qizheng Zhang (Stanford University), Kunle Olukotun (Stanford University)
OptimizationComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AITextTabularFinance RelatedRetrieval-Augmented Generation
🎯 What it does: Proposes Agentic Plan Caching (APC), which caches and reuses plan templates through keyword retrieval and lightweight model adaptation in the Plan-Act process of LLM Agents to reduce inference costs.
Agentic RL Scaling Law: Spontaneous Code Execution for Mathematical Problem Solving
Xinji Mai (Fudan University), Wenqiang Zhang (Fudan University)
Large Language ModelReinforcement LearningAgentic AIText
🎯 What it does: This study trains a foundational large language model using Zero Reward Reinforcement Learning (ZeroRL) to autonomously invoke a Python code interpreter to complete mathematical reasoning tasks.
AgentNet: Decentralized Evolutionary Coordination for LLM-based Multi-Agent Systems
Yingxuan Yang (Shanghai Jiao Tong University), Weinan Zhang (Shanghai Jiao Tong University)
Large Language ModelAgentic AITextRetrieval-Augmented Generation
🎯 What it does: A decentralized multi-agent framework called AgentNet based on Retrieval-Augmented Generation (RAG) has been developed, achieving dynamic evolution, task allocation, and collaboration of LLM agents.
Agents Robust to Distribution Shifts Learn Causal World Models Even Under Mediation
Matteo Ceriscioli (Oregon State University), Karthika Mohan (Oregon State University)
Robotic IntelligenceReinforcement LearningAgentic AIWorld ModelSequential
🎯 What it does: This paper proves that in decision-making tasks with mediation, agents that can adapt to distribution shifts must learn the causal model of their environment. It proposes an algorithm based on optimal strategy oracles (LearnCID) to extract causal structures from robust agents, thereby inferring the optimal strategies for other agents or multi-agent systems, POMDPs, and other sequential decision-making tasks.
AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Tasks
Fali Wang (Pennsylvania State University), Suhang Wang (Pennsylvania State University)
OptimizationComputational EfficiencyHyperparameter SearchTransformerLarge Language ModelAgentic AIText
🎯 What it does: This paper presents AgentTTS, an LLM-based agent framework for optimal allocation of computational budgets in multi-stage complex tasks, thereby enhancing overall task performance.
Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations
Olawale Elijah Salaudeen (Massachusetts Institute of Technology), Marzyeh Ghassemi (Massachusetts Institute of Technology)
Domain AdaptationImageBenchmark
🎯 What it does: This paper proposes an unsupervised subset selection method called OODSelect, which aims to discover hidden subsets in OOD data, revealing a negative correlation between ID and OOD accuracy, thereby uncovering the issue of spurious correlations masked by aggregated statistics.
Agnostic Active Learning Is Always Better Than Passive Learning
Steve Hanneke (Purdue University)
🎯 What it does: This paper studies the query complexity of agnostic active learning, providing both lower and upper bounds compared to passive learning, and proposes the AVID algorithm to achieve optimal query complexity.
Agnostic Continuous-Time Online Learning
Pramith Devulapalli (Purdue University), Wojciech Szpankowski (Purdue University)
OptimizationStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes an agnostic theoretical framework for continuous-time online learning and provides optimal regret upper and lower bounds in both observable and adaptive data stream scenarios.
Agnostic Learning under Targeted Poisoning: Optimal Rates and the Role of Randomness
Bogdan Chornomaz (Technion Israel Institute of Technology), Tom Waknine (Technion Israel Institute of Technology)
Adversarial Attack
🎯 What it does: This paper studies agnostic learning under target-oriented data poisoning attacks, providing the best upper and lower bounds on excess error;
Aha! - Predicting What Matters Next: Online Highlight Detection Without Looking Ahead
Aiden Chang (University of Southern California), Stephanie M. Lukin (DEVCOM Army Research Laboratory)
RecognitionOptimizationComputational EfficiencyVideo
🎯 What it does: The AHA framework is proposed, achieving efficient online, task-oriented video highlight detection.
AI Debate Aids Assessment of Controversial Claims
Salman Rahman (University of California), Saadia Gabriel (University of California)
TransformerLarge Language ModelText
🎯 What it does: The study investigates the impact of AI Debate on biased reviewers in the fact judgment tasks related to COVID-19 and climate change controversies.
AI Research Agents for Machine Learning: Search, Exploration, and Generalization in MLE-bench
Edan Toledo (Meta), Yoram Bachrach (Meta)
Large Language ModelAgentic AITabularBenchmark
🎯 What it does: This paper proposes a framework for designing AI research agents as a combination of search strategies and operators, and systematically evaluates different combinations on MLE-Bench Lite.
AI-Generated Video Detection via Perceptual Straightening
Christian Internò (Bielefeld University), David Klindt
ClassificationAnomaly DetectionTransformerContrastive LearningVideoBenchmark
🎯 What it does: An AI video detection method called ReStraV is proposed, which is based on the geometric curvature of visual representations. It utilizes a pre-trained DINOv2 visual encoder to extract representations of video frames, calculates curvature and distance statistics in the representation space as features to distinguish between real and fake videos.
AI-Researcher: Autonomous Scientific Innovation
Jiabin Tang (University of Hong Kong), Chao Huang (University of Hong Kong)
Recommendation SystemTransformerLarge Language ModelAgentic AITextBenchmarkRetrieval-Augmented Generation
🎯 What it does: The AI-Researcher framework is proposed, achieving full-process automation from literature retrieval, idea generation, algorithm implementation to automatic paper writing, and constructing the Scientist-Bench benchmark.
AiDE-Q: Synthetic Labeled Datasets Can Enhance Learning Models for Quantum Property Estimation
Xinbiao Wang (Nanyang Technological University), Dacheng Tao (Wuhan University)
Data SynthesisOptimizationSupervised Fine-TuningTabularPhysics Related
🎯 What it does: The AiDE-Q framework is proposed, which enhances the learning model for quantum property estimation by iteratively generating high-quality synthetic labeled datasets.
AION-1: Omnimodal Foundation Model for Astronomical Sciences
Liam Holden Parker, Shirley Ho (Flatiron Institute)
SegmentationGenerationRetrievalTransformerContrastive LearningImageMultimodality
🎯 What it does: AION-1 has been constructed, a large-scale multimodal foundation model capable of processing 39 different astronomical data modalities (images, spectra, annotations, numerical data, etc.), and self-supervised pre-training on 200 million celestial objects is achieved through a two-stage process.
Algorithm- and Data-Dependent Generalization Bounds for Diffusion Models
Benjamin Dupuis (INRIA), Umut Simsekli (INRIA)
GenerationDiffusion modelImage
🎯 What it does: A generalization analysis framework related to algorithms and data is proposed, which decomposes and defines the generative performance of diffusion models.
Algorithms and SQ Lower Bounds for Robustly Learning Real-valued Multi-Index Models
Ilias Diakonikolas (University of Wisconsin Madison), Lisheng Ren (University of Wisconsin Madison)
Optimization
🎯 What it does: This paper proposes a robust PAC learning algorithm for learning real-valued multi-index models (MIMs) under Gaussian distribution, and provides a statistical query (SQ) lower bound that matches the complexity of the algorithm, proving that it is nearly optimal in terms of dimensionality.
Alias-Free ViT: Fractional Shift Invariance via Linear Attention
Hagay Michaeli (Technion), Daniel Soudry (Technion)
ClassificationRecognitionTransformerImage
🎯 What it does: This paper proposes an Alias-Free Vision Transformer, which enhances the model's robustness to image translation by eliminating aliasing in downsampling and non-linearity, and implementing linear cross-covariance attention to achieve shift-equivariance for both integer and fractional shifts.
Align Your Flow: Scaling Continuous-Time Flow Map Distillation
Amirmojtaba Sabour (NVIDIA), Karsten Kreis (NVIDIA)
GenerationKnowledge DistillationDiffusion modelFlow-based ModelImage
🎯 What it does: This paper proposes Align Your Flow (AYF), a distillation method based on continuous time flow mapping, which can compress diffusion/flow models into few-step generators, supporting multi-step sampling while maintaining high quality.
Align-DA: Align Score-based Atmospheric Data Assimilation with Multiple Preferences
Jing-An Sun (Fudan University), LEI BAI
OptimizationReinforcement LearningDiffusion modelTime Series
🎯 What it does: This paper proposes a soft constraint alignment framework Align-DA based on reinforcement learning, transforming the atmospheric data assimilation problem into a preference optimization task, and adaptively adjusting the background prior using reward signals to enhance analysis quality and predictive capability.
AlignedGen: Aligning Style Across Generated Images
Jiexuan Zhang (Peking University), Jian Zhang (Peking University)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A training-free, DiT-based style-aligned image generation framework called AlignedGen is proposed, enhancing the style consistency of generated images under different text prompts.
Aligning by Misaligning: Boundary-aware Curriculum Learning for Multimodal Alignment
Hua Ye (Nanjing University), Xuan Zhang (Carnegie Mellon University)
RetrievalRepresentation LearningTransformerContrastive LearningImageVideoMultimodalityAudio
🎯 What it does: A boundary-aware curriculum learning framework BACL is proposed, which enhances multimodal alignment by dynamically sampling edge negative samples and emphasizing fine-grained mismatches in local attention loss.
Aligning Compound AI Systems via System-level DPO
Xiangwen Wang (Stanford University), Sanmi Koyejo (Stanford University)
OptimizationDiffusion modelImageMultimodality
🎯 What it does: System-level alignment of multi-component AI systems is conducted, proposing the SysDPO framework and implementing joint optimization.
Aligning Evaluation with Clinical Priorities: Calibration, Label Shift, and Error Costs
Gerardo Flores, Ashia C. Wilson (Massachusetts Institute of Technology)
ClassificationOptimizationExplainability and InterpretabilityBiomedical DataElectronic Health Records
🎯 What it does: A framework is proposed that combines calibration, label shift, and cost of error to evaluate the practical utility of clinical decision support models using adjusted logarithmic loss.
Aligning Text to Image in Diffusion Models is Easier Than You Think
Jaa-Yeon Lee (KAIST), Jong Chul Ye (KAIST)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: This paper proposes SoftREPA, which enhances the internal representation alignment between text and images by incorporating a small number of soft text tokens into a frozen pre-trained diffusion model and using contrastive learning.
Aligning Text-to-Image Diffusion Models to Human Preference by Classification
Longquan Dai (Nanjing University of Science and Technology), Jinhui Tang (Nanjing University of Science and Technology)
ClassificationGenerationReinforcement Learning from Human FeedbackDiffusion modelImage
🎯 What it does: Proposes to transform the text-to-image diffusion model alignment task into a classification problem, and achieves human preference alignment through the ABC (Alignment by Classification) framework;