AAAI 2024 Papers — Page 7
AAAI Conference on Artificial Intelligence · 2331 papers
DS-AL: A Dual-Stream Analytic Learning for Exemplar-Free Class-Incremental Learning
Huiping Zhuang (South China University of Technology), Zhiping Lin (Nanyang Technological University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A dual-stream analytical learning (DS-AL) framework is proposed for exemplar-free class incremental learning (CIL), achieving continuous learning without catastrophic forgetting through the mainstream C-RLS closed-form solution and the compensation stream DAC module.
DSD²: Can We Dodge Sparse Double Descent and Compress the Neural Network Worry-Free?
Victor Quétu (Telecom Paris Institute Polytechnique de Paris), Enzo Tartaglione (Telecom Paris Institute Polytechnique de Paris)
CompressionKnowledge DistillationImage
🎯 What it does: This study proposes a learning framework that avoids sparse double descent through knowledge distillation, enhancing the performance of sparse networks without sacrificing accuracy.
DTF-AT: Decoupled Time-Frequency Audio Transformer for Event Classification
Tony Alex (Surrey Institute for People Centred AI), Philip JB Jackson (Surrey Institute for People Centred AI)
ClassificationTransformerAudio
🎯 What it does: Proposes the DTF-AT audio Transformer, which achieves audio event classification through a time-frequency decoupled convolutional branch and a multi-scale MaxViT structure;
DTL: Disentangled Transfer Learning for Visual Recognition
Minghao Fu (Nanjing University), Jianxin Wu (Nanjing University)
RecognitionDomain AdaptationComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes a Disentangled Transfer Learning (DTL) framework, which achieves the aggregation and adjustment of task-specific information by inserting a lightweight Compact Side Network (CSN) into the Transformer backbone, significantly reducing GPU memory usage while maintaining high accuracy.
DTMFormer: Dynamic Token Merging for Boosting Transformer-Based Medical Image Segmentation
Zhehao Wang (Huazhong University of Science and Technology), Zengqiang Yan (Huazhong University of Science and Technology)
SegmentationTransformerImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes the DTMFormer dynamic Token merging Transformer block, which addresses the attention collapse problem in medical image segmentation and improves convergence and performance.
Dual Self-Paced Cross-Modal Hashing
Yuan Sun (Sichuan University), Peng Hu
RetrievalImageTextMultimodality
🎯 What it does: This paper proposes a Dual-layer Self-adaptive Self-learning Cross-modal Hashing (DSCMH), which gradually learns from easy samples/features to difficult samples/features through instance-level and feature-level self-learning weights to obtain robust binary hash codes.
Dual-Channel Learning Framework for Drug-Drug Interaction Prediction via Relation-Aware Heterogeneous Graph Transformer
Xiaorui Su (University of Chinese Academy of Sciences), Lun Hu (Xinjiang Technical Institutes of Physics and Chemistry)
Drug DiscoveryGraph Neural NetworkTransformerGraphBiomedical Data
🎯 What it does: A Transformer-based dual-channel relation-aware heterogeneous graph network called TIGER is proposed for drug-drug interaction prediction.
Dual-Level Curriculum Meta-Learning for Noisy Few-Shot Learning Tasks
Xiaofan Que (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
Meta LearningImage
🎯 What it does: A dual-layer curriculum meta-learning framework (DCML) is proposed, which constructs curricula at both the category level and the sample level to improve robustness against noisy few-shot learning.
Dual-Perspective Knowledge Enrichment for Semi-supervised 3D Object Detection
Yucheng Han (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
Object DetectionPoint Cloud
🎯 What it does: A Dual-Perspective Knowledge Enrichment (DPKE) method is proposed for semi-supervised 3D object detection, addressing issues of data scarcity and low-quality pseudo-labels.
Dual-Prior Augmented Decoding Network for Long Tail Distribution in HOI Detection
Jiayi Gao (Beijing University of Posts and Telecommunications), Jun Guo (Li Auto)
RecognitionObject DetectionKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a Dual Prior Enhanced Decoding Network (DPADN), which splits the human-object interaction detection task into two sub-tasks: human-object pair detection and interaction recognition. It utilizes prior information provided by external object classifiers and verb discriminators to assist decoding, thereby alleviating the recognition difficulties caused by long-tail distributions.
Dual-View Whitening on Pre-trained Text Embeddings for Sequential Recommendation
Lingzi Zhang (Nanyang Technological University), Zhiqi Shen (Nanyang Technological University)
Recommendation SystemTransformerTextSequential
🎯 What it does: For the sequence recommendation task, this paper only uses pre-trained text embeddings (without using product ID embeddings). By applying whitening to remove the correlation among dimensions of the text embeddings and enhance uniformity, and combining two perspectives of full whitening and group whitening, we construct the DWSRec dual-view model, which utilizes decoupled attention and fusion mechanisms to learn more robust representations of users and items.
Dual-Window Multiscale Transformer for Hyperspectral Snapshot Compressive Imaging
Fulin Luo (Chongqing University), Tan Guo (Chongqing University)
RestorationCompressionTransformerImage
🎯 What it does: This paper proposes a Dual-Window Multi-Scale Transformer (DWMT) based on Transformer for the reconstruction of hyperspectral snapshot compressive imaging, constructing a two-stage U-Net network and integrating a dual-branch encoder.
DUEL: Duplicate Elimination on Active Memory for Self-Supervised Class-Imbalanced Learning
Won-Seok Choi (Seoul National University), Byoung-Tak Zhang (Seoul National University)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: The DUEL framework is proposed, utilizing active memory and a duplicate elimination mechanism to achieve unsupervised class imbalance learning.
DVANet: Disentangling View and Action Features for Multi-View Action Recognition
Nyle Siddiqui (University of Central Florida), Mubarak Shah (University of Central Florida)
RecognitionTransformerContrastive LearningVideo
🎯 What it does: A multi-view action recognition framework DVANet based on a Transformer decoder is proposed, which separates action features from view features using learnable queries and achieves view-invariant action representation through supervised contrastive learning.
DVSAI: Diverse View-Shared Anchors Based Incomplete Multi-View Clustering
Shengju Yu (National University of Defense Technology), Xinwang Liu (Nanjing University of Aeronautics and Astronautics)
OptimizationImage
🎯 What it does: The DVSAI framework is proposed, which utilizes multi-scale, multi-dimensional, and multi-size view shared anchor points to address the problem of incomplete multi-view clustering.
Dynamic Budget Throttling in Repeated Second-Price Auctions
Zhaohua Chen (Peking University), Xiaotie Deng (Peking University)
OptimizationReinforcement Learning
🎯 What it does: This paper studies the theoretical performance of dynamic budget throttling strategies for a single advertiser in repeated second-price auctions and proposes an OGD-CB algorithm that is robust to value models.
Dynamic Feature Pruning and Consolidation for Occluded Person Re-identification
YuTeng Ye (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
RecognitionRetrievalTransformerImage
🎯 What it does: A framework based on Feature Pruning and Fusion (FPC) is proposed to address the problem of occluded person re-identification.
Dynamic Knowledge Injection for AIXI Agents
Samuel Yang-Zhao (Australian National University), Marcus Hutter
OptimizationReinforcement LearningAgentic AIGraph
🎯 What it does: This paper proposes the DynamicHedgeAIXI agent, which achieves precise Bayesian mixing of AIXI through online model injection, addressing the cognitive bias caused by fixed model classes in traditional AIXI approximations.
Dynamic Reactive Spiking Graph Neural Network
Han Zhao (Xidian University), Junchi Yan (Shanghai Jiao Tong University)
Graph Neural NetworkSpiking Neural NetworkGraph
🎯 What it does: This paper proposes a Dynamic Reactive Spiking Graph Neural Network (DRSGNN) that enhances the model's expressive power and energy efficiency through optimizable threshold spiking neurons and learnable graph positional information.
Dynamic Regret of Adversarial MDPs with Unknown Transition and Linear Function Approximation
Long-Fei Li (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies the dynamic regret problem in adversarial Markov decision processes (MDPs) under unknown transitions and linear function approximation, proposing a general framework to address the uncertainties brought by unknown transitions and environmental non-stationarity.
Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition
Jianyang Xie (University of Liverpool), Yalin Zheng (University of Liverpool)
RecognitionGraph Neural NetworkVideo
🎯 What it does: A dynamic semantic-driven spatial graph convolutional network (DS-GCN) is proposed, which achieves more accurate modeling of skeletal actions by dynamically encoding the types of joints and edges in graph convolution.
Dynamic Spiking Graph Neural Networks
Nan Yin (Mohamed bin Zayed University of Artificial Intelligence), Bin Gu (Jilin University)
ClassificationGraph Neural NetworkSpiking Neural NetworkGraph
🎯 What it does: A dynamic spiking graph neural network (Dy-SIGN) is proposed for the node classification task on time-varying graphs, combining SNN and GNN to achieve efficient learning through information compensation and implicit differentiation training.
Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning
Yan Fan (Tianjin University), Qinghua Hu (Tianjin University)
Knowledge DistillationGraph Neural NetworkContrastive LearningImage
🎯 What it does: In semi-supervised continual learning, the issue of catastrophic forgetting with unlabeled data is studied, and a Dynamic Subgraph Distillation (DSGD) method is proposed;
Dynamic Tangled Derivative Logic of Metric Spaces
David Fernández-Duque (University of Barcelona), Yoàv Montacute (University of Cambridge)
🎯 What it does: A dynamic spatial logic that combines Cantor derivatives and temporal logic is proposed, incorporating the μ-operator to achieve spatial fixed-point reasoning, and its decidability and completeness are proven.
Dynamic Weighted Combiner for Mixed-Modal Image Retrieval
Fuxiang Huang (Chongqing University), Suqi Song (Chongqing University)
RetrievalKnowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: A Dynamic Weighted Combiner (DWC) is proposed for Mixed Modal Image Retrieval (MMIR), addressing issues of modality importance imbalance, label noise, and modality gap through feature editing, dynamic soft labels, and multimodal contrastive learning.
E2E-AT: A Unified Framework for Tackling Uncertainty in Task-Aware End-to-End Learning
Wangkun Xu (Imperial College London), Fei Teng (Imperial College London)
OptimizationAdversarial AttackTime Series
🎯 What it does: This paper proposes a unified robust framework E2E-AT, which systematically models and trains two types of uncertainties in end-to-end learning: input features and unpredictable constraint optimization parameters.
E2HQV: High-Quality Video Generation from Event Camera via Theory-Inspired Model-Aided Deep Learning
Qiang Qu (University of Sydney), Tongliang Liu (University of Sydney)
GenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: Generating high-quality video frames from the event stream of event cameras
EAN: An Efficient Attention Module Guided by Normalization for Deep Neural Networks
Jiafeng Li (East China Normal University), Ying Wen (East China Normal University)
ClassificationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: An efficient attention module EAN based on normalization is proposed to integrate normalization and attention mechanisms in deep networks.
EarnHFT: Efficient Hierarchical Reinforcement Learning for High Frequency Trading
Molei Qin (Nanyang Technological University), Bo An (Nanyang Technological University)
OptimizationReinforcement LearningTime SeriesFinance Related
🎯 What it does: A three-stage hierarchical reinforcement learning framework EarnHFT is proposed for high-frequency trading in cryptocurrencies.
Earthfarsser: Versatile Spatio-Temporal Dynamical Systems Modeling in One Model
Hao Wu (University of Science and Technology of China), Kun Wang (University of Science and Technology of China)
Convolutional Neural NetworkTransformerVideoTime SeriesPhysics Related
🎯 What it does: This paper presents EarthFarseer, a unified spatiotemporal modeling framework that balances local convolution with global Fourier-Transformer, and introduces a continuous-time Fourier-temporal transform to capture long-term temporal dependencies.
EarthVQA: Towards Queryable Earth via Relational Reasoning-Based Remote Sensing Visual Question Answering
Junjue Wang (Wuhan University), Yanfei Zhong (Wuhan University)
Object DetectionSegmentationVision Language ModelImage
🎯 What it does: A dataset called EarthVQA based on remote sensing visual question answering is proposed, along with a semantic object awareness framework (SOBA);
EAT: Towards Long-Tailed Out-of-Distribution Detection
Tong Wei (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationAnomaly DetectionConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: To address the problem of external sample detection under long-tail distribution, the EAT framework is proposed, which combines virtual labels, multiple rejection categories, tail category enhancement, model ensemble, and fine-tuning.
ECHO-GL: Earnings Calls-Driven Heterogeneous Graph Learning for Stock Movement Prediction
Mengpu Liu (Zhejiang University), Xiaolin Zheng (Zhejiang University)
ClassificationRecommendation SystemAnomaly DetectionOptimizationRecurrent Neural NetworkGraph Neural NetworkTextMultimodalityGraphTime SeriesFinance RelatedStochastic Differential EquationOrdinary Differential EquationAudio
🎯 What it does: This paper proposes a heterogeneous dynamic graph model ECHO-GL based on earnings call semantics to predict stock price movements over various time windows.
EcomGPT: Instruction-Tuning Large Language Models with Chain-of-Task Tasks for E-commerce
Yangning Li (Tsinghua University), Yong Jiang (Alibaba Group)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: A dataset for instruction tuning aimed at e-commerce, EcomInstruct (approximately 2.5 million instructions, 134 tasks), and a specialized LLM EcomGPT trained on this dataset are proposed.
EDA: Evolving and Distinct Anchors for Multimodal Motion Prediction
Longzhong Lin (Zhejiang University), Yue Wang (Zhejiang University)
Autonomous DrivingTransformerMultimodality
🎯 What it does: Proposed and implemented an Evolvable and Unique Anchor (EDA) method, which significantly enhances regression and scoring capabilities in multimodal motion prediction by gradually updating anchors through a multi-layer decoder and using NMS before matching.
Editing Language Model-Based Knowledge Graph Embeddings
Siyuan Cheng (Zhejiang University), Huajun Chen (Tencent)
Recurrent Neural NetworkTransformerLarge Language ModelGraph
🎯 What it does: This paper proposes knowledge graph embedding (KGE) based on language models for fast and data-efficient editing tasks (EDIT/ADD), and constructs the corresponding datasets.
Effect Size Estimation for Duration Recommendation in Online Experiments: Leveraging Hierarchical Models and Objective Utility Approaches
Yu Liu (Amazon.com Inc), Rui Song (Amazon.com Inc)
Recommendation SystemOptimizationTabularTime Series
🎯 What it does: This paper studies how to automate the recommendation of appropriate Average Effect Size (AES) for large-scale online experiments, thereby determining the duration of experiments and improving their efficiency and accuracy.
Effective Causal Discovery under Identifiable Heteroscedastic Noise Model
Naiyu Yin (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
TabularBiomedical Data
🎯 What it does: A multivariate structural equation model (SEM) that is identifiable under heteroscedastic noise conditions is proposed, along with a two-stage iterative learning framework based on continuously differentiable DAG constraints for learning causal directed acyclic graphs from observational data.
Effective Comparative Prototype Hashing for Unsupervised Domain Adaptation
Hui Cui (Qilu University of Technology), Jingjing Li (University of Electronic Science and Technology of China)
RetrievalDomain AdaptationContrastive LearningImage
🎯 What it does: A comparative prototype hashing method for unsupervised domain adaptation retrieval (CPH) is proposed.
Effectiveness of Constant Stepsize in Markovian LSA and Statistical Inference
Dongyan (Lucy) Huo (Cornell University), Qiaomin Xie (University of Wisconsin-Madison)
Reinforcement LearningTime SeriesSequential
🎯 What it does: This paper studies statistical inference using the Linear Stochastic Approximation (LSA) algorithm with a constant step size under Markov data, and proposes a method for constructing confidence intervals based on average iteration and batch mean covariance estimation, supplemented by Richardson–Romberg extrapolation to reduce bias.
Efficient Algorithms for Non-gaussian Single Index Models with Generative Priors
Junren Chen (University of Hong Kong), Zhaoqiang Liu (University of Electronic Science and Technology of China)
RestorationGenerationOptimizationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: For high-dimensional single-index models (SIM), under the premise that the sensing vector is non-Gaussian and the signal satisfies the generative model prior, two efficient algorithms (projection gradient descent and projection power iteration) are proposed to achieve approximately optimal statistical recovery.
Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction
Yu Zang (Beijing University of Posts and Telecommunications), Yunfei Long (McMaster University)
OptimizationFederated LearningImage
🎯 What it does: An efficient asynchronous federated learning framework FedAC is proposed, which integrates time-gradient-based client weight evaluation, forward-looking weighted momentum for adaptive server updates, and fine-grained gradient correction for client updates.
Efficient Axiomatization of OWL 2 EL Ontologies from Data by Means of Formal Concept Analysis
Francesco Kriegel (Technische Universitat Dresden)
Graph
🎯 What it does: This paper proposes an algorithm based on Formal Concept Analysis for efficiently constructing a complete OWL 2 EL TBox (concept inclusion, range restrictions, and role inclusion) from graph data, and provides an implementable Scala implementation.
Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution
Yutao Yuan (Tsinghua University), Chun Yuan (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes an efficient image super-resolution method called ECDP, which utilizes continuous-time conditional diffusion models and probabilistic flow sampling.
Efficient Constrained K-center Clustering with Background Knowledge
Longkun Guo (Fuzhou University), Minhui Xue (CSIRO)
OptimizationTabular
🎯 What it does: A 2-approximation algorithm is proposed for the k-center clustering problem with must-link (ML) and cannot-link (CL) constraints.
Efficient Constraint Generation for Stochastic Shortest Path Problems
Johannes Schmalz (Australian National University), Felipe Trevizan (Australian National University)
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes an efficient constraint generation method that only considers actions contributing to the optimal solution when solving the Stochastic Shortest Path (SSP) problem, thereby reducing unnecessary Q-value calculations. This method is applied to the iLAO* algorithm, resulting in a new CG-iLAO* algorithm.
Efficient Deweahter Mixture-of-Experts with Uncertainty-Aware Feature-Wise Linear Modulation
Rongyu Zhang (Nanjing University), Shanghang Zhang (Peking University)
ClassificationRestorationSegmentationTransformerMixture of ExpertsImage
🎯 What it does: An efficient Mixture-of-Feature-Modulation-Experts (MoFME) framework is designed for image denoising tasks such as rain removal and fog removal under various weather conditions, while also considering downstream segmentation/classification.
Efficient Learning in Polyhedral Games via Best-Response Oracles
Darshan Chakrabarti (Columbia University), Christian Kroer (MIT)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes an online learning framework called AFW-ROMD, which utilizes the best response (i.e., linear minimization) oracle to perform approximate in-place updates on a polyhedral strategy set, thereby enabling the computation of Nash equilibria and coarse correlated equilibria for two-player zero-sum games and general multiplayer games, requiring only O(log t) calls to the best response at each iteration.
Efficient Learning of PDEs via Taylor Expansion and Sparse Decomposition into Value and Fourier Domains
Md Nasim (Purdue University), Yexiang Xue (Purdue University)
OptimizationComputational EfficiencyTabularPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper studies the algorithm REEL, which accelerates PDE model learning using random projection, Taylor expansion, and frequency domain sparse decomposition.
Efficient Lightweight Image Denoising with Triple Attention Transformer
Yubo Zhou (Xiamen University), Yuan Xie (East China Normal University)
RestorationTransformerImage
🎯 What it does: A lightweight image denoising Transformer (LIDFormer) is proposed, which significantly reduces computational load and improves denoising performance through discrete wavelet transform (DWT) for feature compression, a complementary periodic feature reuse (CPFR) module, and a three-dimensional multi-convolution head transpose attention (TMDTA).
Efficient Look-Up Table from Expanded Convolutional Network for Accelerating Image Super-resolution
Kai Yin (University of Electronic Science and Technology of China), Jie Shen (University of Electronic Science and Technology of China)
RestorationSuper ResolutionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a LUT-accelerated super-resolution method using extended convolution (EC), which significantly improves inference speed while maintaining high quality.
Efficient Nonparametric Tensor Decomposition for Binary and Count Data
Zerui Tao (Tokyo University of Agriculture and Technology), Qibin Zhao (RIKEN Center for Advanced Intelligence Project)
TabularStochastic Differential Equation
🎯 What it does: This paper proposes an efficient non-parametric tensor decomposition model, ENTED, for handling binary and count tensor data, aiming to achieve higher quality tensor completion.
Efficient Online Crowdsourcing with Complex Annotations
Reshef Meir (Technion Israel Institute of Technology), Udi Weinsberg (Meta)
OptimizationTabular
🎯 What it does: This paper proposes three algorithms for online crowdsourcing tasks (OAK, POAK, POAKI) that can adaptively decide whether to continue requesting labels while collecting complex annotations (such as bounding boxes, classification paths, etc.) and estimate worker quality through label similarity, thus achieving a cost-quality trade-off.
Efficient Representation Learning of Satellite Image Time Series and Their Fusion for Spatiotemporal Applications
Poonam Goyal (Birla Institute of Technology and Science), Navneet Goyal (Birla Institute of Technology and Science)
OptimizationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageTime SeriesAgriculture Related
🎯 What it does: Proposes two models, PatchNet and FuSITSNet, for efficient processing and fusion of high spatial resolution satellite image time series to predict crop yield, snow cover, and solar energy.
Efficient Spiking Neural Networks with Sparse Selective Activation for Continual Learning
Jiangrong Shen (Zhejiang University), Huajin Tang (Dalian University of Technology)
ClassificationSpiking Neural NetworkImage
🎯 What it does: A time-series neural network based on suppressive sparse activation (SA-SNN) is proposed for achieving continuous learning without forgetting old knowledge.
Efficient Target Propagation by Deriving Analytical Solution
Yanhao Bao (Tokyo Institute of Technology), Nakamasa Inoue (Tokyo Institute of Technology)
ClassificationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: An analytical feedback function based on Jacobian matching loss is proposed, which significantly improves training efficiency by skipping traditional feedback network training in target propagation.
EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration
Zicheng Cai (Guangdong University of Technology), Yutao Lai (Guangdong University of Technology)
Neural Architecture SearchImage
🎯 What it does: This paper proposes EG-NAS, a neural architecture search framework that combines gradient descent with an improved evolutionary strategy, aiming to reduce search costs and avoid getting trapped in local optima.
Electron Microscopy Images as Set of Fragments for Mitochondrial Segmentation
Naisong Luo (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
SegmentationTransformerImage
🎯 What it does: A fragmented model called FragViT based on visual Transformer is proposed for the three-dimensional segmentation of mitochondria in electron microscope images;
Eliciting Honest Information from Authors Using Sequential Review
Yichi Zhang (University of Michigan), Weijie Su (University of Pennsylvania)
Tabular
🎯 What it does: A sequential review mechanism is proposed, utilizing the relative ranking information of authors when submitting multiple papers to incentivize authors to rank truthfully, thereby improving conference acceptance decisions.
Eliciting Kemeny Rankings
Anne-Marie George (University of Oslo), Christos Dimitrakakis (University of Neuchatel)
🎯 What it does: The paper models the problem of collecting voter preferences as a dueling bandits problem and proposes a method for estimating Kemeny rankings under PAC (Probably Approximately Correct) conditions.
Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift
Shengwei An (Purdue University), Xiangyu Zhang (Purdue University)
GenerationAnomaly DetectionDiffusion modelImage
🎯 What it does: The ELIJAH framework is proposed for detecting and removing backdoor injections in diffusion models.
Embedded Feature Selection on Graph-Based Multi-View Clustering
Wenhui Zhao (Xidian University), Qianqian Wang (Xidian University)
Graph Neural NetworkGraph
🎯 What it does: A multi-view clustering method based on anchor graphs, EFSGMC, is proposed, which employs embedded feature selection and tensor Schatten p-norm to directly obtain clustering labels in one go, without the need for post-processing.
Embracing Language Inclusivity and Diversity in CLIP through Continual Language Learning
Bang Yang (Peking University), Yuexian Zou (Peking University)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes a continuous language learning framework based on CLIP, called CLL-CLIP, which focuses on training only the scalable token embedding layer. It combines cross-modal InfoNCE and cross-language MSE objectives to achieve alignment between images and multilingual text, and employs the TEIR method for unified distribution initialization and frequency regularization of token embeddings to mitigate catastrophic forgetting.
Emergent Communication for Numerical Concepts Generalization
Enshuai Zhou (University of Science and Technology of China), Yunji Chen (Institute of Software)
Convolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This paper proposes a two-stage training framework (NumSen+NumRel) that enables multiple agents to learn numerical concepts through emergent communication in NumGame and achieve counting and arithmetic reasoning for unseen quantities.
EMGAN: Early-Mix-GAN on Extracting Server-Side Model in Split Federated Learning
Jingtao Li (Sony AI), Chaitali Chakrabarti (Arizona State University)
Federated LearningAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an Early Mixed GAN Attack (EMGAN) for Split Federated Learning, which effectively extracts the server-side model through three modules: early learning, multiple GANs, and ProperMix.
Emotion Rendering for Conversational Speech Synthesis with Heterogeneous Graph-Based Context Modeling
Rui Liu (Inner Mongolian University), Haizhou Li (Chinese University of Hong Kong)
GenerationGraph Neural NetworkTransformerContrastive LearningAudio
🎯 What it does: This paper proposes an emotional dialogue speech synthesis model (ECSS) that can accurately capture and generate emotional speech within the context of dialogue.
Empowering CAM-Based Methods with Capability to Generate Fine-Grained and High-Faithfulness Explanations
Changqing Qiu (Beijing Institute of Technology), Yining Zhang (Peking University)
Explainability and InterpretabilityConvolutional Neural NetworkScore-based ModelImage
🎯 What it does: A novel explanation method FG-CAM is studied, which gradually enhances the resolution of explanations by utilizing the relationships between feature maps of adjacent layers, thereby generating fine-grained and highly credible interpretable results at shallow and input layers.
Empowering Dual-Level Graph Self-Supervised Pretraining with Motif Discovery
Pengwei Yan (Zhejiang University), Xiaozhong Liu (Worcester Polytechnic Institute)
ClassificationRepresentation LearningDrug DiscoveryGraph Neural NetworkAuto EncoderContrastive LearningGraph
🎯 What it does: A dual-layer graph self-supervised pre-training framework DGPM is proposed, which automatically discovers subgraph patterns and achieves node-subgraph interactive learning.
Encoding Constraints as Binary Constraint Networks Satisfying BTP
Ruiwei Wang (National University of Singapore)
🎯 What it does: This paper proposes and analyzes a binary constraint network (BTPN) that satisfies the broken triangle property, proving that its consistency checking can be implemented using monotonic circuits of polynomial size, and explores its expressiveness in constraint modeling.
End-to-End Learning of LTLf Formulae by Faithful LTLf Encoding
Hai Wan (Sun Yat-sen University), Bo Peng (Sun Yat-sen University)
ClassificationExplainability and InterpretabilityTabular
🎯 What it does: This study proposes an end-to-end learning method for linear temporal logic (LTL_f) formulas, utilizing 'faithful LTL_f encoding' to ensure that the parameters of the neural network correspond one-to-one with the LTL_f formulas, thereby enabling the automatic discovery of tree-structured formulas from large-scale data.
End-to-End Real-Time Vanishing Point Detection with Transformer
Xin Tong (Intelligent Science and Technology Academy of CASIC), Xuhui Huang (Intelligent Science and Technology Academy of CASIC)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: An end-to-end Transformer-based real-time vanishing point detection method called VPTR is proposed, which directly regresses vanishing point coordinates from images.
End-to-End RGB-D Image Compression via Exploiting Channel-Modality Redundancy
Huiming Zheng (Peking University), Wei Gao (Peking University)
CompressionTransformerImage
🎯 What it does: A Transformer-based end-to-end RGB-D compression framework is proposed, utilizing YUV420 domain processing, cross-modal and single-modal attention, and a conditional entropy model, significantly improving rate-distortion performance.
End-to-End Verification for Subgraph Solving
Stephan Gocht (Lund University), Yong Kiam Tan (Institute for Infocomm Research)
Graph Neural NetworkGraph
🎯 What it does: A toolchain has been designed and implemented for the first time to perform end-to-end formal verification of problems such as maximum clique, subgraph isomorphism, and maximum common connected induced subgraph, covering the entire process from graph problem encoding, proof log generation to final result verification.
Energy Efficient Streaming Time Series Classification with Attentive Power Iteration
Hao Huang (General Electric Vernova Research), Shinjae Yoo (Brookhaven National Lab)
ClassificationComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningTime SeriesSequential
🎯 What it does: A streaming time series classification network named strAPI has been designed and implemented, capable of real-time processing of time series on resource-constrained devices and providing classification results.
Engineering an Exact Pseudo-Boolean Model Counter
Suwei Yang (GrabTaxi Holdings), Kuldeep S. Meel (National University of Singapore)
Tabular
🎯 What it does: This paper presents PBCount, an exact model counter for pseudo-Boolean (PB) formulas.
Enhance Sketch Recognition’s Explainability via Semantic Component-Level Parsing
Guangming Zhu (Xidian University), Liang Zhang (Xidian University)
RecognitionExplainability and InterpretabilityRecurrent Neural NetworkTransformerImage
🎯 What it does: A structured sketch recognition network has been constructed, utilizing a semantic component-level memory module to achieve interpretable sketch recognition and segmentation.
Enhanced Fine-Grained Motion Diffusion for Text-Driven Human Motion Synthesis
Dong Wei (Nanjing University of Science and Technology), Jianfeng Lu (Nanjing University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes DiffKFC, a conditional diffusion model that enables fine-grained human action synthesis through sparse keyframes guided by text.
Enhancing Bilingual Lexicon Induction via Bi-directional Translation Pair Retrieving
Qiuyu Ding (Harbin Institute of Technology), Tiejun Zhao (Harbin Institute of Technology)
RetrievalOptimizationHyperparameter SearchText
🎯 What it does: This paper proposes a bidirectional retrieval strategy (BRTR and BRSR) that simultaneously utilizes retrieval results from the source language to the target language and from the target language to the source language in Bilingual Lexicon Induction (BLI) to address the 'source-side bias' problem caused by unidirectional retrieval.
Enhancing Cognitive Diagnosis Using Un-interacted Exercises: A Collaboration-Aware Mixed Sampling Approach
Haiping Ma (Anhui University), Xingyi Zhang (Anhui University)
Tabular
🎯 What it does: This paper proposes the CMES (Collaborative-aware Mixed Exercise Sampling) framework, which enhances the accuracy of cognitive diagnosis by sampling and mixing non-interactive questions.
Enhancing Ensemble Clustering with Adaptive High-Order Topological Weights
Jiaxuan Xu (Southwestern University of Finance and Economics), Lei Duan (Southwestern University of Finance and Economics)
Tabular
🎯 What it does: This paper proposes a topology-based ensemble clustering algorithm with adaptive weights based on higher-order connections (AWEC). It learns the optimal connection matrix from the multi-order connection information of the co-affinity matrix and embeds topology learning to achieve more robust clustering results.
Enhancing Evolving Domain Generalization through Dynamic Latent Representations
Binghui Xie (Chinese University of Hong Kong), James Cheng (Chinese University of Hong Kong)
Domain AdaptationRecurrent Neural NetworkAuto EncoderTime SeriesSequential
🎯 What it does: This paper proposes the MISTS framework, which simultaneously learns dynamic features of temporal evolution and cross-domain invariant features in the task of Evolution Domain Generalization (EDG), and utilizes an adaptive classifier to predict future domains.
Enhancing Hyperspectral Images via Diffusion Model and Group-Autoencoder Super-resolution Network
Zhaoyang Wang (Xidian University), Maoguo Gong (Xidian University)
RestorationSuper ResolutionDiffusion modelAuto EncoderImage
🎯 What it does: A two-stage framework combining Graph Autoencoders (GAE) and diffusion models is proposed for single image super-resolution of hyperspectral images.
Enhancing Job Recommendation through LLM-Based Generative Adversarial Networks
Yingpeng Du (Peking University), Jie Zhang (Nanyang Technological University)
Recommendation SystemTransformerLarge Language ModelGenerative Adversarial NetworkText
🎯 What it does: This paper proposes an interactive recommendation framework called LGIR, based on large language models and generative adversarial networks, to enhance job recommendation effectiveness on online recruitment platforms.
Enhancing Low-Resource Relation Representations through Multi-View Decoupling
Chenghao Fan (Huazhong University of Science and Technology), Dangyang Chen (Ping An Property and Casualty Insurance Company of China)
Representation LearningTransformerPrompt EngineeringContrastive LearningText
🎯 What it does: A multi-view decoupled Prompt-Tuning method MVRE is proposed to enhance relation extraction performance in low-resource scenarios.
Enhancing Multi-Label Classification via Dynamic Label-Order Learning
Jiangnan Li (Harbin Institute of Technology), Ruifeng Xu (Harbin Institute of Technology)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes a dynamic label order learning method that adaptively generates sample-specific label sequences in multi-label classification tasks by progressively selecting the most difficult labels and concatenating them into a sequence.
Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning
Pengfei Jiao (Hangzhou Dianzi University), Huaming Wu (Tianjin University)
Recurrent Neural NetworkGraph Neural NetworkGenerative Adversarial NetworkGraphTime SeriesSequential
🎯 What it does: The MINDS model is proposed to simultaneously handle two tasks of information diffusion: macro (predicting the final scale of diffusion) and micro (predicting the next affected users);
Enhancing Neural Radiance Fields with Adaptive Multi-Exposure Fusion: A Bilevel Optimization Approach for Novel View Synthesis
Yang Zou (University of Sydney), Jinyuan Liu (Dalian University of Technology)
GenerationOptimizationNeural Radiance FieldImage
🎯 What it does: This paper proposes an unsupervised multi-exposure correction and dual-layer optimization framework that jointly trains NeRF to synthesize high-quality new perspective images under extreme lighting conditions such as low light and overexposure.
Enhancing RAW-to-sRGB with Decoupled Style Structure in Fourier Domain
Xuanhua He (University of Science and Technology of China), Man Zhou (Nanyang Technological University)
Image TranslationRestorationDomain AdaptationImage
🎯 What it does: A RAW-to-sRGB mapping framework called FourierISP is proposed, which utilizes separate sub-networks for structure enhancement, style learning, and color fusion to achieve high-quality mobile RAW to DSLR RGB conversion.
Enhancing Representation of Spiking Neural Networks via Similarity-Sensitive Contrastive Learning
Yuhan Zhang (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
Knowledge DistillationRepresentation LearningSpiking Neural NetworkContrastive LearningImage
🎯 What it does: By introducing a similarity-sensitive contrastive learning framework, spiking neural networks (SNNs) can maximize the mutual information with the corresponding features of artificial neural networks (ANNs), thereby enhancing the representational capacity and classification accuracy of SNNs.
Enhancing Semi-supervised Domain Adaptation via Effective Target Labeling
Jiujun He (Southwestern University of Finance and Economics), Guosheng Yin (University of Hong Kong)
Domain AdaptationImage
🎯 What it does: An effective target sample labeling framework is proposed, which utilizes active learning and pseudo-labeling strategies to automatically select informative target domain samples, thereby enhancing the performance of semi-supervised domain adaptation (SSDA).
Enhancing the Efficiency of Altruism and Taxes in Affine Congestion Games through Signalling
Vittorio Bilò (Università del Salento), Cosimo Vinci (Università del Salento)
Optimization
🎯 What it does: A signaling mechanism is introduced in convex combination conflict games, analyzing the impact of θ-charitable players and personalized taxes on price disorder when the optimal strategy is signaled, proving that any pure Nash equilibrium is socially optimal under perfect balance of charity level;
Enhancing the Robustness of Spiking Neural Networks with Stochastic Gating Mechanisms
Jianhao Ding (Peking University), Jian K. Liu (University of Birmingham)
Computational EfficiencyAdversarial AttackSpiking Neural NetworkReinforcement LearningImage
🎯 What it does: A stochastic gating model is introduced in deep SNNs to randomly filter spikes, enhancing adversarial robustness and energy efficiency.
Enhancing Training of Spiking Neural Network with Stochastic Latency
Srinivas Anumasa (Mohamed bin Zayed University of Artificial Intelligence), Bin Gu (Jilin University)
Spiking Neural NetworkImage
🎯 What it does: A direct training method called Stochastic Latency Training (SLT) is proposed, enabling a single SNN model to maintain high accuracy across different inference latencies while significantly reducing training time.
Enhancing Zero-Shot Multi-Speaker TTS with Negated Speaker Representations
Yejin Jeon (POSTECH), Gary Geunbae Lee
GenerationData SynthesisTransformerAudio
🎯 What it does: A new framework for zero-shot multi-speaker TTS is proposed, utilizing negative feature learning and multi-stream Transformer to extract decoupled speaker representations, and achieving the fusion of speaker and text through adaptive layer normalization.
EnMatch: Matchmaking for Better Player Engagement via Neural Combinatorial Optimization
Kai Wang (Fuxi AI Lab NetEase Inc), Changjie Fan (Fuxi AI Lab NetEase Inc)
Recommendation SystemOptimizationGraph Neural NetworkTransformerReinforcement LearningTabular
🎯 What it does: This paper proposes EnMatch, a k-vsk scene matching framework focused on player engagement, which achieves high-quality matching across levels by constructing a matching strategy based on Neural Combinatorial Optimization (NCO) and a fine-grained engagement prediction model.
Entropic Open-Set Active Learning
Bardia Safaei (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an Entropic Open-Set Active Learning framework that selects informative samples by utilizing two entropy scores for known and unknown categories, thereby achieving better sample selection in open-set active learning.
Entropy Induced Pruning Framework for Convolutional Neural Networks
Yiheng Lu (Xidian University), Cai Xu (Xidian University)
Convolutional Neural NetworkImage
🎯 What it does: A structured pruning framework based on information entropy, AFIE, is proposed, which can evaluate the importance of convolution kernels and achieve one-time pruning when the model is trained for only one round or has been sufficiently trained.
Envy-Free House Allocation under Uncertain Preferences
Haris Aziz (University of New South Wales), Mashbat Suzuki (University of New South Wales)
Optimization
🎯 What it does: The study investigates the problem of envy-free housing allocation under uncertain preferences, proposing and analyzing algorithms and complexities for maximum envy-free probability allocation under various preference uncertainty models.
Episodic Return Decomposition by Difference of Implicitly Assigned Sub-trajectory Reward
Haoxin Lin (Nanjing University), Yang Yu (Nanjing University)
Recurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: A new return decomposition method based on sub-trajectory differences, called Diaster, is proposed to generate immediate agent rewards for reinforcement learning in environments with extremely delayed rewards.
EPSD: Early Pruning with Self-Distillation for Efficient Model Compression
Dong Chen (Jilin University), Jian Tang (Midea Group)
CompressionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: The EPSD framework is proposed, which combines early pruning with self-distillation to compress models directly from random initialization.