AAAI 2023 Papers — Page 3
AAAI Conference on Artificial Intelligence · 1578 papers
CMNet: Contrastive Magnification Network for Micro-Expression Recognition
Mengting Wei (Southeast University), Jiateng Liu (Southeast University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes a neural network for contrastive amplification and trend correction of micro-expressions to improve the accuracy of micro-expression recognition.
CMVAE: Causal Meta VAE for Unsupervised Meta-Learning
Guodong Qi (Zhejiang University), Huimin Yu (Zhejiang University)
Meta LearningAuto EncoderImage
🎯 What it does: This paper proposes an unsupervised meta-learning model based on causal structures, called CMVAE, to eliminate background bias and improve few-shot classification performance.
Co-imitation: Learning Design and Behaviour by Imitation
Chang Rajani (University of Helsinki), Ville Kyrki (Aalto University)
Robotic IntelligenceReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: A Co-Imitation Learning framework (CoIL) is designed to enable robots to learn behaviors and morphology simultaneously based solely on demonstration data without a reward function, achieving precise imitation of the demonstrator's morphology and behavior.
Coarse2Fine: Local Consistency Aware Re-prediction for Weakly Supervised Object Localization
Yixuan Pan (Southeast University), Xiaobo Lu (Southeast University)
Object DetectionSegmentationKnowledge DistillationTransformerImage
🎯 What it does: Using weakly supervised annotations, a local consistency-aware re-prediction framework (LCAR) is constructed to achieve the conversion from inconsistent activation maps to fine-grained complete masks.
COCA: COllaborative CAusal Regularization for Audio-Visual Question Answering
Mingrui Lao (Leiden University), Michael S. Lew (Dalian University of Technology)
RecognitionVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a Collaborative Causal Regularization (COCA) method to eliminate the unimodal and joint modality biases present in audio-visual question answering (AVQA) models, enhancing the model's ability to integrate reasoning across multimodal information.
Code-Aware Cross-Program Transfer Hyperparameter Optimization
Zijia Wang (Xiamen University), Jinsong Su (Xiamen University)
Recommendation SystemOptimizationHyperparameter SearchRecurrent Neural NetworkTransformerReinforcement LearningTabular
🎯 What it does: This study proposes the CaTHPO framework, which can accelerate the Bayesian Optimization (BO) process in cross-program hyperparameter optimization by transferring knowledge from similar programs.
Cogito Ergo Summ: Abstractive Summarization of Biomedical Papers via Semantic Parsing Graphs and Consistency Rewards
Giacomo Frisoni (University of Bologna), Gianluca Moro (University of Bologna)
GenerationGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningTextBiomedical DataReview/Survey Paper
🎯 What it does: This paper proposes an abstract medical literature summarization framework COGITOERGOSUMM based on semantic graphs (event extraction and AMR), which can generate more concise and readable summaries while maintaining factual consistency.
COLA: Improving Conversational Recommender Systems by Collaborative Augmentation
Dongding Lin (Hong Kong Polytechnic University), Wenjie Li (Hong Kong Polytechnic University)
Recommendation SystemGraph Neural NetworkTransformerTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a Collaborative Enhancement (COLA) method that combines user-item interaction graphs with retrieval of similar dialogues to simultaneously improve item representation and user preference modeling, thereby enhancing the recommendation effectiveness and generation quality of conversational recommendation systems.
Collective Intelligence in Human-AI Teams: A Bayesian Theory of Mind Approach
Samuel Westby (Northeastern University), Christoph Riedl (Northeastern University)
Text
🎯 What it does: A multi-agent network based on Bayesian inference was constructed to infer and enhance the collective intelligence of human teams in real-time during team chats.
Collusion-Proof and Sybil-Proof Reward Mechanisms for Query Incentive Networks
Youjia Zhang (Tsinghua University), Pingzhong Tang (Tsinghua University)
🎯 What it does: This paper designs two types of reward mechanisms for query incentive networks aimed at single-answer tasks, with the goal of encouraging agents to disseminate information and respond honestly while controlling the total rewards to stay within budget.
CoMAE: Single Model Hybrid Pre-training on Small-Scale RGB-D Datasets
Jiange Yang (Nanjing University), Limin Wang (Nanjing University)
ClassificationRepresentation LearningTransformerAuto EncoderContrastive LearningImageMultimodality
🎯 What it does: A single-model self-supervised mixed pre-training framework called CoMAE is proposed, which uses a unified Transformer encoder to perform cross-modal contrastive learning and multi-modal masked autoencoder pre-training on RGB and depth data, thereby learning high-quality multi-modal representations.
Combating Mode Collapse via Offline Manifold Entropy Estimation
Haozhe Liu (King Abdullah University of Science and Technology), Yefeng Zheng (Tencent)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: By extending the discriminator into a feature embedding network and maximizing entropy in its embedding space, MaEM-GAN is proposed to alleviate mode collapse in GANs.
Combating Unknown Bias with Effective Bias-Conflicting Scoring and Gradient Alignment
Bowen Zhao (Tsinghua University), Shu-Tao Xia (Tsinghua University)
ClassificationOptimizationImage
🎯 What it does: A two-stage unknown bias elimination framework is proposed, which first identifies conflicting samples through effective bias conflict scoring (including peer selection and periodic integration), and then uses gradient alignment to dynamically balance the contributions of bias alignment and conflicting samples' gradients, forcing the model to focus on intrinsic features.
Combinatorial Causal Bandits
Shi Feng (Tsinghua University), Wei Chen (Microsoft Research)
Reinforcement LearningTabular
🎯 What it does: A Combinatorial Causal Bandit (CCB) framework is proposed, where the learning agent selects up to K variables for intervention in each round, aiming to minimize the expected regret of the target variable Y.
Combinatorial Civic Crowdfunding with Budgeted Agents: Welfare Optimality at Equilibrium and Optimal Deviation
Sankarshan Damle (International Institute of Information Technology), Sujit Gujar (International Institute of Information Technology)
Optimization
🎯 What it does: This study investigates how budget-constrained agents donate under refund mechanisms in combinatorial civic crowdfunding and proves that it is impossible to achieve welfare optimality in equilibrium under any monotonic refund scheme; additionally, it proves that finding the optimal deviation strategy for an agent, given the donations of other agents, is NP-hard; subsequently, five heuristic strategies are proposed and evaluated through simulations for their trade-offs between social welfare and agent utility.
Combining Adversaries with Anti-adversaries in Training
Xiaoling Zhou (Tianjin University), Ou Wu (Tianjin University)
Adversarial AttackMeta LearningConvolutional Neural NetworkImage
🎯 What it does: The paper proposes a method that combines adversarial samples and anti-adversarial samples for training (CAAT) to enhance the robustness, accuracy, and fairness of deep networks.
Combining Slow and Fast: Complementary Filtering for Dynamics Learning
Katharina Ensinger (Bosch), Sebastian Trimpe (RWTH Aachen University)
Recurrent Neural NetworkTime SeriesSequentialPhysics Related
🎯 What it does: A dynamic modeling method based on complementary filtering is proposed, which learns high-frequency details and low-frequency long-term behavior from different models and fuses both during prediction.
Commitment Games with Conditional Information Disclosure
Anthony DiGiovanni (Center on Long-Term Risk), Jesse Clifton (Center on Long-Term Risk)
🎯 What it does: This paper proposes a new framework for commitment games that allows players to make conditional decisions based on the commitments and information disclosures of their opponents through commitment devices. It provides a corresponding folk theorem, proving that under this framework, any feasible payment vector that satisfies individual rationality constraints (including all achievable optimal payments) can be realized in a Bayesian Nash equilibrium. Additionally, the author designs a program implementation called ε-GroundedFairSIRBot, which utilizes robust program equilibrium to achieve conditional information disclosure and commitment, and demonstrates that under this implementation, a δ-Bayesian Nash equilibrium can be obtained (where δ can be controlled by the error probability ε).
Common Knowledge of Abstract Groups
Merlin Humml (Friedrich-Alexander-Universität Erlangen-Nürnberg), Lutz Schröder (Friedrich-Alexander-Universität Erlangen-Nürnberg)
🎯 What it does: An abstract group epistemic logic (AGEL) is proposed to describe the common knowledge of agent groups defined by attributes;
Communication-Efficient Collaborative Best Arm Identification
Nikolai Karpov (Indiana University Bloomington), Qin Zhang (Indiana University Bloomington)
Recommendation SystemOptimizationComputational EfficiencyTabular
🎯 What it does: A top-m arm identification algorithm under the Collaborative MAB model is proposed, aiming to significantly reduce communication costs while maintaining a high speedup (close to K times that of a single machine).
Compact Transformer Tracker with Correlative Masked Modeling
Zikai Song (Huazhong University of Science and Technology), Wei Yang (La Trobe University)
Object TrackingTransformerImageVideo
🎯 What it does: A compact Transformer tracker (CTTrack) based on raw self-attention is proposed, with a pluggable mutual mask decoder added during the training phase to enhance feature representation capabilities.
Competition or Cooperation? Exploring Unlabeled Data via Challenging Minimax Game for Semi-supervised Relation Extraction
Yu Hong (Fudan University), Wei Wang (Fudan University)
ClassificationGenerationTransformerGenerative Adversarial NetworkContrastive LearningText
🎯 What it does: This paper proposes a competitive semi-supervised relation extraction framework called AdvSRE, which constructs an extreme minimax game on unlabeled text, utilizing the adversarial interplay between a generator and a discriminator to fully exploit diverse relational expressions.
Competition, Alignment, and Equilibria in Digital Marketplaces
Meena Jagadeesan (University of California), Nika Haghtalab (University of California)
Recommendation SystemOptimizationReinforcement Learning
🎯 What it does: This paper studies the competition and user welfare alignment of digital market platforms when using reinforcement learning algorithms through a theoretical model.
Complement Sparsification: Low-Overhead Model Pruning for Federated Learning
Xiaopeng Jiang (New Jersey Institute of Technology), Cristian Borcea (New Jersey Institute of Technology)
Federated LearningImageTextBenchmark
🎯 What it does: Proposes the Complement Sparsification (CS) mechanism, which generates complementary sparse models through collaboration between the server and clients in federated learning, reducing communication and computation costs while maintaining approximate model performance.
Complex Dynamic Neurons Improved Spiking Transformer Network for Efficient Automatic Speech Recognition
Qingyu Wang (Chinese Academy of Sciences), Bo Xu (Chinese Academy of Sciences)
RecognitionComputational EfficiencySpiking Neural NetworkTransformerAudio
🎯 What it does: A pulse Transformer network (DyTr-SNN) that combines complex dynamic neurons is proposed for efficient automatic speech recognition.
Complexity of Probabilistic Inference in Random Dichotomous Hedonic Games
Saar Cohen (Bar-Ilan University), Noa Agmon (Bar-Ilan University)
Graph
🎯 What it does: This paper studies the probabilistic issues of inferring stability and optimality in Randomly Participating Dichotomous Benefit Games (RDHG) and provides various complexity results.
Complexity of Reasoning with Cardinality Minimality Conditions
Nadia Creignou (Aix Marseille University), Johannes Schmidt (Jönköping University)
🎯 What it does: In this paper, the author provides a complete complexity classification of the SAT variant CARDMINSAT based on cardinality minimization conditions within the Schaefer framework, determining when it can be solved in polynomial time and when it is ΘP₂-complete;
Complexity of Safety and coSafety Fragments of Linear Temporal Logic
Alessandro Artale (Free University of Bozen-Bolzano), Angelo Montanari (University of Udine)
Computational Efficiency
🎯 What it does: This paper studies the complexity of the satisfiability, validity, and realizability problems of the safety and liveness sublanguages of linear temporal logic (LTL) under infinite and finite execution traces, providing a complete classification of different complexities such as polynomial space, exponential time, and polynomial time.
Compositional Prototypical Networks for Few-Shot Classification
Qiang Lyu (University of Chinese Academy of Sciences), Weiqiang Wang (University of Chinese Academy of Sciences)
ClassificationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: Proposed a Compositional Prototypical Networks that learns attribute-level component prototypes and reuses these prototypes to construct category prototypes in few-shot classification.
Compressed Decentralized Learning of Conditional Mean Embedding Operators in Reproducing Kernel Hilbert Spaces
Boya Hou (University of Illinois Urbana Champaign), Subhonmesh Bose (University of Illinois Urbana Champaign)
Federated LearningReinforcement LearningTime Series
🎯 What it does: This paper proposes a distributed learning algorithm based on consistency compression for collaboratively approximating the Conditional Mean Embedding (CME) operator in a network.
Compressed Heterogeneous Graph for Abstractive Multi-Document Summarization
Miao Li (University of Melbourne), Jey Han Lau (University of Melbourne)
Graph Neural NetworkTransformerTextGraph
🎯 What it does: A multi-document summarization model based on compressed heterogeneous graphs, HGSUM, is proposed. It constructs a heterogeneous graph using three types of nodes (words, sentences, documents) and various edge types, and encodes it through a graph attention network. The compressed output is then used as input for the text decoder.
Compressing Transformers: Features Are Low-Rank, but Weights Are Not!
Hao Yu (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationObject DetectionSegmentationCompressionKnowledge DistillationTransformerImageText
🎯 What it does: A Transformer compression framework based on feature low-rank approximation is proposed, utilizing AFM, AAFM, and GFM to achieve unsupervised and few-shot model compression.
Computably Continuous Reinforcement-Learning Objectives Are PAC-Learnable
Cambridge Yang (Massachusetts Institute of Technology), Michael Carbin (Massachusetts Institute of Technology)
Reinforcement Learning
🎯 What it does: This paper proposes a general sufficient condition for the PAC learnability of reinforcement learning objectives and proves the PAC learnability of the objective by unifying continuity (from an information-theoretic perspective) and computability (from a computational-theoretic perspective). The framework is then applied to three types of existing objectives, including reward machines and LTL-in-the-limit, for validation.
Computing Divergences between Discrete Decomposable Models
Loong Kuan Lee (Monash University), Geoffrey I. Webb (Monash University)
GraphTabular
🎯 What it does: This paper proposes an exact method for calculating the αβ-divergence (including KL divergence, Hellinger distance, Chi-squared divergence, etc.) between two decomposable models (decomposable Markov networks) and presents the corresponding Junction Forest Computation (JFComp) algorithm.
Conceptual Reinforcement Learning for Language-Conditioned Tasks
Shaohui Peng (Institute of Computing Technology, Chinese Academy of Sciences), Yunji Chen (Institute of Computing Technology, Chinese Academy of Sciences)
Reinforcement LearningText
🎯 What it does: This paper proposes a Concept Reinforcement Learning (CRL) framework that constructs language-conditioned policies by learning transferable, compact, and interpretable concepts.
Concurrent Multi-Label Prediction in Event Streams
Xiao Shou (Rensselaer Polytechnic Institute), Kristin P. Bennett (Rensselaer Polytechnic Institute)
ClassificationAnomaly DetectionTransformerTime SeriesSequentialBiomedical DataElectronic Health RecordsFinance Related
🎯 What it does: The TCMBN model is proposed, which combines Transformer and Conditional Mixture Bernoulli Network to predict concurrent multi-labels in event streams, and learns a real-time label graph structure through least squares sparse precision matrix learning.
Conditional Diffusion Based on Discrete Graph Structures for Molecular Graph Generation
Han Huang (Beihang University), Weifeng Lv (Beihang University)
GenerationData SynthesisDrug DiscoveryGraph Neural NetworkDiffusion modelGraphStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A conditional diffusion model based on discrete graph structures is proposed for molecular graph generation.
Conditional Syntax Splitting for Non-monotonic Inference Operators
Jesse Heyninck (Open Universiteit), Christoph Beierle (FernUniversität in Hagen)
🎯 What it does: This paper introduces the concept of Conditional Syntax Splitting, which extends traditional Syntax Splitting to support conditional independence. It demonstrates that Lexicographic reasoning and System W satisfy Conditional Syntax Splitting and explains the Drowning Effect as a violation of Conditional Syntax Splitting, revealing the relationship between the Drowning Effect and conditional independence.
Confidence-Aware Training of Smoothed Classifiers for Certified Robustness
Jongheon Jeong (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
ClassificationAdversarial AttackConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: A sample confidence adaptive random smoothing training method (CAT-RS) is proposed, which enhances confidence calibration and adversarial robustness by dynamically controlling the robustness target of each sample.
Consensus Learning for Cooperative Multi-Agent Reinforcement Learning
Zhiwei Xu (Institute of Automation Chinese Academy of Sciences), Guoliang Fan (Institute of Automation Chinese Academy of Sciences)
Knowledge DistillationReinforcement LearningContrastive LearningSequential
🎯 What it does: The COLA framework is proposed, achieving multi-agent collaborative consensus without communication through contrastive learning and knowledge distillation, with consensus serving as an additional input to enhance the performance of CTDE algorithms.
Constrained Market Share Maximization by Signal-Guided Optimization
Bo Hui (Auburn University), Wei-Shinn Ku (Auburn University)
OptimizationGraph Neural NetworkReinforcement LearningTabular
🎯 What it does: This paper proposes a method to maximize an airline's market share by adjusting flight frequency under budget constraints.
Constrained Submodular Optimization for Vaccine Design
Zheng Dai (Massachusetts Institute of Technology), David K. Gifford (Massachusetts Institute of Technology)
OptimizationDrug DiscoveryBiomedical Data
🎯 What it does: A reduction gain optimization framework based on constrained submodular functions (Optivax-P) is proposed for designing peptide vaccines under redundancy and similarity constraints.
Constraint Optimization over Semirings
A. Pavan (Iowa State University), Arnab Bhattacharyya (National University of Singapore)
Optimization
🎯 What it does: This paper defines the problem of optimal interpretation (optSemVal) and the optimal interpretation itself (optSem) of propositional formulas over a given semiring, with a particular focus on the Viterbi semiring. It systematically analyzes its computational complexity, approximability, and its relationship with MaxSat.
Context-Aware Safe Medication Recommendations with Molecular Graph and DDI Graph Embedding
Qianyu Chen (Beijing Institute of Technology), Mingzhong Wang (University of the Sunshine Coast)
Recommendation SystemDrug DiscoveryGraph Neural NetworkGraphBiomedical DataElectronic Health Records
🎯 What it does: A drug recommendation framework CARMEN based on context-aware graph neural networks is proposed, integrating patient history, molecular graphs, and drug-drug interaction graphs.
Context-Aware Transformer for 3D Point Cloud Automatic Annotation
Xiaoyan Qian (University of Hong Kong), Ngai Wong (University of Hong Kong)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: An end-to-end model based on Transformer (CAT) is used to automatically generate high-quality 3D box annotations from weak labels containing only 2D bounding boxes, replacing traditional multi-stage segmentation, generation, and multimodal fusion processes.
ConTextual Masked Auto-Encoder for Dense Passage Retrieval
Xing Wu (Institute of Information Engineering Chinese Academy of Sciences), Songlin Hu (Kuaishou Technology)
RetrievalTransformerAuto EncoderText
🎯 What it does: A new generative pre-training method called CoT-MAE is proposed, which utilizes contextual information across text spans for self-supervised and context-supervised masked autoencoding to enhance the quality of text representations in dense retrieval.
Continual Graph Convolutional Network for Text Classification
Tiandeng Wu (Huawei Technologies), Jiandong Ding (Huawei Technologies)
ClassificationGraph Neural NetworkLarge Language ModelContrastive LearningText
🎯 What it does: A continuous learning graph convolutional network, ContGCN, is proposed for online text classification.
Continual Learning with Scaled Gradient Projection
Gobinda Saha (Purdue University), Kaushik Roy (Purdue University)
ClassificationReinforcement LearningImage
🎯 What it does: A continuous learning method based on gradient projection is proposed—Scaled Gradient Projection (SGP), which achieves a balance between learning new tasks and retaining old tasks by scaling gradients in the important gradient space of old tasks according to their importance.
Continual Variational Autoencoder via Continual Generative Knowledge Distillation
Fei Ye (University of York), Adrian G. Bors (University of York)
GenerationKnowledge DistillationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised, task-agnostic continuous generative model framework that utilizes short-term memory (STM) and long-term memory (teacher) to achieve continuous learning from infinite data streams through a knowledge incremental assimilation mechanism and continuous generative knowledge distillation.
Continuous Mixtures of Tractable Probabilistic Models
Alvaro H.C. Correia (Eindhoven University of Technology), Robert Peharz (Eindhoven University of Technology)
GenerationData SynthesisComputational EfficiencyImageTabular
🎯 What it does: This study investigates a hybrid model that combines continuous latent space with probabilistic circuits (PC), utilizing numerical integration (such as randomized quasi-Monte Carlo) to transform continuous mixtures into executable PCs, thereby achieving efficient probabilistic inference and density estimation.
Continuous Trajectory Generation Based on Two-Stage GAN
Wenjun Jiang (Beihang University), Jiawei Jiang (Beihang University)
GenerationData SynthesisAutonomous DrivingRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningGenerative Adversarial NetworkTime SeriesSequential
🎯 What it does: This paper proposes TS-TrajGen, a two-stage generative adversarial network that generates continuous trajectories on road networks by combining A* heuristics and neural networks.
ContraFeat: Contrasting Deep Features for Semantic Discovery
Xinqi Zhu (University of Sydney), Dacheng Tao (University of Sydney)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: The ContraFeat model is proposed and implemented, which can automatically discover decoupled and editable semantic directions in the W space of StyleGAN, eliminating the need for manual hierarchical selection steps.
Contrastive Attention Networks for Attribution of Early Modern Print
Nikolai Vogler (University of California), Taylor Berg-Kirkpatrick (Carnegie Mellon University)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A metric learning model based on contrastive attention is proposed to automatically identify the similarity of damaged prints in early modern printed books, thereby inferring the printer.
Contrastive Classification and Representation Learning with Probabilistic Interpretation
Rahaf Aljundi (Toyota Motor Europe), Daniel Olmeda Reino (Toyota Motor Europe)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study investigates a joint learning representation and classifier loss, combining the robustness of contrastive learning with the probabilistic interpretation of cross-entropy.
Contrastive Identity-Aware Learning for Multi-Agent Value Decomposition
Shunyu Liu (Zhejiang University), Mingli Song (Zhejiang University)
Reinforcement LearningContrastive LearningSequentialBenchmark
🎯 What it does: Proposes a Contrastive Identity-Aware (CIA) learning module that utilizes contrastive learning to enhance the distinguishability of credit allocation among different agents in the Value Decomposition (VD) network, thereby promoting diverse behaviors in multi-agent collaboration;
Contrastive Learning Reduces Hallucination in Conversations
Weiwei Sun (Shandong University), Zhaochun Ren (Shandong University)
GenerationKnowledge DistillationTransformerLarge Language ModelContrastive LearningText
🎯 What it does: A hybrid strategy based on contrastive learning, MixCL, is proposed to reduce the hallucination generation of pre-trained language models in knowledge-driven dialogues.
Contrastive Learning with the Feature Reconstruction Amplifier
Wentao Cui (Shanxi University), Liang Bai (Shanxi University)
Representation LearningContrastive LearningImage
🎯 What it does: A general Feature Reconstruction Amplifier (FRA) module is proposed and validated, which enhances representation learning by injecting high-dimensional reconstruction information into low-dimensional features within a contrastive learning framework.
Contrastive Masked Autoencoders for Self-Supervised Video Hashing
Yuting Wang (Tsinghua University), Shu-Tao Xia (Tsinghua University)
RetrievalTransformerAuto EncoderContrastive LearningVideo
🎯 What it does: A single-stage self-supervised video hashing method called ConMH is proposed, which directly generates high-quality binary hash codes through a high ratio of temporal masking autoencoders and debiased contrastive learning.
Contrastive Multi-Task Dense Prediction
Siwei Yang (Tongji University), Dan Xu (Hong Kong University of Science and Technology)
Object DetectionSegmentationContrastive LearningImage
🎯 What it does: A regularization method based on cross-task contrastive consistency is proposed to achieve unified learning and inference in a multi-task dense prediction framework.
Contrastive Open Set Recognition
Baile Xu (Nanjing University), Jian Zhao (Nanjing University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: This paper proposes a method for open set recognition based on supervised contrastive learning called ConOSR, which enhances feature quality using soft target contrastive learning and enables the detection of unknown samples.
Contrastive Pre-training with Adversarial Perturbations for Check-In Sequence Representation Learning
Letian Gong (Beijing Jiaotong University), Huaiyu Wan (Beijing Jiaotong University)
Recommendation SystemRepresentation LearningRecurrent Neural NetworkContrastive LearningTime SeriesSequential
🎯 What it does: This paper proposes a contrastive pre-training based checkpoint sequence representation learning method (CACSR), which generates more challenging positive and negative samples through spatial-temporal augmentation and adversarial perturbations in the latent space, thereby learning mobile trajectory representations with higher-level semantics.
Contrastive Predictive Autoencoders for Dynamic Point Cloud Self-Supervised Learning
Xiaoxiao Sheng (Shanghai Jiao Tong University), Gang Xiao (Shanghai Jiao Tong University)
RecognitionRepresentation LearningTransformerAuto EncoderContrastive LearningVideoPoint Cloud
🎯 What it does: A unified Contrastive Prediction and Reconstruction (CPR) self-supervised framework is proposed for feature learning of dynamic point cloud sequences.
Controllable Image Captioning via Prompting
Ning Wang (Huawei Inc), Linlin Li (Huawei Inc)
GenerationTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: A prompt learning-based image description framework is proposed, which can control the generated description style, length, and domain through different prompts.
Controlling Class Layout for Deep Ordinal Classification via Constrained Proxies Learning
Cong Wang (Nanjing University), Qing Gu (Nanjing University)
ClassificationAgentic AIImage
🎯 What it does: A framework for deep ordinal classification through constrained proxy layout is proposed, called Constrained Proxies Learning (CPL).
Converge to the Truth: Factual Error Correction via Iterative Constrained Editing
Jiangjie Chen (Fudan University), Yanghua Xiao (Fudan University)
TransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a fact error correction method called VENCE based on iterative constraint editing, which utilizes the gradient of a fact verification model to guide editing positions and incorporates authenticity scores into the energy function, enabling minimal edits to correct factual errors under unsupervised conditions.
ConvMatch: Rethinking Network Design for Two-View Correspondence Learning
Shihua Zhang (Wuhan University), Jiayi Ma (Wuhan University)
Pose EstimationConvolutional Neural NetworkOptical FlowImage
🎯 What it does: This paper proposes the ConvMatch network, which utilizes convolutional neural networks to learn the correspondence between two views for more accurate outlier rejection and geometric estimation.
ConvNTM: Conversational Neural Topic Model
Hongda Sun (Renmin University of China), Rui Yan (Renmin University of China)
Graph Neural NetworkTransformerAuto EncoderText
🎯 What it does: This paper proposes a Conversational Neural Topic Model (ConvNTM) for dialogue scenarios, capturing multi-turn structures through hierarchical encoding and modeling the interactions between speakers and audiences using a multi-role graph network to achieve utterance-level topic distributions. Additionally, a word co-occurrence constraint is introduced as an auxiliary objective to enhance topic coherence.
Cooperative and Adversarial Learning: Co-enhancing Discriminability and Transferability in Domain Adaptation
Hui Sun (Nanjing University), Ming Li (Nanjing University)
Domain AdaptationConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImage
🎯 What it does: A Cooperative and Adversarial Learning (CALE) framework is proposed, which achieves the mutual cooperation and confrontation of discriminative and transfer modules by generating easy and hard samples, unifying the optimization objectives of both.
CoopInit: Initializing Generative Adversarial Networks via Cooperative Learning
Yang Zhao (Baidu Research), Ping Li (Baidu Research)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes CoopInit, a GAN initialization strategy based on cooperative learning, which first trains the discriminator and generator using an energy-based model (EBM) and maximum likelihood estimation (MLE) during the cooperative phase to cover data patterns, and then switches to traditional adversarial training.
CoordFill: Efficient High-Resolution Image Inpainting via Parameterized Coordinate Querying
Weihuang Liu (University of Macau), Jue Wang (Tencent AI Lab)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: By generating spatially adaptive parameters through downsampling high-resolution images, and then using implicit representation of coordinate queries (multi-layer perceptron) to reconstruct missing areas pixel by pixel, efficient high-resolution image inpainting is achieved.
Coordinate Descent Methods for DC Minimization: Optimality Conditions and Global Convergence
Ganzhao Yuan (Peng Cheng Laboratory)
OptimizationTabular
🎯 What it does: This paper proposes a Coordinate Descent algorithm based on Sequential Nonconvex Approximation (CD-SNCA) to solve the Differential Convex (DC) minimization problem, and proves its convergence to a saddle point in the coordinate dimension, achieving Q-linear convergence under the assumption of error bounds.
CoP: Factual Inconsistency Detection by Controlling the Preference
Shuaijie She (Nanjing University), Jiajun Chen (Nanjing University)
GenerationAnomaly DetectionTransformerPrompt EngineeringText
🎯 What it does: This paper proposes an unsupervised factual consistency detection framework called CoP, which incorporates prompts in an additional reasoning step to control the preferences of the generative model, thereby calculating probability differences to identify factual inconsistencies in summaries.
Copyright-Certified Distillation Dataset: Distilling One Million Coins into One Bitcoin with Your Private Key
Tengjun Liu (Fudan University), Wanxuan Gu (NVIDIA)
ClassificationKnowledge DistillationMixture of ExpertsImage
🎯 What it does: A method is proposed to embed watermarks in the dataset distillation process, allowing models trained on the distilled dataset to maintain original functionality while carrying verifiable copyright watermarks.
Correct for Whom? Subjectivity and the Evaluation of Personalized Image Aesthetics Assessment Models
Samuel Goree (Indiana University), David J. Crandall (Indiana University)
Image
🎯 What it does: This study investigates the subjectivity issue in personalized image aesthetic quality assessment and conducts empirical analysis using the PR-AADB dataset.
Correlation Loss: Enforcing Correlation between Classification and Localization
Fehmi Kahraman (Middle East Technical University), Emre Akbas (Middle East Technical University)
Object DetectionImage
🎯 What it does: A loss function for directly optimizing the correlation between classification and localization (Correlation Loss) is proposed and validated, and it is applied as a plugin to various object detectors based on NMS and NMS-free, significantly improving detection accuracy.
Correspondence-Free Domain Alignment for Unsupervised Cross-Domain Image Retrieval
Xu Wang (Sichuan University), Peng Hu (Centre for Frontier AI Research)
RetrievalDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised cross-domain image retrieval method called CoDA, which can retrieve images from different domains without category labels and corresponding relationships.
Corruption-Tolerant Algorithms for Generalized Linear Models
Bhaskar Mukhoty (Mohamed Bin Zayed University of Artificial Intelligence), Purushottam Kar (Indian Institute of Technology Kanpur)
ClassificationOptimizationTabular
🎯 What it does: The SVAM (Sequential Variance-Altered MLE) framework is proposed to uniformly address robust learning of generalized linear models (GLM) under adversarially tampered training data labels, covering least squares, logistic regression, gamma regression, and more.
Counterfactual Dynamics Forecasting – a New Setting of Quantitative Reasoning
Yanzhu Liu (Institute for Infocomm Research and Centre for Frontier AI Research, A*STAR), Joo-Hwee Lim (Institute for Infocomm Research and Centre for Frontier AI Research, A*STAR)
Time SeriesSequentialPhysics RelatedOrdinary Differential Equation
🎯 What it does: This study investigates the problem of counterfactual dynamics prediction, using deep neural networks to simulate multi-agent dynamic systems and predict future trajectories under different counterfactual interventions.
Counterfactual Learning with General Data-Generating Policies
Yusuke Narita (Yale University), Kohei Yata (University of Wisconsin-Madison)
Reinforcement LearningTabular
🎯 What it does: A method is proposed for offline policy evaluation (OPE) of missing support (deterministic) logging strategies, utilizing Approximate Propensity Score (APS) to achieve local estimation of action mean differences.
Coupling Artificial Neurons in BERT and Biological Neurons in the Human Brain
Xu Liu (Northwestern Polytechnical University), Xintao Hu
TransformerLarge Language ModelTextTime SeriesMagnetic Resonance Imaging
🎯 What it does: A framework is proposed that couples fine-grained artificial neurons in the Transformer model (the query/key dimensions of each multi-head self-attention in BERT) with functional neural networks in the human brain (functional brain networks obtained through fMRI), establishing a correspondence between the two using temporal activation synchronization, and providing a neurolinguistic interpretation of the coupling results using semantic information such as part-of-speech tags.
Covariate-Shift Generalization via Random Sample Weighting
Yue He (Tsinghua University), Peng Cui (Tsinghua University)
Domain AdaptationOptimizationTabular
🎯 What it does: The study investigates covariate shift generalization under model underfitting conditions, proposing a Random Sample Weighting (RSW) method that achieves robust predictions by simulating heterogeneous environments and combining it with invariant learning.
CowClip: Reducing CTR Prediction Model Training Time from 12 Hours to 10 Minutes on 1 GPU
Zangwei Zheng (National University of Singapore), Yang You (Bytedance Inc.)
Recommendation SystemOptimizationComputational EfficiencyTabular
🎯 What it does: In the CTR prediction task, the feasibility of large-batch training was studied, and the CowClip algorithm was proposed to significantly shorten single-GPU training time.
CP-Rec: Contextual Prompting for Conversational Recommender Systems
Keyu Chen (East China Normal University), Shiliang Sun (East China Normal University)
Recommendation SystemTransformerPrompt EngineeringText
🎯 What it does: A context-aware conversational recommendation system called CP-Rec is proposed, which can achieve task planning and topic expansion in multi-task dialogues while utilizing knowledge graphs to enhance user profiles and recommendation quality.
CRAFT: Camera-Radar 3D Object Detection with Spatio-Contextual Fusion Transformer
Youngseok Kim (Korea Advanced Institute of Science and Technology), Dongsuk Kum (Korea Advanced Institute of Science and Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A camera-radar 3D object detection framework called CRAFT based on early fusion is proposed.
Crafting Monocular Cues and Velocity Guidance for Self-Supervised Multi-Frame Depth Learning
Xiaofeng Wang (Institute of Automation, Chinese Academy of Sciences), Xingang Wang (PhiGent Robotics)
Depth EstimationAutonomous DrivingImageBenchmark
🎯 What it does: A self-supervised multi-frame deep learning framework MOVEDepth is proposed, which constructs a cost volume using monocular depth prioritization and camera speed guidance to integrate multi-view geometric information, significantly improving depth accuracy.
CRIN: Rotation-Invariant Point Cloud Analysis and Rotation Estimation via Centrifugal Reference Frame
Yujing Lou (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
ClassificationSegmentationPose EstimationPoint Cloud
🎯 What it does: This paper proposes the Centrifugal Rotation-Invariant Network (CRIN), which achieves rotation-invariant feature extraction of point clouds through a rotation-invariant 'Centrifugal Reference Frame' (CRF), and builds continuous rotation distribution, attention-based downsampling, and relational modules on this basis, ultimately achieving unsupervised rotation estimation.
Cross-Category Highlight Detection via Feature Decomposition and Modality Alignment
Zhenduo Zhang (Tencent)
RecognitionDomain AdaptationTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: A cross-category video highlight detection framework is proposed, which achieves model transfer across video categories by learning category-independent highlight features.
Cross-Domain Adaptative Learning for Online Advertisement Customer Lifetime Value Prediction
Hongzu Su (University of Electronic Science and Technology of China), Ke Lu (Shandong Normal University)
Domain AdaptationRecommendation SystemTabular
🎯 What it does: A Cross-Domain Adaptation Framework (CDAF) is proposed to alleviate the data scarcity problem in the target domain by leveraging rich data from the source domain, and to enhance the customer lifetime value (LTV) prediction on online advertising platforms using domain-invariant information.
Cross-Domain Few-Shot Graph Classification with a Reinforced Task Coordinator
Qiannan Zhang (King Abdullah University of Science and Technology), Xiangliang Zhang (University of Notre Dame)
ClassificationMeta LearningGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes a cross-domain few-shot graph classification framework (CDTC) that utilizes a small number of labeled target domain samples as prompt tasks, constructs a task bipartite graph, and achieves adaptive meta-task selection through reinforcement learning to improve the generalization ability of cross-domain graph classification.
Cross-Domain Graph Anomaly Detection via Anomaly-Aware Contrastive Alignment
Qizhou Wang (Monash University), Christopher Leckie (University of Melbourne)
Domain AdaptationAnomaly DetectionGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A cross-domain graph anomaly detection method (ACT) is proposed, which achieves efficient anomaly detection on unlabeled target graphs through unsupervised contrastive learning on the target graph and joint optimization of anomaly-aware domain alignment with the representation of the target graph.
Cross-Modal Contrastive Learning for Domain Adaptation in 3D Semantic Segmentation
Bowei Xing (Peking University), Taiyan Chen (Peking University)
SegmentationDomain AdaptationAutonomous DrivingTransformerContrastive LearningImagePoint Cloud
🎯 What it does: A cross-modal contrastive learning and neighborhood feature aggregation module is proposed for unsupervised domain adaptation in 3D semantic segmentation.
Cross-Modal Distillation for Speaker Recognition
Yufeng Jin (Tongji University), Cairong Zhao (Tongji University)
RecognitionKnowledge DistillationMultimodalityAudio
🎯 What it does: This paper proposes a cross-modal distillation framework called VGSR, which uses a facial recognition model as a teacher to transfer recognition knowledge to a student model that only uses voice, thereby enhancing speaker recognition performance.
Cross-Modal Label Contrastive Learning for Unsupervised Audio-Visual Event Localization
Peijun Bao (Nanyang Technological University), Alex C. Kot (Nanyang Technological University)
RecognitionOptimizationContrastive LearningVideoMultimodalityAudio
🎯 What it does: A novel unsupervised audio-video event localization framework is proposed, which automatically generates pseudo-labels using cross-modal feature learning and label contrastive learning, and gradually optimizes the localization model through the EM algorithm.
Cross-Modality Earth Mover’s Distance for Visible Thermal Person Re-identification
Yongguo Ling (Xiamen University), Nicu Sebe (University of Trento)
RecognitionRetrievalConvolutional Neural NetworkImageMultimodalityBenchmark
🎯 What it does: A distribution alignment method based on Cross-Modal Earth Mover's Distance (CM-EMD) is proposed, combined with CM-DL and MGS to enhance the performance of visible-thermal person re-identification.
Cross-Modality Person Re-identification with Memory-Based Contrastive Embedding
De Cheng (Xidian University), Xinbo Gao (University of Science and Technology of China)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: A memory-augmented cross-modal deep metric learning framework is proposed for the task of person re-identification between visible light and infrared images.
Cross-View Geo-Localization via Learning Disentangled Geometric Layout Correspondence
Xiaohan Zhang (University of Vermont), Safwan Wshah (University of Science and Technology of China)
RetrievalDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a cross-view geographic localization model named GeoDTR, which can separate low-level visual features from global geometric layout information and jointly represent them through a geometric layout extractor.
Crowd-Level Abnormal Behavior Detection via Multi-Scale Motion Consistency Learning
Linbo Luo (Xidian University), Wentong Cai (Nanyang Technological University)
Anomaly DetectionGraph Neural NetworkOptical FlowVideo
🎯 What it does: An unsupervised video anomaly detection method for detecting abnormal behaviors in large crowds (such as counterflow and congested turbulence) is proposed.
CrysGNN: Distilling Pre-trained Knowledge to Enhance Property Prediction for Crystalline Materials
Kishalay Das (Indian Institute of Technology Kharagpur), Niloy Ganguly (Indo Korea Science and Technology Center)
Knowledge DistillationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This study proposes the CrysGNN pre-training framework, which utilizes 800K unlabeled crystal graph structures to self-supervised learn localized chemical features and global crystal structures at the node and graph levels, respectively, and injects the pre-trained knowledge into existing material property prediction models through knowledge distillation.
CSTAR: Towards Compact and Structured Deep Neural Networks with Adversarial Robustness
Huy Phan (Rutgers University), Saman Zonouz (Georgia Institute of Technology)
CompressionOptimizationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A CSTAR method is proposed that can simultaneously achieve model compression, structuring, and adversarial robustness within a single training framework.
Curriculum Multi-Negative Augmentation for Debiased Video Grounding
Xiaohan Lan (Tsinghua University), Wenwu Zhu (Tsinghua University)
RecognitionRetrievalGraph Neural NetworkContrastive LearningVideoText
🎯 What it does: By employing multi-level negative sample augmentation (cross-video clip/video-level and self-shuffling masking) combined with multi-stage curriculum learning, we suppress the video localization model's dependence on the distribution bias of temporal annotations and enhance cross-modal semantic matching capabilities.