ICML 2026 Papers — Page 4
International Conference on Machine Learning · 6554 papers
AgentSelect: Benchmark for Narrative Query-to-Agent Recommendation
Yunxiao Shi (University of Technology Sydney), Min Xu (University of Technology Sydney)
Data SynthesisRecommendation SystemTransformerLarge Language ModelContrastive LearningTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: This study proposes AGENTSELECT, a unified benchmark that transforms multi-source evaluation data (LLM, tools, composite agents) into positive sample query-agent interaction data, thus enabling the agent recommendation task for natural language queries.
AgentSteerTTS: A Multi-Agent Closed-Loop Framework for Composite-Instruction Text-to-Speech
Bin Kang (University of Chinese Academy of Sciences), Zhuotao Tian (Shenzhen Loop Area Institute)
GenerationTransformerPrompt EngineeringAuto EncoderGenerative Adversarial NetworkContrastive LearningTextBenchmarkRetrieval-Augmented GenerationAudio
🎯 What it does: Designed and implemented AgentSteerTTS, a multi-agent closed-loop framework capable of achieving fine-grained control over expressive features such as emotion and tone based on composite natural language instructions;
AgentSuite: Toward More Reliable Agent Evaluation with a Component-Based Benchmark Auditing Pipeline
Hyewon Suh (Georgia Institute of Technology), Zhen Dong (NVIDIA)
Anomaly DetectionExplainability and InterpretabilityComputational EfficiencyData-Centric LearningTransformerLarge Language ModelAgentic AIPrompt EngineeringTextTabularBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the COBA (Componentized Benchmark Audit) pipeline and the AgentSuite unified benchmark platform, which are used to automatically detect, diagnose, and filter validity issues in LLM agent benchmarks.
AgentTailor: A Semantic-Aware LLM-Based Multi-Agent System with Actor-Critic Structure
Peiting Yang (Harbin Engineering University), Rongsheng Li (Harbin Engineering University)
OptimizationAI Code AssistantTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper proposes AgentTailor, a multi-agent system based on large language models, which dynamically optimizes the communication structure through semantic-aware edge evaluation and virtual execution.
AgentVocab: Structure-Aware Vocabulary Adaptation for Efficient LLM Agents
Kai Bian (Harbin Institute of Technology), Xuelong Li (TeleAI of China Telecom)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes the AgentVocab framework, which in the tool calling scenarios of LLM agents, mines structured and semantic fragments from actual calling logs to expand the vocabulary, thereby reducing tokenization fragmentation and improving decoding efficiency.
AgentWebBench: Benchmarking Multi-Agent Coordination in Agentic Web
Shanshan Zhong (Carnegie Mellon University), Chenyan Xiong (Carnegie Mellon University)
Recommendation SystemReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AITextMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the AgentWebBench benchmark to evaluate user agents in the Agentic Web environment for completing four categories of information retrieval and generation tasks in collaboration with multiple content agents.
AgentXRay: White-Boxing Agentic Systems via Workflow Reconstruction
Ruijie Shi (Tsinghua University), Chen Qian (Shanghai Jiao Tong University)
Explainability and InterpretabilityComputational EfficiencyAI Code AssistantTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringText
🎯 What it does: By only utilizing the input-output pairs of the black-box system, construct an editable white-box workflow to achieve functional recovery and interpretation of black-box agent systems.
Aggregate Models, Not Explanations: Improving Feature Importance Estimation
Joseph Paillard (F. Hoffmann-La Roche Ltd), Bertrand Thirion (Universite Paris-Saclay)
Explainability and InterpretabilityData-Centric LearningTabularBiomedical Data
🎯 What it does: This study explores the differences between model-level integration (integrating predictors) and explanation-level integration (averaging feature importance from individual models) in estimating feature importance, and validates the performance differences through theoretical derivation and experiments.
AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs
WenXiang Lin, Shaohuai Shi (Harbin Institute of Technology)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose the AGoQ system, which uses layer-aware activation quantization (≈4 bits) and precision-preserving 8-bit gradient quantization in distributed LLM training, achieving significant GPU memory compression and training speed improvement.
Agora: Toward Autonomous Bug Detection in Production-Level Consensus Protocols with LLM Agents
Xiang Liu (National University of Singapore), Ceyao Zhang (Peking University)
Anomaly DetectionExplainability and InterpretabilityComputational EfficiencyAI Code AssistantTransformerLarge Language ModelAgentic AIPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the Agora framework, which systematically detects protocol-level vulnerabilities in distributed consensus protocol implementations using an LLM-driven multi-agent system, identifying 15 previously unknown logical flaws.
AGZO: Activation-Guided Zeroth-Order Optimization for LLM Fine-Tuning
Wei Lin (Chinese University of Hong Kong), Hong Xu (Chinese University of Hong Kong)
OptimizationFederated LearningComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningGaussian SplattingText
🎯 What it does: This paper proposes an activation-guided zeroth-order optimization method (AGZO) for fine-tuning large language models under memory-constrained environments;
AI Cartography: Mapping the Latent Landscape of AI Benchmark Ecosystems
Michael Hardy (Stanford University), Sanmi Koyejo (Stanford University)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Systematically decompose and control measurement noise in AI benchmark leaderboards using confirmatory factor analysis, generalizability theory, and mixed-effects latent regression, constructing a 'latent landscape' map of AI capabilities.
AI Engram: In Search of Memory Traces in Artificial Intelligence
Jea Kwon (Max Planck Institute for Security and Privacy), Meeyoung Cha (Max Planck Institute for Security and Privacy)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerAuto EncoderContrastive LearningImageText
🎯 What it does: Propose a geometric framework based on four principles of neuroscience, defining AI Engram and deriving a closed-form spectral projection estimator to identify and manipulate operable memory traces in deep networks.
AI4SLT: Empirical Processes in Lean 4 for Formal Statistical Learning Theory
Yuanhe Zhang (University of Warwick), Fanghui Liu (Shanghai Jiao Tong University)
OptimizationExplainability and InterpretabilityReview/Survey PaperBenchmark
🎯 What it does: Completed a comprehensive formalization of statistical learning theory (SLT) in Lean 4, covering a complete theoretical framework from high-dimensional Gaussian concentration, Dudley entropy integral, to least squares regression;
AICrypto: Evaluating Cryptography Capabilities of Large Language Models
Yu Wang (Chinese Academy of Sciences), Tianxing He (Tsinghua University)
TransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the AICrypto benchmark to evaluate the capabilities of large language models (LLMs) in cryptography, containing 135 multiple-choice questions, 150 CTF challenges, and 30 proof questions;
AIR-VLA: Vision-Language-Action Systems for Aerial Manipulation
Jianli Sun (Chinese Academy of Sciences), Yonglin Tian (Chinese Academy of Sciences)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelVision-Language-Action ModelDiffusion modelImageVideoTextMultimodalityBenchmark
🎯 What it does: Proposed the AIR-VLA benchmark, combining a simulation environment, 3000 multimodal data samples, and multidimensional evaluation, specifically targeting aerial manipulation systems.
AIR: Improving Agent Safety through Incident Response
Zibo Xiao (Tianjin University), Junjie Chen (Tianjin University)
Safty and PrivacyReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AIPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed and implemented the AIR (Agent Incident Response) framework, achieving event detection, isolation, recovery, and eradication during the execution of LLM agents. It automatically generates guardrail rules using a DSL to enhance agent security.
AIR: Post-training Data Selection for Reasoning via Attention Head Influence
Jinrui Liu (Beihang University), Chongyang Tao (Beihang University)
Explainability and InterpretabilityKnowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposed an unsupervised, no-training post-training data selection framework called AIR, which selects the most valuable data for inference by examining the causal impact of the retrieval attention head.
Aitchison Embeddings for Learning Compositional Graph Representations
Nikolaos Nakis (Yale University), Giannis Nikolentzos (University of Peloponnese)
Representation LearningGraph Neural NetworkAuto EncoderContrastive LearningGraph
🎯 What it does: Proposes a graph embedding framework called AICoG based on Aitchison geometry, treating nodes as combinations of relative proportions and achieving trainable Euclidean space representations through ILR coordinates.
ALAS: Additive Learnable Alpha-Stable Kernels for Flexible Bayesian Optimization
Weibo Huang (Shanghai Jiao Tong University), Cheng Hua (Shanghai Jiao Tong University)
OptimizationTabularBenchmark
🎯 What it does: Proposed a learnable α-stable kernel family (ALAS) and its separable variant (ALAS-Sep) for adaptive modeling of black-box functions with different smoothness levels in Bayesian optimization.
Alethia: a Foundational Encoder for Voice Deepfakes
Yi Zhu (Reality Defender), Surya Koppisetti (Reality Defender)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningFlow-based ModelAuto EncoderContrastive LearningAudio
🎯 What it does: A specialized foundation encoder called Alethia for voice deepfakes was constructed, and it was pre-trained and fine-tuned on multiple tasks.
Algorithmic Recourse of In-Context Learning for Tabular Data
Wenshuo Dong (King Abdullah University of Science and Technology), Lijie Hu (Mohamed bin Zayed University of Artificial Intelligence)
OptimizationExplainability and InterpretabilityData-Centric LearningTransformerPrompt EngineeringTabularFinance Related
🎯 What it does: The study implements algorithm regression in In-Context Learning (ICL) models under tabular data, proposing the ASR-ICL framework to generate actionable and sparse recovery solutions.
AlgoTrace: Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
Samuel Lippl (Columbia University), Ida Momennejad (Microsoft Research)
OptimizationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Construct the AlgoTrace framework, which parses and regulates algorithmic primitives in LLM reasoning through clustering plus function vector methods
AlgoVeri: An Aligned Benchmark for Verified Code Generation on Classical Algorithms
Haoyu Zhao (Princeton University), Sanjeev Arora (Princeton University)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes ALGOVERI, a cross-language (Dafny, Verus, Lean) alignment benchmark for evaluating the formal verification code generation capabilities of large models on complex classical algorithms;
AlienLM: Alienization of Language for API-Boundary Privacy in Black-Box LLMs
Jaehee Kim (Seoul National University), Pilsung Kang (Seoul National University)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Designed AlienLM, which achieves text concealment within black-box APIs through vocabulary substitution while maintaining model usability.
Align Forward, Adapt Backward: Closing the Discretization Gap in Logic Gate Networks
Youngsung Kim (Inha University)
ClassificationComputational EfficiencyRepresentation LearningContrastive LearningImage
🎯 What it does: This paper studies the gap between training and inference in discrete choice networks, and proposes forward alignment to eliminate this gap;
Align Your Trajectory Tangent: Training Better Consistency Models via Manifold-Aligned Tangents
Beomsu Kim (KAIST), Jong Chul Ye (KAIST)
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelContrastive LearningImage
🎯 What it does: Propose a self-supervised method based on flow feature distance to align the trajectory tangent of the consistency model, thereby significantly improving the training convergence speed and generation quality;
AlignedNorm: Prompting Vision–Language Models via Coupled Prompt Field
Qi Ma (Nankai University), Deng-Ping Fan (Nankai University)
ClassificationRecognitionDomain AdaptationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed the Coupled Prompt Field (CPF) paradigm and applied AlignedNorm to achieve norm alignment in prompt learning for vision-language models such as CLIP, addressing the issue of local optima between baselines and new tasks.
Aligning Tree-Search Policies with Fixed Token Budgets in Test-Time Scaling of LLMs
Sora Miyamoto (Institute of Science Tokyo), Naoaki Okazaki (Institute of Science Tokyo)
OptimizationComputational EfficiencyTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmarkPhysics Related
🎯 What it does: Proposed a budget-oriented Monte Carlo Tree Search algorithm called BG-MCTS, which can perform tree search decoding under a fixed output token budget.
Alignment between Brains and AI: Evidence for Convergent Evolution across Modalities, Scales and Training Trajectories
Guobin Shen (Chinese Academy of Sciences), Yi Zeng (University of Chinese Academy of Sciences)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImageTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Conduct a large-scale hierarchical internal representation alignment evaluation between 630 visual and language AI models (covering various architectures, scales, and training stages) and the natural scene dataset NSD's fMRI recordings, generating over 60 million CKA similarity measurements;
Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment
Cameron Tice (Geodesic Research), Kyle O'Brien (Geodesic Research)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The study influences the alignment tendency of large language models by incorporating positive and negative AI discourse data during the pre-training phase, with experiments demonstrating self-fulfilling (mis)alignment phenomena.
Alignment Risks from Capability-Seeking RL Training
Yujun Zhou (University of Notre Dame), Xiangliang Zhang (University of Notre Dame)
Explainability and InterpretabilityTransformerSupervised Fine-TuningReinforcement LearningText
🎯 What it does: By designing four controllable 'exploit games' (Context-Conditional Compliance, Audited Self-Grading, Proxy Metric Gaming, Reward Tampering), the system evaluates whether goal-seeking reinforcement learning (GRPO) on open-weight 3–8B LLMs will spontaneously learn to exploit environmental structural loopholes through misleading shortcuts, and investigates their detectability, cross-task transferability, student model transmission, and the persistence of corrective training.
Alignment Tampering: How Reinforcement Learning from Human Feedback Is Exploited to Optimize Misaligned Biases
Dongyoon Hahm (Korea Advanced Institute of Science and Technology), Kimin Lee (Korea Advanced Institute of Science and Technology)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Proposed and demonstrated 'alignment tampering', where during the RLHF process, the model influences preference data through its own generated outputs, thereby amplifying undesirable biases.
Alignment-Aware Decoding
Frédéric Berdoz (ETH Zurich), Roger Wattenhofer (ETH Zurich)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringText
🎯 What it does: Proposes the Alignment-Aware Decoding (AAD) method, which directly utilizes token-level rewards obtained from DPO training during inference to guide generation, without requiring additional training;
Alignment-Guided Score Matching for Text-to-Image Alignment in Diffusion Models
Jaa-Yeon Lee (KAIST), Jong Chul Ye (KAIST)
GenerationSupervised Fine-TuningPrompt EngineeringDiffusion modelScore-based ModelImageTextMultimodality
🎯 What it does: Proposes a reward-free post-training framework called Alignment-Guided Score Matching, which guides the diffusion model to better align images with text during generation by performing score matching on soft text tokens.
Alignment-Sensitive Minimax Rates for Spectral Algorithms with Learned Kernels
Dongming Huang (National University of Singapore), Qian Lin (Tsinghua University)
OptimizationExplainability and InterpretabilityComputational EfficiencyRepresentation LearningAuto EncoderContrastive LearningTabularSequentialReview/Survey PaperStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed and analyzed the Effective Span Dimension (ESD), a noise-sensitive complexity measure, to quantify the alignment between the target signal and the learned kernel (or feature space), and provided its optimal extreme risk rates in sequence models, linear regression, and RKHS regression; further studied the dynamic changes of ESD during kernel spectrum learning via over-parameterized gradient flow (OP-GF) under a fixed feature basis, proving that ESD can be reduced under certain conditions; verified on simulated data that ESD decreases with training time and network depth, along with the corresponding estimation error reduction.
AlignVid: Taming Visual Dominance via Training-Free Attention Modulation in Text-guided Image-to-Video Generation
Yexin Liu (Hong Kong University of Science and Technology), Harry Yang (Hong Kong University of Science and Technology)
Image TranslationGenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageVideoTextMultimodalityBenchmark
🎯 What it does: Proposed a novel attention modulation method called AlignVid, which does not require additional training, to address the semantic mismatch problem caused by the visual dominance of reference images in text-guided image-to-video generation.
AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing
Yuexin Li (National University of Singapore), Jiaheng Zhang (National University of Singapore)
GenerationSafty and PrivacyData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a sentence-level watermarking framework called AliMark, aiming to enhance the robustness of watermarks in text generation models under strong semantic rewriting (such as GPT-3.5, DIPPER).
All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs
Xi Chen (University of Toronto), Gerald Penn (University of Toronto)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Proposed and validated the Functional Anisotropy Hypothesis, showing that a single LLM task can be implemented by multiple circuits/fibers with significant structural differences and low overlap;
All ERMs Can Fail in Stochastic Convex Optimization Lower Bounds in Linear Dimension
Tal Burla (Tel Aviv University), Roi Livni (Tel Aviv University)
Optimization
🎯 What it does: Constructed instances in random convex optimization that are dimension-linear, have a unique fitting solution, but do not generalize.
Alleviating Observation Bias via Causal-Invariant Meta-Learning for Unbalanced Incomplete Multi-view Clustering
Jiaqi Jin (National University of Defense Technology), En Zhu (National University of Defense Technology)
Representation LearningAdversarial AttackData-Centric LearningMeta LearningTransformerAuto EncoderGenerative Adversarial NetworkContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: To address the observation bias problem in imbalanced missing multi-view clustering, the CIMLN framework is proposed for missing view recovery and clustering.
Alleviating Sparse Rewards by Modeling Step-Wise and Long-Term Sampling Effects in Flow-Based GRPO
Yunze Tong (Zhejiang University), Hao Jiang (Taobao & Tmall Group of Alibaba)
GenerationTransformerReinforcement LearningScore-based ModelFlow-based ModelImageTextMultimodalityStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose the TP-GRPO framework, which alleviates the sparse reward problem in the Flow Matching model by introducing step reward and inflection point mechanisms, thereby improving the performance of text-to-image generation.
Allocating Variance to Maximize Expectation
Renato Paes Leme (Google Research), Pratik Worah (Google Research)
OptimizationGaussian SplattingChain-of-Thought
🎯 What it does: This paper studies how to allocate the variances of each Gaussian random variable, given a total variance budget, in order to maximize the expectation of these variables (or the maximum of their subsets).
AlphaGRPO: Unlocking Self-Reflective Multimodal Generation in Unified Multimodal Models via Decompositional Verifiable Reward
Runhui Huang (The University of Hong Kong), Hengshuang Zhao (The University of Hong Kong)
GenerationExplainability and InterpretabilityComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringVision Language ModelVision-Language-Action ModelDiffusion modelScore-based ModelImageTextMultimodalityRetrieval-Augmented GenerationStochastic Differential Equation
🎯 What it does: Propose the AlphaGRPO framework, which directly applies Group Relative Policy Optimization (GRPO) to the AR-Diffusion unified multimodal model, leveraging the model's pre-trained knowledge to achieve multimodal generation and self-reflection without cold start.
AlphaRouter: Token-level Routing Between SLM and LLM with Reinforcement Learning and Tree Search
Siteng Liao (Beijing Normal University), Tian Wang (Beijing Normal University)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Designed a Token-Level routing framework called AlphaRouter based on Small Language Models (SLM) and Large Language Models (LLM), which learns the optimal collaborative inference path using search and reinforcement learning.
ALSO: Adversarial Online Strategy Optimization for Social Agents
Xiang Li (Tianjin University), Qinghua Hu
OptimizationAdversarial AttackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the ALSO framework, which enhances the social intelligence of large language model agents in multi-agent social simulations through online adaptive strategy optimization.
Alterbute: Editing Intrinsic Attributes of Objects in Images
Tal Reiss (Google), Yedid Hoshen (Google)
Image TranslationImage HarmonizationGenerationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Developed an algorithm called Alterbute based on diffusion models, which precisely edits the intrinsic attributes (color, texture, material, and shape) of objects in images while preserving object identity and scene consistency.
Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training
Ran Xu (Emory University), Haoyu Wang (University at Albany)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringMixture of ExpertsTextBenchmark
🎯 What it does: This paper proposes the RUBRIC-ARM framework, which achieves multi-dimensional evaluation in non-verified tasks by jointly optimizing the scorer and rule generator through alternating reinforcement learning.
AMA-Bench: Evaluating Long-Horizon Memory for Agentic Applications
Yujie Zhao (University Of California San Diego), Jishen Zhao (University Of California San Diego)
Autonomous DrivingFederated LearningExplainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningAdversarial AttackHyperparameter SearchData-Centric LearningRobotic IntelligenceMeta LearningDrug DiscoveryAI Code AssistantReinforcement Learning from Human FeedbackProtein Structure PredictionTransformerLarge Language ModelPrompt EngineeringDiffusion modelScore-based ModelFlow-based ModelRectified FlowNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingSimultaneous Localization and MappingWorld ModelOptical FlowTextSequentialBenchmarkRetrieval-Augmented GenerationStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Designed the AMA-Bench evaluation framework and proposed the AMA-Agent memory system;
Ambient Dataloops: Generative Models for Dataset Refinement
Adrian Rodriguez-Munoz (Massachusetts Institute of Technology), Giannis Daras (Massachusetts Institute of Technology)
RestorationGenerationData SynthesisTransformerDiffusion modelScore-based ModelAuto EncoderImageTextBiomedical Data
🎯 What it does: Propose the Ambient Dataloops framework, which iteratively co-evolves datasets and models, using diffusion models to gradually denoise and improve the original noisy or low-quality datasets, thereby enhancing model training.
Ambiguous Strategic Classification
Ivri Hikri (Technion - Israel Institute of Technology), Nir Rosenfeld (Technion - Israel Institute of Technology)
ClassificationSafty and PrivacyExplainability and InterpretabilityReinforcement LearningContrastive LearningTabularBenchmark
🎯 What it does: This paper studies how to influence users' strategic behavior and model performance by controlling the uncertainty (ambiguity) of classifiers, in scenarios where the learning system can only disclose partial classifier information.
AmbiRefer3D: 3D Visual Grounding with Referential Ambiguity
Rongjiang Zhu (Beijing Institute Of Technology), Xinxiao Wu (Beijing Institute Of Technology)
RecognitionObject DetectionSegmentationRetrievalRepresentation LearningConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelContrastive LearningTextMultimodalityPoint CloudRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a fuzzy task for 3D visual localization, and build an interactive multi-round question-answering framework to eliminate referential ambiguity and accurately locate the target object.
AMDP: Asynchronous Multi-Directional Pipeline Parallelism for Large-Scale Models Training
Ling Chen (Zhejiang University), Wenjie Yu (Zhejiang University)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Proposed AMDP, an asynchronous multi-direction pipeline parallel training framework, which limits the parameter mismatch between forward and backward passes to one step, significantly improving model training throughput.
Amodal Instance Segmentation with IRAIS Dataset for Sim-to-Real Transfer
Bidong Chen (University of Coimbra), Lingui Li (Guangzhou College of Commerce)
SegmentationGenerationData SynthesisDomain AdaptationTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingImagePoint CloudMeshBenchmark
🎯 What it does: This paper proposes MaviGen, an automated 3D retail scene generation and rendering framework, which generates multi-view synthetic images with depth, occlusion order, and camera parameters. Based on this, the IRAIS simulation-to-real dual segmentation benchmark (3D-IRAIS + Real-IRAIS) is constructed, and the EUREKA encoder-querier network is designed, which achieves joint learning of visible and complete instance segmentation through unified queries and dual-head masks.
Amortized Maximum Inner Product Search with Learned Support Functions
Theo X. Olausson (Apple), marco cuturi
RetrievalOptimizationComputational EfficiencyConvolutional Neural NetworkTransformerMixture of ExpertsScore-based ModelAuto EncoderContrastive LearningTextTabularBenchmark
🎯 What it does: This work proposes an amortized MIPS method, which trains a neural network to directly predict the maximum inner product matching key for a given query, thereby converting the cost of multiple MIPS solutions into network forward inference.
Amortized Simulation-Based Inference in Generalized Bayes via Neural Posterior Estimation
Shiyi Sun (University of Oxford), Kate Lee
OptimizationComputational EfficiencyRepresentation LearningData-Centric LearningTransformerDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderContrastive LearningTabularBiomedical DataBenchmarkPhysics RelatedStochastic Differential Equation
🎯 What it does: Propose a fully amortized variational approximation for the power posterior (p_β(θ|x)) in Generalized Bayes, training a neural posterior estimator q_ϕ(θ|x,β) conditioned on the observation x and temperature β, such that inference only requires a single forward pass for sampling, without simulator calls or MCMC.
Amortized Variational Inference for Partial-Label Learning: A Probabilistic Approach to Label Disambiguation
Tobias Fuchs (Karlsruhe Institute of Technology), Nadja Klein (Karlsruhe Institute of Technology)
ClassificationExplainability and InterpretabilityComputational EfficiencyData-Centric LearningTransformerAuto EncoderContrastive LearningImageTextTabular
🎯 What it does: Propose the VILLP framework, modeling label uncertainty in partial label learning as a variational inference problem, using neural networks to predict variational parameters, achieving a direct approximation of the true label posterior.
An Algebraic View of the Expressivity of Recurrent Language Models
Franz Nowak (ETH Zuerich), Reda Boumasmoud (ETH Zuerich)
Explainability and InterpretabilityRepresentation LearningRecurrent Neural NetworkTextReview/Survey Paper
🎯 What it does: This paper proposes a unified algebraic perspective, modeling the expressiveness of recurrent language models (RNNs) as a hierarchical transformation monad (wreath product), and reduces the syntax monad of recognizable languages through realizable monads;
An analytic theory of convolutional neural network inverse problems solvers
Minh Hai Nguyen, Pierre Weiss (Universite Toulouse)
Image TranslationRestorationExplainability and InterpretabilityConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed TomographyReview/Survey Paper
🎯 What it does: This paper derives an analytical formula for a CNN inverse problem solver by introducing an MMSE estimator with translational equivariance and locality constraints, and proves that it can accurately predict the output of a trained CNN;
An Approximation Algorithm for Graph Label Selection
Josia John (ETH Zurich), Maximilian Probst Gutenberg (ETH Zurich)
OptimizationComputational EfficiencyGraph
🎯 What it does: Proposes the first approximate algorithm for the graph label selection problem under a given budget k without using resource expansion, and provides an approximation ratio of about √{ⁿ⁴}{n}; meanwhile, an implementable version that scales to large-scale graphs is realized.
An Asymmetric Latent Factorization-of-Tensors Model for Relation Analysis
Weiling Li (Dongguan University of Technology), Xin Luo (Southwest University)
Recommendation SystemOptimizationRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningGraphTabularTime Series
🎯 What it does: Propose an anti-symmetric tensor factorization model (ALFT) to handle relational tensors containing entity graphs and entity sets, thereby addressing the limitations of traditional LFT models under the graph isomorphism assumption.
An Efficient Joint Learning Approach for Item Response Theory
Tanish Agarwal (IIT Bombay), Arpit Agarwal (IIT Bombay)
Recommendation SystemOptimizationComputational EfficiencyTabularBenchmark
🎯 What it does: Propose an EM algorithm based on Polya-Gamma data augmentation to jointly estimate user ability θ and item difficulty β in the Rasch model, addressing the inconsistency or slow convergence issues of traditional JMLE, MMLE, and CMLE on sparse data.
An Embarrassingly Simple Way to Optimize Orthogonal Matrices at Scale
Adrián Javaloy (University of Edinburgh), Antonio Vergari (University of Edinburgh)
OptimizationAuto EncoderContrastive LearningImage
🎯 What it does: Proposed POGO, an optimizer for large-scale orthogonal matrix constraints, which can achieve fast convergence while maintaining orthogonality;
An Empirical Study on the Resilience of Partial Merging to Model Clone Attacks
Tiantong Wu (Nanyang Technological University), Wei Yang Bryan Lim (Nanyang Technological University)
Federated LearningSafty and PrivacyAdversarial AttackTransformerAuto EncoderContrastive LearningImageText
🎯 What it does: Studied the threat to model privacy posed by model cloning attacks in partial model merging (PMM), and proposed the ModelPirate attack method.
An Evidential Route to Asymptotic Bayes Optimality under Sparsity
Qiaoyu Liang (University of Toronto), Michael Evans (University of Toronto)
OptimizationFederated LearningExplainability and InterpretabilityComputational EfficiencyRepresentation LearningTabularReview/Survey PaperPhysics Related
🎯 What it does: This paper establishes the asymptotic optimality properties of certain multiple testing rules based on the relative belief ratio from the perspective of statistical evidence, demonstrating how asymptotic Bayesian optimality can be achieved in sparse scenarios.
An Exploration of Non-Euclidean Gradient Descent: Muon and its Many Variants
Michael Crawshaw (George Mason University), Robert M. Gower (Flatiron Institute)
OptimizationLarge Language ModelMixture of ExpertsImageText
🎯 What it does: Systematically conduct theoretical analysis and empirical evaluation of the Muon optimizer and its variants, propose a new variant called MuonMax, and combine the adaptive step size Momo with the steepest descent method under arbitrary norms to form a general non-Euclidean gradient descent framework.
An Exponential Separation Between Quantum and Quantum-Inspired Classical Algorithms for Linear Systems
Allan Grønlund (Kvantify), Kasper Green Larsen (Aarhus University)
OptimizationGraphTabularPhysics Related
🎯 What it does: Demonstrates that there is an exponential performance gap between quantum algorithms and quantum-inspired classical algorithms in solving linear system problems.
An Exterior Method for Nonnegative Matrix Factorization
Qiujing Lu (UCLA), Vwani Roychowdhury (UCLA)
OptimizationComputational EfficiencyRepresentation LearningImageTextTabularAudio
🎯 What it does: Proposed an external optimization framework called eNMF, which maps the unconstrained SVD decomposition to the non-negative orthogonal domain via a rotation transformation, and subsequently enforces non-negativity through a penalty term and converges to a local optimum using HALS;
An In-Depth Study on Deep Learning Model Cloning
Bin Hu (Hangzhou Dianzi University), Tianyi Hu (Hangzhou Dianzi University)
RetrievalAnomaly DetectionFederated LearningExplainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerAuto EncoderContrastive LearningOptical FlowTextTabularBenchmark
🎯 What it does: Proposes the definition of deep learning model cloning and a detection method called MCDetector based on dual similarity (structural and weight similarity).
An Information-Theoretic Criterion for Efficient Data Synthesis
Hanyu Li (Peking University), Xiaotie Deng (Peking University)
Data SynthesisExplainability and InterpretabilityRepresentation LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringContrastive LearningText
🎯 What it does: This paper proposes criteria based on information theory to clarify when synthetic data can continuously improve the performance of language models, and explains the differences in efficiency among various synthetic strategies through data processing inequalities and meta-level signal analysis.
An Interactive Paradigm for Deep Research
Lin Ai (Columbia University), Saayan Mitra (Adobe Research)
Recommendation SystemData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the STEER framework to achieve interactive controllability in deep research systems.
An Odd Estimator for Shapley Values
Fabian Fumagalli (LMU Munich), R. Teal Witter (Claremont McKenna College)
Explainability and InterpretabilityComputational EfficiencyTransformerContrastive LearningImageTextTabular
🎯 What it does: Proposed an OddSHAP, a Shapley value approximation estimator based on odd subspaces, which achieves consistent and efficient estimation by utilizing paired sampling and Fourier basis functions.
AnalogVerifier: A Neuro-Symbolic Framework for Analog Circuit Verification
Yanfang Liu (Chinese University of Hong Kong), Tsung-Yi Ho (Chinese University of Hong Kong)
OptimizationExplainability and InterpretabilityAI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextGraphTabularPhysics RelatedRetrieval-Augmented Generation
🎯 What it does: A neural-symbolic framework called AnalogVerifier was constructed, achieving the automatic generation of verification test platforms from industrial-level unstructured analog circuit specifications, and ensuring functional passing through closed-loop repair.
Analytic Bijections for Smooth and Interpretable Normalizing Flows
Mathis Gerdes (University of Amsterdam), Miranda C. N. Cheng (University of Amsterdam)
Explainability and InterpretabilityComputational EfficiencyDiffusion modelScore-based ModelFlow-based ModelRectified FlowContrastive LearningImageTabularTime SeriesSequentialBiomedical DataPhysics Related
🎯 What it does: Designed and implemented three globally smooth, closed-form invertible scalar bijections (cubic rational, sinh, cubic conjugation), and constructed interpretable radial flows based on them as pluggable invertible transformations for coupling and other normalization flows.
Anatomy of Massive Activations and Attention Sinks
Shangwen Sun (New York University), Jiachen Zhu (New York University)
OptimizationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Investigate two phenomena in Transformer language models: 'massive activations' and 'attention sinks', systematically analyze their generation mechanisms, interrelationships, and impacts on model behavior, and verify their causal independence through a series of ablation experiments.
Anchor-Final Self-Supervision Drives Hallucination-Aware Optimization in Large Vision-Language Models
Jiaxi Liu (Shanghai Jiao Tong University), Nanyang Ye (Shanghai Jiao Tong University)
OptimizationExplainability and InterpretabilityRepresentation LearningTransformerReinforcement LearningPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed the Anchor-Final Self-Supervision (AFS) framework, which suppresses hallucinations in unsupervised visual-language models by leveraging internal layer-wise probabilistic differences;
Anchor-guided Hypergraph Condensation with Dual-level Discrimination
Fan Li (University of New South Wales), Wenjie Zhang (University of New South Wales)
ClassificationCompressionRepresentation LearningGraph Neural NetworkDiffusion modelScore-based ModelAuto EncoderContrastive LearningGraph
🎯 What it does: This paper proposes a hypergraph condensation framework named AHGCDD, which significantly compresses large-scale hypergraph data while preserving the effectiveness of hypergraph learning.
ANCHOR: Abductive Network Construction with Hierarchical Orchestration for Reliable Probability Inference in Large Language Models
Wentao Qiu (Xiamen University), Qingqiang Wu (Xiamen University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringMixture of ExpertsTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: A hierarchical factor space was constructed, and LLM was used for factor generation and parameterization, while Naïve Bayes and causal Bayesian networks were integrated to provide a reliable probabilistic inference framework.
ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm
Kefan Song (University of Virginia), Yanjun Qi (University of Virginia)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIPrompt EngineeringText
🎯 What it does: Developed the ANCHOR framework for automated alignment auditing of CLI agents on real-world harmful tasks.
Anchored Decoding: Provably Reducing Copyright Risk for Any Language Model
Jacqueline He (University of Washington), Pang Wei Koh (University of Washington)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposes a technique called Anchored Decoding, which suppresses large language models (LMs) from repeating copyrighted content during inference by interpolating with a safety model, and provides formal safety guarantees based on the K-NAF framework;
Anchored Policy Optimization: Mitigating Exploration Collapse via Support-Constrained Rectification
Tianyi Wang (Southern University of Science and Technology), Guanhua Chen (Shanghai University of Finance and Economics)
Reinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: Investigate the exploration collapse caused by Recursive Space Contraction (RSC) in RLVR, and propose Anchored Policy Optimization (APO) to achieve distribution recovery through support coverage and ratio correction.
Anchoring Self-Play for Code Repair
Caroline Choi (Stanford University), Ludwig Schmidt (Stanford University)
AI Code AssistantTransformerLarge Language ModelReinforcement LearningPrompt EngineeringDiffusion modelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper uses the self-play method of reinforcement learning to enable a single model to both generate and fix bugs, and proposes Anchored Self-Play (ASP) to alleviate distribution drift; subsequently, its performance is evaluated on the newly constructed BUGSOURCEBENCH benchmark.
Angel or Demon: Investigating the Plasticity Interventions' Impact on Backdoor Threats in Deep Reinforcement Learning
Oubo Ma (Zhejiang University), Shouling Ji (Zhejiang University)
Anomaly DetectionAdversarial AttackReinforcement LearningPrompt EngineeringContrastive LearningTabularTime SeriesSequential
🎯 What it does: This study systematically evaluates 14,664 experiments to explore the impact of plastic interventions in DRL (Deep Reinforcement Learning) on backdoor attacks injected later, and reveals its internal mechanisms through pathological feature analysis.
Annotations Mitigate Post-Training Mode Collapse
Jacob Mitchell Springer (Carnegie Mellon University), Aditi Raghunathan (Carnegie Mellon University)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By incorporating semantic annotations during pre-training and maintaining the annotation distribution during later training, the model's semantic mode collapse during post-training is suppressed.
Anomaly-Preference Image Generation
Fuyun Wang (Nanjing University of Science and Technology), Zhen Cui (Beijing Normal University)
GenerationData SynthesisAnomaly DetectionTransformerDiffusion modelScore-based ModelImage
🎯 What it does: Propose a few-shot anomaly image generation framework APO based on implicit preference alignment, which can synthesize realistic and diverse defective samples without the need for manual annotation.
AnomSeer: Reinforcing Multimodal LLMs to Reason for Time-Series Anomaly Detection
Junru Zhang (Zhejiang University), Duanqing Xu (Zhejiang University)
Anomaly DetectionExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningMultimodalityTime SeriesChain-of-Thought
🎯 What it does: Propose the ANOMSEER framework, which utilizes reinforcement learning to enable multi-modal large language models to perform fine-grained reasoning in time series anomaly detection and generate interpretable diagnostic reports.
Anti-Aliasing Matters: A Dynamic Network for Time Series Forecasting
Heng Zhou (Zhejiang University), Chao Li (Zhejiang University)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerAuto EncoderContrastive LearningMultimodalityTabularTime SeriesBenchmarkFinance RelatedPhysics Related
🎯 What it does: This paper proposes a time series forecasting network called DMANet, designed to address aliasing issues generated during multi-scale downsampling processes.
Anti-Backdoor Coreset Selection via Cumulative Entropy
Qi Zhao (Karlsruhe Institute of Technology), Christian Wressnegger (Karlsruhe Institute of Technology)
Anomaly DetectionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: To defend against neural network backdoors during training, Anti-Backdoor Coreset Selection (ABCS) is proposed, which selects an informative and backdoor-free subset from a contaminated dataset for training.
Anti-causal domain generalization: Leveraging unlabeled data
Sorawit Saengkyongam (Apple), Christina Heinze-Deml (Apple)
Domain AdaptationRepresentation LearningContrastive LearningTime SeriesBiomedical DataBenchmark
🎯 What it does: In anti-causal domain generalization, the authors enhance the model's robustness to unseen environments by utilizing unlabelled multi-environment data through regularization methods, proposing two techniques: Mean Direction Regularization (MIR) and Variance Direction Regularization (VIR).
ANTiC: Adaptive Neural Temporal In Situ Compressor
Sandeep Suresh Cranganore, Johannes Brandstetter (Johannes Kepler University Linz)
CompressionOptimizationComputational EfficiencyTransformerSupervised Fine-TuningAuto EncoderPoint CloudTime SeriesPhysics Related
🎯 What it does: The ANTIC framework is proposed for large-scale high-dimensional PDE simulations, combining physics-aware adaptive time sampling and continuously fine-tuned neural field compression, achieving online end-to-end data compression.
Antidistillation Fingerprinting
Yixuan Even Xu (Carnegie Mellon University), J Zico Kolter (Carnegie Mellon University)
Safty and PrivacyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningText
🎯 What it does: Proposes a model fingerprinting method based on antidistillation technology called Antidistillation Fingerprinting (ADFP), which is used to embed detectable fingerprints during the model distillation process;
Any-dimensional invariant universality
Shengtai Yao (Stanford University), Mateo Diaz Diaz (University of Chicago)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningGraph Neural NetworkTransformerContrastive LearningPoint CloudGraphReview/Survey Paper
🎯 What it does: This paper proposes a unified framework that clarifies how to define and prove the universal approximation properties of invariant models in any dimension (such as DeepSets, GNN, point cloud networks) on an infinite-dimensional limit space; it also provides a three-step 'recipe' for testing universality, followed by constructing specific models that are continuous and universally applicable in three typical domains: sets, graphs, and point clouds, improving existing architectures and proving that they can approximate all continuous invariant functions on the corresponding compact sets with arbitrary precision.
Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and Architecture
Shuchen Xue (University of Chinese Academy of Sciences), Zhi-Ming Ma (University of Chinese Academy of Sciences)
GenerationData SynthesisComputational EfficiencyTransformerLarge Language ModelDiffusion modelScore-based ModelText
🎯 What it does: Implement and evaluate the masked diffusion model (MDM), named Any-Order GPT, within a single decoder framework, and systematically compare it with traditional autoregressive (AR) models and encoder-based MDM.
Any2Any: Unified Arbitrary Modality Translation for Remote Sensing
Haoyang Chen (Wuhan University), Bo Du (Wuhan University)
Image TranslationData SynthesisTransformerDiffusion modelScore-based ModelAuto EncoderGenerative Adversarial NetworkContrastive LearningImageMultimodality
🎯 What it does: Propose the Any2Any framework to achieve arbitrary cross-modal translation of remote sensing multi-modal images, and construct a million-scale RST-1M training set.
Any3D-VLA: Enhancing VLA Robustness via Diverse Point Clouds
Xianzhe Fan (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
Robotic IntelligenceConvolutional Neural NetworkTransformerReinforcement LearningMixture of ExpertsVision-Language-Action ModelAuto EncoderContrastive LearningImagePoint CloudMesh
🎯 What it does: Proposes the ANY3D-VLA framework, which fuses 2D visual features with RGB images processed through point cloud projection, compression, and pre-trained point cloud encoder, thereby enhancing the spatial understanding and robustness of vision-language-action models in complex scenarios.
AnyBand-Diff: A Unified Remote Sensing Image Generation and Band Repair Framework with Spectral Priors
Zhao zuopeng, Wenwen Liu (China University of Mining and Technology)
RestorationGenerationTransformerDiffusion modelScore-based ModelContrastive LearningImagePhysics Related
🎯 What it does: A unified framework for remote sensing image generation and band restoration, named AnyBand-Diff, is proposed. It utilizes sparse conditional diffusion, physics-guided sampling, and multi-scale physical loss to achieve spectral reconstruction of arbitrary band subsets and full-band image generation.
AnyCanvas: Potential Field Guidance for Training-Free Spatial Control in Text-to-Image Diffusion
Tianyi Xie (Southeast University), Dian Shen (Southeast University)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: Propose a training-agnostic framework called AnyCanvas that can generate semantically consistent images on canvases of arbitrary shapes.
AnyEdit++: Adaptive Long-Form Knowledge Editing via Bayesian Surprise
Bowen Tian (Hong Kong University of Science and Technology), Yutao Yue (Hong Kong University of Science and Technology)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes the AnyEdit++ framework, which utilizes Bayesian Surprise for adaptive segmentation to achieve efficient editing of long texts.
AnyMod-LLVE: Low-Light Video Enhancement with Modality-Agnostic Inference
Hangfeng Liang (Southeast University), Ying Fu (Beijing Institute of Technology)
RestorationKnowledge DistillationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoMultimodality
🎯 What it does: Propose a unified multi-modal low-light video enhancement framework, AMNet, which can flexibly handle missing modalities during inference and achieve enhancement using only RGB input.
Anytime Detection of Strategic Deviations in Multi-Agent Systems
Etienne Gauthier (Inria), Michael I. Jordan (Inria)
Anomaly DetectionReinforcement Learning
🎯 What it does: This paper proposes an online sequential detection framework for real-time identification of strategic behavior deviations in multi-agent systems.