AAAI 2024 Papers — Page 15
AAAI Conference on Artificial Intelligence · 2331 papers
Multimodal Graph Neural Architecture Search under Distribution Shifts
Jie Cai (Tsinghua University), Wenwu Zhu (Tsinghua University)
Domain AdaptationNeural Architecture SearchGraph Neural NetworkReinforcement LearningMultimodalityGraph
🎯 What it does: To address the issue of distribution drift in multimodal graph data, a multimodal graph neural network architecture search method named OMG-NAS is proposed, which automatically searches for MGNNs with excellent out-of-distribution (OOD) generalization capabilities.
Multiobjective Lipschitz Bandits under Lexicographic Ordering
Bo Xue (City University of Hong Kong), Qingfu Zhang (University of Waterloo)
Optimization
🎯 What it does: This paper studies the multi-objective bandit problem under lexicographic order, proposing a new multi-objective Lipschitz bandit model and developing an improved algorithm to simultaneously maximize multiple objectives.
Multiple Hypothesis Dropout: Estimating the Parameters of Multi-Modal Output Distributions
David D. Nguyen (CSIRO Data61 Cybersecurity CRC), Surya Nepal (CSIRO Data61 Cybersecurity CRC)
GenerationData SynthesisTransformerGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes Multiple Hypothesis Dropout (MH Dropout), which transforms a single-output network into a multi-output network and learns the mean and variance of the multimodal output distribution through a random winner-take-all loss.
Multiscale Attention Wavelet Neural Operator for Capturing Steep Trajectories in Biochemical Systems
Jiayang Su (Guangxi Normal University), Minghan Chen (Wake Forest University)
Biomedical DataOrdinary Differential Equation
🎯 What it does: A multi-scale attention wavelet neural operator (MAWNO) is proposed for accurately simulating the nonlinear differential equations of biochemical systems with steep trajectories, using a physics-driven training approach.
Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks
Fuzhi Wu (Southeast University), Lotfi Senhadji (Univ Rennes)
ClassificationSegmentationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a Multi-Scale Low-Frequency Memory Network (MLFM), which maintains the original structure of CNN while inserting Low-Frequency Memory Units (LFMU) at different scales to preserve and interact with low-frequency information, thereby enhancing the performance of CNN in tasks such as image classification and semantic segmentation.
MULTISCRIPT: Multimodal Script Learning for Supporting Open Domain Everyday Tasks
Jingyuan Qi (Virginia Tech), Lifu Huang (Virginia Tech)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: The MULTISCRIPT benchmark is proposed, defining two tasks: multimodal script generation and subsequent step prediction, and providing 6,655 task videos and text data based on WikiHow.
MultiSum: A Multi-Facet Approach for Extractive Social Summarization Utilizing Semantic and Sociological Relationships
Tanglong Zhao (Tianjin University), Bo Wang (Tianjin University)
Graph Neural NetworkText
🎯 What it does: The MultiSum model is proposed, which uses Graph Convolutional Networks (GCN) to integrate social and semantic relationships, achieving unsupervised extractive social text summarization.
Multitarget Device-Free Localization via Cross-Domain Wi-Fi RSS Training Data and Attentional Prior Fusion
Na Fan (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
Transformer
🎯 What it does: Using Wi-Fi RSS signals, an end-to-end Transformer with attention prior fusion module is proposed for multi-target device-free localization.
MusER: Musical Element-Based Regularization for Generating Symbolic Music with Emotion
Shulei Ji (Xi'an Jiaotong University), Xinyu Yang (Xi'an Jiaotong University)
GenerationTransformerAuto EncoderAudio
🎯 What it does: This paper presents MusER, a music element decoupling and emotion generation framework based on VQ-VAE, which can separate elements such as pitch, rhythm, and dynamics in the latent space and control the emotion of generated music through element transfer.
Music Style Transfer with Time-Varying Inversion of Diffusion Models
Sifei Li (Institute of Automation Chinese Academy of Sciences), Changsheng Xu (Institute of Automation Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelAudio
🎯 What it does: A music style transfer method based on diffusion models and time-varying text inversion has been developed, capable of transferring any audio (instruments, natural sounds, synthetic effects) into the target music with very few samples.
MuST: Robust Image Watermarking for Multi-Source Tracing
Guanjie Wang (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
RecognitionData SynthesisConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A multi-source watermark tracking method MuST is proposed to identify and extract copyright information from multi-source synthesized images.
Mutual-Modality Adversarial Attack with Semantic Perturbation
Jingwen Ye (National University of Singapore), Xinchao Wang (National University of Singapore)
Adversarial AttackConvolutional Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: Utilizing the visual-text alignment space of the pre-trained CLIP, a cross-modal adversarial attack and defense framework is proposed, generating semantic perturbations on visual inputs and dynamically updating them in text prompts to restore matching.
MWSIS: Multimodal Weakly Supervised Instance Segmentation with 2D Box Annotations for Autonomous Driving
Guangfeng Jiang (University of Science and Technology of China), Pai Peng (COWAROBOT)
SegmentationAutonomous DrivingKnowledge DistillationImageMultimodalityPoint Cloud
🎯 What it does: Using only 2D bounding box annotations, we jointly train an instance segmentation model for images and LiDAR, and enhance segmentation quality through a multi-modal pseudo-label generation and correction module.
N-gram Unsupervised Compoundation and Feature Injection for Better Symbolic Music Understanding
Jinhao Tian (Wuhan University), Ping Wang (Wuhan University)
ClassificationGenerationRecurrent Neural NetworkTransformerSequentialAudio
🎯 What it does: The NG-Midiformer model is proposed, combining unsupervised composite tokenization and N-gram Transformer encoding to achieve a better understanding of symbolic music sequences.
NaMa: Neighbor-Aware Multi-Modal Adaptive Learning for Prostate Tumor Segmentation on Anisotropic MR Images
Runqi Meng (ShanghaiTech University), Dinggang Shen (Shanghai United Imaging Intelligence Co., Ltd.)
SegmentationConvolutional Neural NetworkImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper proposes a Neighborhood-Aware Multimodal Adaptive Learning Network (NaMa) for accurate segmentation of prostate tumors in multimodal heterogeneous MRI.
Narrowing the Gap between Supervised and Unsupervised Sentence Representation Learning with Large Language Model
Mingxin Li (Beihang University), Yongyi Mao (University of Ottawa)
Representation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: This paper conducts detailed experiments on the training process of supervised and unsupervised contrastive sentence embedding (CSE) and finds that the complexity of similarity patterns determines the performance gap; subsequently, it utilizes the contextual learning ability of large language models to generate data with complex similarity patterns and introduces hierarchical triplet loss to fully leverage the hierarchical scoring of STS data, ultimately significantly narrowing the performance gap between supervised and unsupervised CSE.
NaRuto: Automatically Acquiring Planning Models from Narrative Texts
Ruiqi Li (Australian National University), Patrik Haslum (Australian National University)
TransformerLarge Language ModelText
🎯 What it does: A fully automated, unsupervised system called NaRuto is proposed for extracting planning action models from narrative texts.
Natural Strategic Ability in Stochastic Multi-Agent Systems
Raphaël Berthon (RWTH Aachen University), Aniello Murano (University of Naples Federico II)
Reinforcement LearningAgentic AI
🎯 What it does: This paper proposes the definitions of probabilistic ATL and ATL* using natural strategies in stochastic multi-agent systems (i.e., Nat PATL and Nat PATL*), and analyzes the complexity of their model checking.
NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models
Gengze Zhou (University of Adelaide), Qi Wu (University of Adelaide)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposes NavGPT, a zero-shot visual-language navigation agent based on large language models, which completes instruction following by transforming visual perception into natural language and performing explicit reasoning with LLM.
Navigating Open Set Scenarios for Skeleton-Based Action Recognition
Kunyu Peng (Karlsruhe Institute of Technology), Alina Roitberg (University of Stuttgart)
RecognitionGraph Neural NetworkTransformerVideoMultimodalityBenchmark
🎯 What it does: A benchmark for Open Set Skeleton Action Recognition (OS-SAR) has been established to evaluate existing open set methods and propose the CrossMax method to improve the open set recognition performance of skeleton actions.
Navigating Real-World Partial Label Learning: Unveiling Fine-Grained Images with Attributes
Haoran Jiang (University of Chinese Academy of Sciences), YingJie Tian
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: A framework for partial label learning on fine-grained images (PLL-FG) called SoDisam is proposed, which utilizes shared attribute learning for visual representation and achieves discrimination through attribute attention.
ND-MRM: Neuronal Diversity Inspired Multisensory Recognition Model
Qixin Wang (Beijing Normal University), Xia Wu (Beijing Normal University)
RecognitionSpiking Neural NetworkMultimodalityAudio
🎯 What it does: A multi-sensory recognition model inspired by neural diversity (ND-MRM) is proposed, which can achieve cross-sensory interaction through simulating the single-sensory and multi-sensory response characteristics of neurons, significantly enhancing multi-sensory emotion recognition performance.
Near-Optimal Resilient Aggregation Rules for Distributed Learning Using 1-Center and 1-Mean Clustering with Outliers
Yuhao Yi (Sichuan University), Jiancheng Lv (Sichuan University)
Anomaly DetectionFederated LearningImage
🎯 What it does: A robust aggregation rule based on a 1-center/1-mean clustering (including outliers) approximation algorithm is proposed, and a two-phase voting framework (2PRASHB) is designed to enhance the robustness of distributed learning under Byzantine attacks.
Nearly Equitable Allocations beyond Additivity and Monotonicity
Siddharth Barman (Indian Institute of Science), Soumyajit Pyne (Tata Institute of Fundamental Research)
Optimization
🎯 What it does: The study investigates fair allocation under non-additive and non-monotonic conditions, proving the existence and algorithmic implementation of the EQx scheme.
NeBLa: Neural Beer-Lambert for 3D Reconstruction of Oral Structures from Panoramic Radiographs
Sihwa Park (Korea University), Seung Jun Baek (Korea University)
SegmentationGenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImagePoint CloudComputed Tomography
🎯 What it does: This paper proposes the NeBLa model, which can reconstruct the 3D structure of the oral cavity using only a single panoramic X-ray image.
Negative Pre-aware for Noisy Cross-Modal Matching
Xu Zhang (University of Electronic Science and Technology of China), Mang Ye (Wuhan University)
RetrievalTransformerContrastive LearningMultimodality
🎯 What it does: A Negative Pre-aware Cross-modal Matching (NPC) method is proposed, which improves the noise robustness of cross-modal matching by pre-evaluating the negative impact of each sample and assigning adaptive confidence weights to the samples, and fine-tuning on CLIP ViT‑B/32.
NegVSR: Augmenting Negatives for Generalized Noise Modeling in Real-world Video Super-Resolution
Yexing Song (Guangdong University of Technology), Yukai Shi (Guangdong University of Technology)
RestorationSuper ResolutionVideo
🎯 What it does: The NegVSR model is proposed, which improves the robustness of real video super-resolution by utilizing sequence noise extraction, negative sample enhancement, and augmented negative guided loss.
Neighborhood-Enhanced 3D Human Pose Estimation with Monocular LiDAR in Long-Range Outdoor Scenes
Jingyi Zhang (Xiamen University), Cheng Wang (Xiamen University)
Pose EstimationTransformerPoint Cloud
🎯 What it does: In large-scale outdoor scenes, 3D human pose estimation has been improved by utilizing point clouds collected from monocular LiDAR, combined with spatial and structural consistency information from background neighborhood (3BN) and scanning neighborhood (3SN);
NeRF-LiDAR: Generating Realistic LiDAR Point Clouds with Neural Radiance Fields
Junge Zhang (Fudan University), Li Zhang (Fudan University)
SegmentationGenerationAutonomous DrivingNeural Radiance FieldImagePoint Cloud
🎯 What it does: This paper proposes NeRF-LiDAR, which utilizes NeRF for implicit reconstruction of real driving scenes to generate realistic LiDAR point clouds and semantic labels.
NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning
Linsheng Chen (Sun Yat-sen University), Philip H.S. Torr (University of Oxford)
GenerationData SynthesisDepth EstimationPrompt EngineeringNeural Radiance FieldImage
🎯 What it does: The NeRF-VPT method is proposed, which utilizes View Prompt Tuning with cascading view prompts to enhance the view synthesis quality of NeRF by using the RGB output from the previous rendering as a visual prompt in multiple stages.
NestE: Modeling Nested Relational Structures for Knowledge Graph Reasoning
Bo Xiong (University of Stuttgart), Steffen Staab (Griffith University)
Graph
🎯 What it does: This paper proposes a new knowledge graph embedding method called NestE, which is designed to simultaneously capture the semantic relationships between atomic facts (triples) and nested facts.
NESTER: An Adaptive Neurosymbolic Method for Causal Effect Estimation
Abbavaram Gowtham Reddy (Indian Institute of Technology Hyderabad), Vineeth N Balasubramanian (Indian Institute of Technology Hyderabad)
Tabular
🎯 What it does: An adaptive neuro-symbolic method called NESTER is proposed for estimating causal effects from observational data.
NeSyFOLD: A Framework for Interpretable Image Classification
Parth Padalkar (University of Texas at Dallas), Gopal Gupta (University of Texas at Dallas)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes the NeSyFOLD framework, which replaces the later layers of CNN with a set of Answer Set Programming (ASP) rules, capable of generating globally interpretable rule sets and using these rule sets to make predictions, forming an interpretable neuro-symbolic model.
Neural Amortized Inference for Nested Multi-Agent Reasoning
Kunal Jha (Dartmouth College), Tianmin Shu (Johns Hopkins University)
Autonomous DrivingReinforcement LearningSequential
🎯 What it does: This paper studies an I-POMDP method based on neural amortized inference for high-order (up to level 2) multi-agent social reasoning.
Neural Causal Abstractions
Kevin Xia (Columbia University), Elias Bareinboim (Columbia University)
Data SynthesisExplainability and InterpretabilityRepresentation LearningAuto EncoderGenerative Adversarial NetworkImageTabular
🎯 What it does: This paper proposes a neural causal abstraction framework based on 'abstract functions (τ)', utilizing interpretable variable clustering (inter-/intravariable) to construct high-level causal models, and provides theoretical tools such as hierarchical consistency (Q_τ-consistency) and abstract conditions (AIC) for inference.
Neural Embeddings for kNN Search in Biological Sequence
Zhihao Chang (Zhejiang University), Wentao Hu (Zhejiang University)
RetrievalRepresentation LearningConvolutional Neural NetworkBiomedical Data
🎯 What it does: An end-to-end embedding framework for biological sequence kNN search, called Bio-kNN, is proposed.
Neural Gaussian Similarity Modeling for Differential Graph Structure Learning
Xiaolong Fan (Xidian University), Jieyi Liu (Xidian University)
OptimizationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: A differentiable graph structure learning framework is proposed, using a neural Gaussian similarity model to sample neighbors and reduce complexity through a transition graph.
Neural Network Approximation for Pessimistic Offline Reinforcement Learning
Di Wu (Wuhan University), Xiliang Lu (Wuhan University)
Reinforcement Learning
🎯 What it does: In offline reinforcement learning, the authors propose an adversarial framework using deep neural networks for policy and value function approximation, and provide a non-asymptotic upper bound on estimation error under conditions of geometrical C-mixing and partial coverage of data.
Neural Network Approximators for Marginal MAP in Probabilistic Circuits
Shivvrat Arya (University of Texas at Dallas), Vibhav Gogate (University of Texas at Dallas)
Tabular
🎯 What it does: This paper proposes a self-supervised learning-based neural network approximator for approximating the marginal maximum a posteriori (MMAP) inference problem in differentiable probabilistic circuits (PC).
Neural Oscillators for Generalization of Physics-Informed Machine Learning
Taniya Kapoor (Delft University of Technology), Rolf Dollevoet (Delft University of Technology)
Recurrent Neural NetworkTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: In Physics-Informed Machine Learning (PIML), by combining neural oscillators (CoRNN, LEM) with Physics-Informed Neural Networks (PINN), the PINN is first trained within the training domain, and its output is then used as a time series input to the oscillator, achieving accurate extrapolation to unseen time domains (or parameter domains);
Neural Physical Simulation with Multi-Resolution Hash Grid Encoding
Haoxiang Wang (Tsinghua University), Qionghai Dai (Tsinghua University)
OptimizationComputational EfficiencyMeshPhysics RelatedOrdinary Differential Equation
🎯 What it does: A meshless physical simulation framework based on multi-resolution hash grid encoding is constructed to solve spatiotemporal partial differential equations (such as elasticity, transport, and incompressible flow), achieving efficient approximation by combining numerical gradients with dynamic geometric sampling.
Neural Reasoning about Agents’ Goals, Preferences, and Actions
Matteo Bortoletto (University of Stuttgart), Andreas Bulling (University of Stuttgart)
Graph Neural NetworkTransformerSequentialBenchmark
🎯 What it does: A neural network model named IRENE is proposed, which intuitively infers the goals, preferences, and behaviors of agents from sequences of observed actions, achieving new optimal performance on the Baby Intuitions Benchmark (BIB);
Neural Time-Reversed Generalized Riccati Equation
Alessandro Betti (Inria), Matteo Tiezzi (University of Siena)
OptimizationRecurrent Neural NetworkTime SeriesOrdinary Differential Equation
🎯 What it does: This paper proposes a forward-time neural network optimization method that estimates the cost function's costate by using the time-reversed generalized Riccati equation in continuous-time control problems, achieving optimal control learning with only forward propagation.
Neuromorphic Event Signal-Driven Network for Video De-raining
Chengjie Ge (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationConvolutional Neural NetworkSpiking Neural NetworkVideo
🎯 What it does: Combining the event stream generated by the event camera as dynamic prior, a deep unfolding network is proposed to achieve video de-raining.
NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views
Han Huang (Tsinghua University), Yu-Shen Liu (Tsinghua University)
GenerationOptimizationNeural Radiance FieldPoint Cloud
🎯 What it does: The NeuSurf framework is proposed, utilizing a sparse viewpoint surface point prior to achieve sparse viewpoint 3D surface reconstruction.
New Classes of the Greedy-Applicable Arm Feature Distributions in the Sparse Linear Bandit Problem
Koji Ichikawa (NEC Corporation), Ken-ichi Kawarabayashi (National Institute of Informatics)
OptimizationReinforcement Learning from Human FeedbackTabular
🎯 What it does: In the context of sparse linear bandits, the author analyzes how the greedy strategy can still ensure sample diversity under a broader distribution of arm characteristics, thus obtaining a sublinear upper bound on returns that is of the same order as known greedy algorithms.
NightRain: Nighttime Video Deraining via Adaptive-Rain-Removal and Adaptive-Correction
Beibei Lin (National University of Singapore), Robby T. Tan (National University of Singapore)
RestorationDomain AdaptationKnowledge DistillationTransformerDiffusion modelVideo
🎯 What it does: This paper proposes the NightRain method, which uses adaptive rain removal and adaptive correction dual processes to remove rain from real nighttime rain videos, addressing the domain gap issue between synthetic data and real scenes.
NILUT: Conditional Neural Implicit 3D Lookup Tables for Image Enhancement
Marcos V. Conde (University of Wurzburg), Radu Timofte (York University)
Computational EfficiencyImage
🎯 What it does: An implicit 3D LUT (NILUT) that can run efficiently on mobile devices and a condition version (CNILUT) with multiple styles are proposed for image enhancement and color adjustment.
NN-Steiner: A Mixed Neural-Algorithmic Approach for the Rectilinear Steiner Minimum Tree Problem
Andrew B. Kahng (University of California San Diego), Chien-Yi Yang (University of California San Diego)
OptimizationPoint Cloud
🎯 What it does: A new hybrid neural algorithm framework called NN-Steiner is proposed for solving the Rectilinear Steiner Minimum Tree (RSMT) problem, which combines Arora's PTAS algorithm.
No Head Left Behind – Multi-Head Alignment Distillation for Transformers
Tianyang Zhao (Amazon Web Services AI Labs), Ying Nian Wu (Amazon Web Services AI Labs)
Knowledge DistillationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes an Attention Map Alignment Distillation (AMAD) method that enables knowledge distillation when the number of attention heads in the teacher and student Transformers does not match, and experiments are conducted in visual language tasks.
No Internal Regret with Non-convex Loss Functions
Dravyansh Sharma (Carnegie Mellon University)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: This paper studies the problem of minimizing internal regret in continuous action spaces and proposes an algorithm that achieves sublinear internal regret in both full information and partial feedback settings.
No More Shortcuts: Realizing the Potential of Temporal Self-Supervision
Ishan Rajendrakumar Dave (University of Central Florida), Mubarak Shah (Adobe Research)
RecognitionRetrievalTransformerContrastive LearningVideo
🎯 What it does: Proposes a frame-level temporal self-supervised task (OFL and TSP) and frame-independent augmentation to train a video Transformer to learn high-level temporal features.
No Prejudice! Fair Federated Graph Neural Networks for Personalized Recommendation
Nimesh Agrawal (Indian Institute of Technology), Jayadeva (Indian Institute of Technology)
Recommendation SystemFederated LearningSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: A fair federated graph neural network framework F2PGNN is proposed, achieving fair personalized recommendations for user groups in a federated learning environment while protecting the privacy of users' sensitive attributes.
No Prior Mask: Eliminate Redundant Action for Deep Reinforcement Learning
Dianyu Zhong (Tsinghua University), Qianchuan Zhao (Tsinghua University)
Reinforcement Learning
🎯 What it does: A framework called No Prior Mask (NPM) is proposed, which can filter redundant actions in the action space without prior knowledge. It dynamically generates state-related action masks by learning the similarity factors between actions, reducing redundant exploration in reinforcement learning (RL).
NodeMixup: Tackling Under-Reaching for Graph Neural Networks
Weigang Lu (Xidian University), Long Jin (Xidian University)
Graph Neural NetworkGraph
🎯 What it does: The NodeMixup method is proposed, which enhances the reachability of GNNs and alleviates the under-reaching problem by performing mixup at the node level between labeled nodes and pseudo-labeled unlabeled nodes.
Noise-Aware Image Captioning with Progressively Exploring Mismatched Words
Zhongtian Fu (Nanjing University of Science and Technology), Yang Yang (Nanjing University of Science and Technology)
RecognitionGenerationRecurrent Neural NetworkTransformerImageText
🎯 What it does: A fine-grained word-level denoising framework for noisy image-text pairs, NIC (Noise-aware Image Captioning), is proposed. It adaptively identifies and corrects erroneous words, making the image captioning model more robust on noisy data.
Noise-Free Optimization in Early Training Steps for Image Super-resolution
MinKyu Lee (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
Super ResolutionOptimizationKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper addresses the issue of traditional pixel-level L1 loss neglect in single image super-resolution (SISR) training by proposing to decompose the target HR image into 'desired centroid' and 'inherent noise', and estimating the centroid based on a pre-trained network to achieve a noise-independent optimization objective.
Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation
Zhuohang Dang (Xi'an Jiaotong University), Jingdong Wang (Baidu Inc)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a framework called SREM to address the noisy correspondence problem in cross-modal retrieval.
Non-excludable Bilateral Trade between Groups
Yixuan Even Xu (Tsinghua University), Vincent Conitzer (Carnegie Mellon University)
Tabular
🎯 What it does: This paper studies the non-excludable bilateral trade problem where both buyers and sellers have multiple self-interested members, and all members share the trading results, and designs a corresponding trading mechanism.
Non-exemplar Domain Incremental Object Detection via Learning Domain Bias
Xiang Song (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
Object DetectionDomain AdaptationAutonomous DrivingTransformerImage
🎯 What it does: A sample-free domain incremental object detection method is proposed, utilizing a frozen baseline model to achieve continuous learning and generalization by learning the domain bias of each new domain.
Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement
Fushuo Huo (Hong Kong Polytechnic University), Yunfeng Fan (Hong Kong Polytechnic University)
ClassificationOptimizationContrastive LearningImage
🎯 What it does: For the single-channel online class incremental learning (NO-CL) scenario without example buffers, we propose the Dual-Prototype Self-Augment and Refinement (DSR) method, which utilizes dual prototypes and bi-level optimization to achieve the retention of base class knowledge and efficient learning of new classes.
Non-flat ABA Is an Instance of Bipolar Argumentation
Markus Ulbricht (Leipzig University), Francesca Toni (Imperial College London)
🎯 What it does: This paper transforms the non-flat assumption-based argumentation framework (ABAF) into a bipolar argumentation framework (BAF) and proposes a new BAF semantics to achieve this transformation; it then introduces premise-augmented BAF (pBAF) to fully capture the acceptability semantics in ABAF, and finally analyzes the computational complexity of these frameworks.
Non-monotone Sequential Submodular Maximization
Shaojie Tang (University of Texas at Dallas), Jing Yuan (University of North Texas)
Recommendation SystemOptimizationVideo
🎯 What it does: This study investigates the problem of maximizing non-monotonic sequence submodular functions, proposing a sampling-greedy algorithm to achieve approximate solutions for both flexible and fixed lengths, and validating the effectiveness in a video recommendation scenario.
Non-parametric Representation Learning with Kernels
Pascal Esser (Technical University of Munich), Debarghya Ghoshdastidar (Technical University of Munich)
Representation LearningAuto EncoderContrastive LearningImage
🎯 What it does: This paper studies kernel-based unsupervised/self-supervised representation learning methods, proposing kernel contrastive learning (simple and spectral contrast) and kernel autoencoders, and provides their representation theorems and generalization error bounds.
Non-stationary Projection-Free Online Learning with Dynamic and Adaptive Regret Guarantees
Yibo Wang (Nanjing University), Lijun Zhang (Zhejiang University)
Recommendation SystemOptimizationTabular
🎯 What it does: This paper studies non-stationary projection-free online learning, proposing dynamic regret and adaptive regret as performance metrics, and provides a new dynamic regret analysis for the existing projection-free method BOGD IP.
NondBREM: Nondeterministic Offline Reinforcement Learning for Large-Scale Order Dispatching
Hongbo Zhang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
OptimizationReinforcement LearningTabular
🎯 What it does: A nondeterministic order scheduling framework based on offline reinforcement learning, NondBREM, has been developed for real-time order dispatching in large-scale ride-hailing platforms.
Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models
Liang Li (Meituan), Xiangxiang Chu (Meituan)
CompressionOptimizationTransformerLarge Language ModelText
🎯 What it does: A technique called Norm Tweaking is proposed for low-bit quantization of large language models without sacrificing accuracy, aimed at improving the performance of quantized models.
Not All Tasks Are Equally Difficult: Multi-Task Deep Reinforcement Learning with Dynamic Depth Routing
Jinmin He (Institute of Automation, Chinese Academy of Sciences), Jian Cheng (University of Chinese Academy of Sciences)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A dynamic depth routing framework D2R is proposed to dynamically select the number of modules based on task difficulty in multi-task reinforcement learning.
Novel Class Discovery in Chest X-rays via Paired Images and Text
Jiaying Zhou (Peking University), Qingchao Chen (Peking University)
ClassificationSegmentationAnomaly DetectionContrastive LearningImageTextMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A new multimodal novel category discovery method based on paired chest X-ray images and text is proposed, utilizing two networks to encode images and text respectively, and evaluating the quality of pseudo-labels through internal consistency, thereby generating weighted synthetic pseudo-labels for joint training.
Novelty vs. Potential Heuristics: A Comparison of Hardness Measures for Satisficing Planning
Simon Dold (University of Basel), Malte Helmert (University of Basel)
🎯 What it does: This paper compares two types of metrics for measuring task difficulty in classical planning—novelty width (o/eNW) and correlation complexity (CC)—and proposes a new metric called River Measure (RM).
Null Space Matters: Range-Null Decomposition for Consistent Multi-Contrast MRI Reconstruction
Jiacheng Chen (Zhejiang University of Technology), Jianwei Zheng (Zhejiang University of Technology)
RestorationConvolutional Neural NetworkTransformerImageMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: In the multi-contrast MRI reconstruction task, a deep unfolding network (Range-Null Empowered Unfolding Network, RNU) is proposed that utilizes Range-Null Decomposition and Correlation Decoupling Learning Block. It achieves data consistency by performing proximal mapping only in the null space and separates low-frequency global (isotropy) and high-frequency local (anisotropy) features in multi-modal fusion, ultimately achieving higher quality MRI reconstruction.
NuScenes-QA: A Multi-Modal Visual Question Answering Benchmark for Autonomous Driving Scenario
Tianwen Qian (Fudan University), Yu-Gang Jiang (Fudan University)
Autonomous DrivingRecurrent Neural NetworkVision Language ModelImageTextMultimodalityPoint CloudBenchmark
🎯 What it does: This study investigates the visual question answering (VQA) task in autonomous driving scenarios, constructing a large-scale dataset called NuScenes-QA that includes images, LiDAR, and multi-frame information, and proposes a baseline model for evaluation.
O^2-Recon: Completing 3D Reconstruction of Occluded Objects in the Scene with a Pre-trained 2D Diffusion Model
Yubin Hu (Tsinghua University), Yong-Jin Liu (Beijing Jiaotong University)
RestorationGenerationDiffusion modelImage
🎯 What it does: A framework named O-Recon is proposed for image inpainting of occluded regions using a pre-trained 2D diffusion model, and complete 3D object reconstruction is achieved based on the repaired images and depth information utilizing neural implicit surfaces.
Object Attribute Matters in Visual Question Answering
Peize Li (Jilin University), Yan Wang (Jilin University)
Knowledge DistillationRepresentation LearningGraph Neural NetworkTransformerContrastive LearningImageMultimodality
🎯 What it does: A visual question answering model OAM-VQA is proposed, which utilizes object attributes to achieve visual and language alignment.
Object-Aware Adaptive-Positivity Learning for Audio-Visual Question Answering
Zhangbin Li (Hefei University of Technology), Meng Wang (Hefei University of Technology)
RecognitionObject DetectionTransformerContrastive LearningTextMultimodalityAudio
🎯 What it does: This paper proposes an audio-visual question answering model based on fine-grained visual objects, and enhances the understanding and utilization of cross-modal information through an adaptive positive sample learning strategy.
Object-Aware Domain Generalization for Object Detection
Wooju Lee (Korea Advanced Institute of Science and Technology), Hyun Myung (Korea Advanced Institute of Science and Technology)
Object DetectionDomain AdaptationContrastive LearningImage
🎯 What it does: A single-source domain generalization object detection method OA-DG is proposed, which includes OA-Mix data augmentation and OA-Loss contrastive learning.
Occluded Person Re-identification via Saliency-Guided Patch Transfer
Lei Tan (Xiamen University), Liujuan Cao (Xiamen University)
RecognitionRetrievalTransformerImage
🎯 What it does: This paper proposes a saliency-guided patch transfer (SPT) strategy for generating occluded samples, which enhances the robustness of occluded face re-identification by dynamically selecting identity and occlusion patches on a visual transformer and transferring real occluded objects to the target image.
OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning
Fan Wu (Institute of Software, Chinese Academy of Sciences), Ling Li (Institute of Software, Chinese Academy of Sciences)
Reinforcement LearningTabular
🎯 What it does: Developed the OCEAN plugin, which addresses the issue of biased exploration in model-based offline RL by incorporating a conservative exploration mechanism during the rollout phase.
OctOcc: High-Resolution 3D Occupancy Prediction with Octree
Wenzhe Ouyang (Harbin Institute of Technology), Zenglin Xu (Huawei Noah's Ark Lab)
Object DetectionSegmentationAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Designed and implemented a high-resolution 3D occupancy prediction network called OctOcc based on octrees, which maps image features from multiple cameras to 3D space and outputs the semantic occupancy state of each voxel.
ODTrack: Online Dense Temporal Token Learning for Visual Tracking
Yaozong Zheng (Guangxi Normal University), Xianxian Li (Wuzhou University)
Object TrackingTransformerVideo
🎯 What it does: A video-level object tracking framework called ODTrack is proposed, which achieves dense associations between consecutive frames in video streams using an online token propagation method.
Offline and Online Optical Flow Enhancement for Deep Video Compression
Chuanbo Tang (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
CompressionOptimizationSupervised Fine-TuningOptical FlowVideo
🎯 What it does: To address the issue of insufficient robustness of optical flow motion information used in deep video compression, an optical flow enhancement method combining offline fine-tuning and online gradient optimization is proposed, making the optical flow more suitable for video compression needs.
Offline Model-Based Optimization via Policy-Guided Gradient Search
Yassine Chemingui (Washington State University), Janardhan Rao Doppa (Washington State University)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes a framework that treats offline black-box optimization as offline reinforcement learning, where the learning policy determines the step size for gradient search, guided by a surrogate model.
Omni-Kernel Network for Image Restoration
Yuning Cui (Shenzhen Campus of Sun Yat-sen University), Alois Knoll (Technical University of Munich)
RestorationConvolutional Neural NetworkImage
🎯 What it does: An efficient convolutional network named OKNet is proposed for image restoration tasks such as dehazing, snow removal, and defocus correction.
Omnidirectional Image Super-resolution via Bi-projection Fusion
Jiangang Wang (Shenzhen Campus of Sun Yat-sen University), Xiaochun Cao (Shenzhen Campus of Sun Yat-sen University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: This paper proposes a dual-projection panoramic image super-resolution network (BPOSR), which utilizes both equirectangular projection (ERP) and cubic mapping projection (CMP) to enhance the super-resolution quality of 360° images.
Omnipotent Distillation with LLMs for Weakly-Supervised Natural Language Video Localization: When Divergence Meets Consistency
Peijun Bao (Nanyang Technological University), Alex C. Kot (Peking University)
RetrievalKnowledge DistillationTransformerLarge Language ModelVideoText
🎯 What it does: This paper proposes the OmniD algorithm, which utilizes large language models for knowledge distillation in weakly supervised natural language video localization, enhancing the model's adaptability to diverse queries.
On Alternating-Time Temporal Logic, Hyperproperties, and Strategy Sharing
Raven Beutner (CISPA Helmholtz Center for Information Security), Bernd Finkbeiner (CISPA Helmholtz Center for Information Security)
🎯 What it does: This paper proposes a new multi-agent system temporal logic called HyperATL∗S, which can compare multiple execution paths' hyperproperties in a single formula while enforcing multiple agents to share the same strategy, thus achieving previously unexpressed requirements for strategy comparison and sharing.
On Computing Makespan-Optimal Solutions for Generalized Sliding-Tile Puzzles
Marcus Gozon (University of Michigan), Jingjin Yu (Rutgers University)
Optimization
🎯 What it does: This paper studies the optimal solution for the shortest completion time of the Generalized Sliding Tile Puzzle (GSTP), proves its NP-completeness, and provides tight lower/upper bounds on the minimum/maximum completion time for any number of escorts. Furthermore, it proposes a multi-queue row/column shuffling algorithm to achieve near-optimal multi-threaded scheduling.
On Disentanglement of Asymmetrical Knowledge Transfer for Modality-Task Agnostic Federated Learning
Jiayi Chen (University of Virginia), Aidong Zhang (University of Virginia)
Federated LearningKnowledge DistillationMixture of ExpertsMultimodality
🎯 What it does: This paper studies a federated learning framework for multimodal tasks, called AFL, which achieves positive knowledge transfer among heterogeneous clients through knowledge decoupling.
On Estimating the Gradient of the Expected Information Gain in Bayesian Experimental Design
Ziqiao Ao (University of Birmingham), Jinglai Li
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes two gradient-based methods for estimating expected information gain (EIG) (UEEG-MCMC and BEEG-AP) for gradient optimization in Bayesian experimental design.
On Inference Stability for Diffusion Models
Viet Nguyen (VinAI Research), Toan Tran (VinAI Research)
GenerationDiffusion modelImage
🎯 What it does: A new sequence-aware loss function is proposed to reduce the estimation gap between predicted trajectories and actual trajectories in diffusion models, thereby improving image generation quality.
On Optimal Tradeoffs between EFX and Nash Welfare
Michal Feldman (Tel Aviv University), Tomasz Ponitka
Optimization
🎯 What it does: Under additive or sub-additive preferences, a multi-stage algorithm is proposed that can achieve a Nash welfare approximation of 1/(α+1) while ensuring approximate EFX fairness.
On Partial Optimal Transport: Revising the Infeasibility of Sinkhorn and Efficient Gradient Methods
Anh Duc Nguyen (National University of Singapore), Kim-Chuan Toh (National University of Singapore)
OptimizationPoint Cloud
🎯 What it does: The research addresses and solves the infeasibility problem of Partial Optimal Transport (POT) in large-scale issues, proposing efficient algorithms such as Sinkhorn, APDAGD, and Dual Extrapolation.
On the Affinity, Rationality, and Diversity of Hierarchical Topic Modeling
Xiaobao Wu (Nanyang Technological University), Anh Tuan Luu (Nanyang Technological University)
Auto EncoderText
🎯 What it does: A hierarchical topic model called TraCo is proposed, which is based on optimal transport plans and context-aware decoupling, to generate topic hierarchies with high affinity, coherence, and diversity.
On the Computational Complexity of Plan Verification, (Bounded) Plan-Optimality Verification, and Bounded Plan Existence
Songtuan Lin (Australian National University), Pascal Bercher (University of Basel)
Optimization
🎯 What it does: This paper systematically studies the computational complexity of reasoning tasks such as plan verification, existence of finite-length plans, and optimality verification in classical planning and Hierarchical Task Network (HTN) planning (including both grounded and lifted representations). It provides various forms of upper and lower bounds (NP-complete, PSPACE-hard, NEXPTIME-complete, etc.) and reveals the complexity differences under binary and unary encodings.
On the Convergence of an Adaptive Momentum Method for Adversarial Attacks
Sheng Long (National University of Defense Technology), Jun Zhang (National University of Defense Technology)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: An adaptive momentum and adaptive step size adversarial attack algorithm, AdaMSI-FGM, is designed to achieve theoretical convergence guarantees for adversarial sample generation.
On the Expressivity of Recurrent Neural Cascades
Nadezda Alexandrovna Knorozova (RelationalAI), Alessandro Ronca (University of Oxford)
Recurrent Neural Network
🎯 What it does: This paper conducts a theoretical analysis of the expressive power of the acyclic recursive neural cascade network (RNC), proving that when using symbolic or hyperbolic sine activation with positive weights, the RNC can only recognize star-free regular languages; it also demonstrates how to extend the RNC to recognize all regular languages by constructing flip-flop neurons and second-order neurons.
On the Outcome Equivalence of Extensive-Form and Behavioral Correlated Equilibria
Brian Hu Zhang (Carnegie Mellon University), Tuomas Sandholm (Carnegie Mellon University)
Optimization
🎯 What it does: The paper presents a proof of 'result equivalence' between Generalized Form Correlated Equilibrium (EFCE) and Behaviorally Correlated Equilibrium (BCE) in generalized game trees, and provides a polynomial-time algorithm to convert any given EFCE into an equivalent BCE, achieving the first breakthrough in polynomial-time computation of BCE.
On the Robustness of Neural-Enhanced Video Streaming against Adversarial Attacks
Qihua Zhou (Hong Kong Polytechnic University), Zhenda Xu (Hong Kong Polytechnic University)
Object TrackingPose EstimationSuper ResolutionCompressionAdversarial AttackVideo
🎯 What it does: A new adversarial attack method called codec hijacking is proposed and implemented for the NeVS (Neural Enhanced Video Stream) system, which combines perturbation injection and macroblock quantization parameter (QP) control. This method can degrade the final super-resolved video quality and lead to the failure of downstream perception tasks without affecting the video bitrate.
On the Role of Server Momentum in Federated Learning
Jianhui Sun (University of Virginia), Aidong Zhang (University of Maryland)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a general server momentum framework called FedGM, introducing multi-stage scheduling and asynchronous system heterogeneity processing, along with theoretical convergence analysis and experimental validation.
On the Structural Hardness of Answer Set Programming: Can Structure Efficiently Confine the Power of Disjunctions?
Markus Hecher (Massachusetts Institute of Technology), Rafael Kiesel (TU Wien)
🎯 What it does: The paper studies the structural complexity of disjunctive answer set programming (ASP) with disjunctive statements, exploring the impact of various graph structural parameters (such as vertex cover, tree width, tree depth, feedback vertex set, path width, bandwidth, and cut width) on problem-solving time, and provides corresponding algorithms and lower bounds.