AAAI 2023 Papers with AI Summaries
AAAI Conference on Artificial Intelligence · 1578 papers
→ AAAI 2023 papers with code (696)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 1578 AAAI 2023 papers by keyword, author or institution —
free trial on arXivSub.
3D Assembly Completion
Weihao Wang (Tongji University), Bin He (Tongji University)
Pose EstimationRobotic IntelligenceTransformerPoint Cloud
🎯 What it does: A Transformer framework named FiT is proposed to complete incomplete 3D assembly tasks, specifically selecting missing components from a toolbox and predicting their 6-DoF poses to achieve a complete assembly.
3D Human Pose Lifting with Grid Convolution
Yangyuxuan Kang (Institute of Software, Chinese Academy of Sciences), Enhua Wu (University of Macau)
Pose EstimationConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: A 3D human pose enhancement network based on grid convolution is proposed, which significantly improves the regression effect from 2D to 3D poses by transforming irregular skeletal graphs into regular grids for convolutional learning.
3D-TOGO: Towards Text-Guided Cross-Category 3D Object Generation
Zutao Jiang (Xi'an Jiaotong University), Hang Xu (Huawei Noah's Ark Lab)
GenerationData SynthesisTransformerNeural Radiance FieldGenerative Adversarial NetworkContrastive LearningImageText
🎯 What it does: The 3D-TOGO model is proposed to achieve text description-driven cross-category 3D object generation.
A Benchmark and Asymmetrical-Similarity Learning for Practical Image Copy Detection
Wenhao Wang (University of Technology Sydney), Yi Yang (Zhejiang University)
RetrievalContrastive LearningImageBenchmark
🎯 What it does: This paper first constructs an image copy detection benchmark NDEC that includes a large number of hard negative samples, and proposes an Asymmetric Similarity Learning (ASL) method based on feature norm ratios to distinguish edited copies from hard-to-differentiate hard negative queries.
A Coreset Learning Reality Check
Fred Lu (Booz Allen Hamilton), James Holt (Laboratory for Physical Sciences)
OptimizationTabular
🎯 What it does: Systematically evaluated and compared the performance of various subsampling methods based on coresets and optimal subsampling in large-scale Logistic regression, using a unified experimental framework and three evaluation metrics.
A Data Source for Reasoning Embodied Agents
Jack Lanchantin (Meta AI), Arthur Szlam (Meta AI)
TransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: This paper proposes an automated data generator that can generate the world state, questions, and answers for embedded intelligent agents, providing two types of context representations (text sequences and graph structures) and baseline models for evaluation.
A Disentangled-Attention Based Framework with Persona-Aware Prompt Learning for Dialogue Generation
Pingsheng Liu (East China Normal University), Liang He (Hasso Plattner Institute)
GenerationTransformerPrompt EngineeringText
🎯 What it does: A dialogue generation framework based on decoupled attention is proposed, which combines character perception prompt learning and A* heuristic keyword constraints to achieve dynamic decoupling of character selection, character information expansion, and generation in dialogues.
A Domain-Knowledge-Inspired Music Embedding Space and a Novel Attention Mechanism for Symbolic Music Modeling
Zixun Guo (Singapore University of Technology and Design), Dorien Herremans (Singapore University of Technology and Design)
GenerationTransformerAudio
🎯 What it does: This paper proposes a Fundamental Music Embedding (FME) based on biased sine encoding and a RIPO attention mechanism that combines relative pitch and starting time, constructing a RIPO Transformer for modeling and generating symbolic music.
A Domain-Transfer Meta Task Design Paradigm for Few-Shot Slot Tagging
Fengyi Yang (Xinjiang Technical Institute of Physics and Chemistry), Abibulla Atawulla (Xinjiang Technical Institute of Physics and Chemistry)
Domain AdaptationMeta LearningTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes a domain transfer-based meta-task design paradigm and a language feature enhancement task adaptation network to address the overlapping slot issue in few-shot slot filling.
A Dynamics and Task Decoupled Reinforcement Learning Architecture for High-Efficiency Dynamic Target Intercept
Dora D. Liu (DeepBlue Academy of Sciences), Zhong Yuan Lai (University of Sydney)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningTime SeriesSequential
🎯 What it does: This paper proposes a DTTD architecture that decouples dynamic control from task planning for dynamic target interception by drones.
A Fair Generative Model Using LeCam Divergence
Soobin Um (KAIST), Changho Suh (KAIST)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A fair generative model is proposed that utilizes LeCam divergence constraints in the absence of sensitive attribute labels.
A Faster Practical Approximation Scheme for the Permanent
Juha Harviainen (University of Helsinki), Mikko Koivisto (University of Helsinki)
OptimizationComputational EfficiencyTabular
🎯 What it does: A new fast approximate matrix constant rejection sampling scheme is proposed, along with a new nested upper bound.
A Formal Metareasoning Model of Concurrent Planning and Execution
Amihay Elboher (Ben-Gurion University), Eyal Shimony (Ben-Gurion University)
OptimizationRobotic IntelligenceReinforcement LearningTabularBenchmark
🎯 What it does: This paper studies a meta-reasoning model called CoPE that can execute actions in parallel during the planning process and proposes various solving algorithms.
A Framework to Design Approximation Algorithms for Finding Diverse Solutions in Combinatorial Problems
Tesshu Hanaka (Kyushu University), Yota Otachi (Nagoya University)
Optimization
🎯 What it does: A general framework is proposed for designing approximate algorithms for diverse solutions;
A Generalized Scalarization Method for Evolutionary Multi-Objective Optimization
Ruihao Zheng (Southern University of Science and Technology), Zhenkun Wang (Southern University of Science and Technology)
Optimization
🎯 What it does: A general Lp scalarization method (GLp) is proposed and embedded into the global replacement strategy of MOEA/D to address the mismatch between subproblems and solutions caused by different Lp scalarizations.
A Generalized Unbiased Risk Estimator for Learning with Augmented Classes
Senlin Shu (Chongqing University), Lei Feng (Chongqing University)
ClassificationOptimizationImageTabular
🎯 What it does: This paper proposes an unbiased risk estimator compatible with any loss function for learning about augmented class problems that may appear in the testing phase but not in the training phase.
A Generative Approach for Script Event Prediction via Contrastive Fine-Tuning
Fangqi Zhu (Harbin Institute of Technology), Ruifeng Xu (Peng Cheng Laboratory)
GenerationTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: A two-stage generative model is proposed, using event-centered pre-training and contrastive fine-tuning to predict the next event in a script.
A Gift from Label Smoothing: Robust Training with Adaptive Label Smoothing via Auxiliary Classifier under Label Noise
Jongwoo Ko (KAIST), Se-Young Yun (KAIST)
ClassificationSupervised Fine-TuningImage
🎯 What it does: A framework named ALASCA is proposed, which utilizes adaptive label smoothing combined with auxiliary classifiers to achieve robust training in the presence of label noise.
A Graph Fusion Approach for Cross-Lingual Machine Reading Comprehension
Zenan Xu (Sun Yat-sen University), Daxin Jiang (Microsoft)
Graph Neural NetworkTransformerText
🎯 What it does: A Graph Fusion Model (GFMRC) is proposed, which constructs a graph of cross-language alignment and monolingual syntactic information, and learns the attention matrix in a Transformer to enhance cross-language reading comprehension performance.
A Latent-Variable Model for Intrinsic Probing
Karolina Stańczak (University of Copenhagen), Isabelle Augenstein (ETH Zürich)
TransformerReinforcement LearningText
🎯 What it does: A latent variable-based intrinsic detection model is proposed to locate a subset of neurons encoding linguistic properties in pre-trained contextual representations.
A Learnable Radial Basis Positional Embedding for Coordinate-MLPs
Sameera Ramasinghe (Amazon), Simon Lucey (University of Adelaide)
RestorationGenerationData SynthesisImagePoint Cloud
🎯 What it does: A learnable radial basis function position encoding method is proposed, achieving efficient position embedding for coordinate MLP through graph Laplacian regularization.
A Machine with Short-Term, Episodic, and Semantic Memory Systems
Taewoon Kim (Vrije Universiteit Amsterdam), Piek Vossen (Vrije Universiteit Amsterdam)
Recurrent Neural NetworkGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Designed and implemented a reinforcement learning agent with short-term, episodic, and semantic memory systems, maximizing question-and-answer rewards in the 'Room' environment through memory encoding, storage, and retrieval.
A Model-Agnostic Heuristics for Selective Classification
Andrea Pugnana (Scuola Normale Superiore), Salvatore Ruggieri (University of Pisa)
ClassificationImageTabular
🎯 What it does: This paper proposes a model-agnostic selective classification method SCROSS, which implements a rejection strategy for probabilistic binary classifiers through cross-fitting and sample-specific quantile estimation.
A Neural Span-Based Continual Named Entity Recognition Model
Yunan Zhang (Harbin Institute of Technology), Qingcai Chen (Peng Cheng Laboratory)
RecognitionKnowledge DistillationTransformerSupervised Fine-TuningText
🎯 What it does: A span-based continual learning named entity recognition model, SpanKL, is proposed to address the conflict between forward compatibility and backward compatibility in traditional sequence labeling methods.
A Noise-Tolerant Differentiable Learning Approach for Single Occurrence Regular Expression with Interleaving
Rongzhen Ye (Sun Yat-sen University), Pingjia Liang (Sun Yat-sen University)
Text
🎯 What it does: This paper proposes a noise-tolerant differentiable learning framework called SOIREDL for learning single occurrence interleaved regular expressions (SOIRE) from positive and negative samples.
A Pair-Approximation Method for Modelling the Dynamics of Multi-Agent Stochastic Games
Chen Chu (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
Reinforcement LearningTabularStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A method based on pair-approximation is proposed, constructing a partial differential equation model that describes the learning dynamics of infinitely large multi-agent systems in stochastic games based on Q-learning, capable of predicting the evolution of both strategy distribution and environmental state distribution over time.
A Parameterized Theory of PAC Learning
Cornelius Brand (TU Wien), Kirill Simonov (Hasso Plattner Institute)
🎯 What it does: The paper proposes a parameterized PAC learning theory and provides two definitions of fixed-parameter learnability, analyzing the learnability boundaries of CNF/DNF and graph problems.
A Proof That Using Crossover Can Guarantee Exponential Speed-Ups in Evolutionary Multi-Objective Optimisation
Duc-Cuong Dang (University of Passau), Dirk Sudholt (University of Passau)
Optimization
🎯 What it does: In multi-objective evolutionary optimization (EMO), researchers constructed a class of multi-objective problems RR MO and proved that the crossover operator can achieve exponential speedup in the GSEMO and NSGA-II algorithms; without crossover, the algorithm requires exponential time to find any Pareto optimal points.
A Provable Framework of Learning Graph Embeddings via Summarization
Houquan Zhou (Institute of Computing Technology, Chinese Academy of Sciences), Xueqi Cheng (Institute of Computing Technology, Chinese Academy of Sciences)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a node embedding learning framework based on graph summarization called GELSUMM, provides theoretical proofs, and presents closed-form recovery formulas for DeepWalk, LINE, and GCN, validating its effectiveness through experiments.
A Question-Answering Approach to Key Value Pair Extraction from Form-Like Document Images
Kai Hu (University of Science and Technology of China), Qiang Huo (Microsoft Research Asia)
TransformerImageText
🎯 What it does: A QA-based Transformer encoder-decoder model called KVPFormer is proposed to automatically extract key-value pair relationships from table-like document images.
A Scope Sensitive and Result Attentive Model for Multi-Intent Spoken Language Understanding
Lizhi Cheng (Shanghai Jiao Tong University), Weijia Jia (Beijing Normal University)
RecognitionTransformerTextAudio
🎯 What it does: This paper proposes an SSRAN model for multi-intent speech language understanding, aimed at joint slot filling and intent detection.
A Semi-parametric Model for Decision Making in High-Dimensional Sensory Discrimination Tasks
Stephen Keeley (Fordham University), Michael Shvartsman (Meta)
Tabular
🎯 What it does: A semi-parametric psychometric model is proposed, combining traditional unidimensional sigmoid parameterization with a Gaussian process (GP) prior for contextual dimensions beyond intensity;
A Set of Control Points Conditioned Pedestrian Trajectory Prediction
Inhwan Bae (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)
Graph Neural NetworkMixture of ExpertsGraph
🎯 What it does: A pedestrian trajectory prediction framework called Graph-TERN is proposed, based on multi-control points and multi-relation graph convolutional networks.
A Simple Baseline for Multi-Camera 3D Object Detection
Yunpeng Zhang (PhiGent Robotics), Jie Zhou (Tsinghua University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint CloudBenchmark
🎯 What it does: Proposes SimMOD, a two-stage multi-camera 3D object detection framework;
A Simple Unified Approach to Testing High-Dimensional Conditional Independences for Categorical and Ordinal Data
Ankur Ankan (Radboud University), Johannes Textor (Radboud University)
Tabular
🎯 What it does: This paper proposes a residual-based conditional independence (CI) testing method suitable for categorical and ordinal data.
A Simple Yet Effective Subsequence-Enhanced Approach for Cross-Domain NER
Jinpeng Hu (Shenzhen Research Institute of Big Data), Tsung-Hui Chang (Shenzhen Research Institute of Big Data)
RecognitionDomain AdaptationRecurrent Neural NetworkSupervised Fine-TuningText
🎯 What it does: A cross-domain named entity recognition model is proposed, which significantly improves cross-domain transfer effects by combining subsequence-level features with global features.
A Speaker Turn-Aware Multi-Task Adversarial Network for Joint User Satisfaction Estimation and Sentiment Analysis
Kaisong Song (Northeastern University), Xiaozhong Liu (Alibaba Group)
ClassificationRecommendation SystemOptimizationRecurrent Neural NetworkGenerative Adversarial NetworkText
🎯 What it does: This paper proposes the Speaker Turn-Aware Multi-Task Adversarial Network (STMAN), which achieves joint learning of dialogue-level user satisfaction estimation (USE) and turn-level sentiment analysis (SA) in service dialogues.
A Structural Complexity Analysis of Synchronous Dynamical Systems
Eduard Eiben (Royal Holloway University of London), Viktoriia Korchemna (TU Wien)
🎯 What it does: This paper studies three classic problems in synchronous dynamic systems (SyDS) - Reachability (REACHABILITY), Convergence (CONVERGENCE), and Convergence Guarantee (CONVERGENCE GUARANTEE), and systematically analyzes their computational difficulty from the perspective of parameterized complexity under structural parameters such as tree width and tree depth.
A Survey on Model Compression and Acceleration for Pretrained Language Models
Canwen Xu (University of California San Diego), Julian McAuley (University of California San Diego)
CompressionComputational EfficiencyKnowledge DistillationTransformerTextReview/Survey PaperBenchmark
🎯 What it does: This paper provides a systematic review of compression and acceleration techniques for pre-trained language models during the inference phase, covering various methods, metrics, benchmarks, and sustainability issues.
A Tale of Two Latent Flows: Learning Latent Space Normalizing Flow with Short-Run Langevin Flow for Approximate Inference
Jianwen Xie (Baidu Research), Ping Li (Baidu Research)
RestorationGenerationAnomaly DetectionFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Proposes the use of reversible flow models as a prior in latent space, and achieves approximate inference through short-run Langevin sampling, jointly training the latent flow and generative network via maximum likelihood.
A Vector Quantized Approach for Text to Speech Synthesis on Real-World Spontaneous Speech
Li-Wei Chen (Carnegie Mellon University), Alexander Rudnicky (Carnegie Mellon University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkAudio
🎯 What it does: A multi-codebook vector quantization TTS system (MQTTS) for spontaneous speech in the real world is proposed and implemented, enhancing synthesis quality through multi-codebook discrete representation, single-head cross-attention, monotonic alignment, and silent audio prompts.
Abstract Argumentation Framework with Conditional Preferences
Gianvincenzo Alfano (University of Calabria), Irina Trubitsyna (University of Calabria)
🎯 What it does: A new abstract argumentation framework is proposed - the Conditional Preference-based Argumentation Framework (CPAF), which combines traditional abstract argumentation frameworks (AF) with conditional preference rules to provide a method for selecting the 'best' extension based on conditional preferences during the argumentation process.
AC-Band: A Combinatorial Bandit-Based Approach to Algorithm Configuration
Jasmin Brandt (Paderborn University), Kevin Tierney (Bielefeld University)
OptimizationHyperparameter SearchTabular
🎯 What it does: Proposes AC-Band, an algorithm configuration method based on combinatorial multi-armed bandits, which can find near-optimal configurations within a given budget.
Accelerating the Training of Video Super-resolution Models
Lijian Lin (Tencent), Ying Shan (Tencent)
RestorationSuper ResolutionLarge Language ModelVideo
🎯 What it does: Using multi-grid training and large-batch training to accelerate the training process of video super-resolution (VSR) models.
Acceleration of Large Transformer Model Training by Sensitivity-Based Layer Dropping
Yujie Zeng (Samsung Research Institute China), Lin Chen (Samsung Research Institute China)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: This paper proposes a Transformer layer dropout method based on layer sensitivity (SBLD), which significantly accelerates the training of large-scale Transformer models while maintaining or improving accuracy.
Accommodating Audio Modality in CLIP for Multimodal Processing
Ludan Ruan (Renmin University of China), Qin Jin (Renmin University of China)
GenerationRetrievalTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Based on CLIP, CLIP4VLA is designed with an audio encoder that is consistent with the structure of the visual encoder, and distinguishes between speech and non-speech information in audio through audio type tokens; during the pre-training phase, cross-modal and single-modal contrastive learning is used to learn the associations between vision, text, and audio; subsequently, fine-tuning is performed on downstream tasks such as video retrieval and video caption generation.
ACE: Cooperative Multi-Agent Q-learning with Bidirectional Action-Dependency
Chuming Li (University of Sydney), Wanli Ouyang (University of Sydney)
Reinforcement LearningBenchmark
🎯 What it does: This paper proposes the ACE algorithm, which utilizes bidirectional action dependence to transform multi-agent non-stationary MMDP into a single-agent SE-MDP, thereby enabling training and decision-making for multi-agents through standard Q-learning or PPO.
Achieving Zero Constraint Violation for Constrained Reinforcement Learning via Conservative Natural Policy Gradient Primal-Dual Algorithm
Qinbo Bai (Purdue University), Vaneet Aggarwal (University of Maryland)
OptimizationReinforcement LearningTabular
🎯 What it does: A new Conservative Natural Policy Gradient Primal-Dual algorithm (CNPGPD) is proposed, aimed at achieving zero constraint violation while achieving state-of-the-art results in the convergence of the objective value function.
ACL-Net: Semi-supervised Polyp Segmentation via Affinity Contrastive Learning
Huisi Wu (Shenzhen University), Xinrong Guo (Shenzhen University)
SegmentationContrastive LearningImage
🎯 What it does: A semi-supervised colorectal polyp segmentation framework called ACL-Net is proposed, which utilizes affinity map alignment between student and teacher networks to continuously optimize pseudo-labels, thereby improving segmentation performance.
Action-Conditioned Generation of Bimanual Object Manipulation Sequences
Haziq Razali (Imperial), Yiannis Demiris (Imperial)
GenerationRobotic IntelligenceRecurrent Neural NetworkGraph Neural NetworkTime SeriesSequential
🎯 What it does: A modular neural network is proposed to generate complete 3D motion sequences for two-handed manipulation of objects based on given semantic action labels, including wrist and object trajectories, finger joint angles, and other body postures.
Actional Atomic-Concept Learning for Demystifying Vision-Language Navigation
Bingqian Lin (Sun Yat-sen University), Jianzhuang Liu (Huawei Noah's Ark Lab)
Robotic IntelligenceTransformerReinforcement LearningContrastive LearningMultimodality
🎯 What it does: This paper proposes an Actional Atomic-Concept Learning (AACL) framework that maps each visual observation to natural language phrases composed of actions and objects (actional atomic concepts). It extracts object concepts using CLIP, maps action concepts through action direction, and then reorders object concepts using a concept refinement adapter. Finally, it regularizes visual features with concept features in the observation co-embedding module to achieve better cross-modal alignment.
Actionness Inconsistency-Guided Contrastive Learning for Weakly-Supervised Temporal Action Localization
Zhilin Li (University of Science and Technology of China), Qinying Liu (University of Science and Technology of China)
RecognitionContrastive LearningVideo
🎯 What it does: A contrastive learning method based on two-branch inconsistency guidance, AICL, is proposed for weakly supervised temporal action localization.
Active Token Mixer
Guoqiang Wei (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: An adaptive token-mixing mechanism called Active Token Mixer (ATM) is proposed, and the ATMNet backbone and ATMFPN neck are constructed around it.
AdaBoost.C2: Boosting Classifiers Chains for Multi-Label Classification
Jiaxuan Li (Xi'an Jiaotong University), Jiayin Wang (Xi'an Jiaotong University)
ClassificationSupervised Fine-TuningTabular
🎯 What it does: Proposes the AdaBoost.C2 multi-path AdaBoost framework, combining AdaBoost with Classifier Chains for multi-label classification;
AdaCM: Adaptive ColorMLP for Real-Time Universal Photo-Realistic Style Transfer
Tianwei Lin (Baidu Inc.), Yong Liu (Zhejiang University)
Image TranslationImage
🎯 What it does: Proposes the AdaCM framework to achieve real-time high-resolution global photo-realistic style transfer;
AdapSafe: Adaptive and Safe-Certified Deep Reinforcement Learning-Based Frequency Control for Carbon-Neutral Power Systems
Xu Wan (Zhejiang University), Fei Teng (Imperial College London)
OptimizationReinforcement LearningTime Series
🎯 What it does: An adaptive and secure authentication deep reinforcement learning frequency control framework called AdapSafe is proposed for real-time frequency regulation in carbon-neutral power grids.
Adapting Object Size Variance and Class Imbalance for Semi-supervised Object Detection
Yuxiang Nie (Sun Yat-sen University), Guanbin Li (Zhejiang Lab)
Object DetectionKnowledge DistillationImage
🎯 What it does: A semi-supervised object detection framework based on adaptive pseudo-labels and multi-scale feature/prediction consistency learning is proposed to alleviate the issue of low pseudo-label quality caused by class imbalance and scale differences.
Adaptive Discrete Communication Bottlenecks with Dynamic Vector Quantization for Heterogeneous Representational Coarseness
Dianbo Liu (Mila Quebec AI Institute), Kenji Kawaguchi (National University of Singapore)
Representation LearningReinforcement LearningImage
🎯 What it does: A dynamic vector quantization (DVQ) method is proposed, which adaptively selects the size of the discrete codebook and the number of segments based on information complexity, achieving a tunable communication bottleneck.
Adaptive Dynamic Filtering Network for Image Denoising
Hao Shen (Hefei University of Technology), Wandi Zhang (Hefei University of Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: An image denoising network ADFNet based on adaptive dynamic filtering is proposed, aiming to efficiently preserve details while removing noise.
Adaptive Hierarchy-Branch Fusion for Online Knowledge Distillation
Linrui Gong (East China Normal University), Lizhuang Ma (East China Normal University)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: An Adaptive Hierarchical-Branch Fusion Framework (AHBF-OKD) is proposed, which constructs a deep incremental hierarchical branch and recursively uses a Teacher Assistant and attention mechanism for online knowledge distillation to address the homogenization problem of traditional OKD.
Adaptive Low-Precision Training for Embeddings in Click-Through Rate Prediction
Shiwei Li (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
Recommendation SystemOptimizationTabular
🎯 What it does: Proposes Low-Precision Training (LPT) to directly compress the embedding table of CTR models during the training phase, and further designs Adaptive Low-Precision Training (ALPT) to learn the step size, achieving lossless compression at 8-bit precision.
Adaptive Mixing of Auxiliary Losses in Supervised Learning
Durga Sivasubramanian (Indian Institute of Technology Bombay), Ganesh Ramakrishnan (Google Research)
Knowledge DistillationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: A framework called AMAL is proposed for instance-level adaptive mixing of auxiliary losses in supervised learning, which learns the mixing weights for each sample on the validation set using bi-level optimization and meta-learning, further applied to knowledge distillation and rule denoising tasks.
Adaptive Perturbation-Based Gradient Estimation for Discrete Latent Variable Models
Pasquale Minervini (University of Edinburgh), Mathias Niepert (University of Stuttgart)
OptimizationGraph
🎯 What it does: An adaptive gradient estimation method called AIMLE is proposed for efficiently solving gradients in deep learning models that include discrete algorithms.
Adaptive Policy Learning for Offline-to-Online Reinforcement Learning
Han Zheng (University of Technology Sydney), Jing Jiang (Southern University of Science and Technology)
Reinforcement LearningTabularBenchmark
🎯 What it does: An adaptive strategy learning framework (APL) is proposed, which simultaneously utilizes the diversity of offline data and the immediacy of online interaction in the offline-to-online reinforcement learning scenario, employing pessimistic updates for offline data and optimistic updates for online data.
Adaptive Texture Filtering for Single-Domain Generalized Segmentation
Xinhui Li (Tianjin University), Xiaojie Guo (Sun Yat-sen University)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This study focuses on semantic segmentation for single-source domain generalization, proposing an adaptive texture filtering mechanism and a hierarchical guidance generalization network to eliminate domain-specific texture influences and learn domain-invariant features.
AdaTask: A Task-Aware Adaptive Learning Rate Approach to Multi-Task Learning
Enneng Yang (Northeastern University), Guibing Guo (Tencent Inc)
Recommendation SystemOptimizationTabular
🎯 What it does: A task-aware adaptive learning rate method called AdaTask is proposed to address the parameter dominance issue in multi-task learning.
ADEPT: A DEbiasing PrompT Framework
Ke Yang (Tsinghua University), Heng Ji (University of Illinois Urbana-Champaign)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes ADEPT, an algorithm for bias removal using continuous prompting, which achieves debiasing through prompt tuning on pre-trained language models while maintaining representational capacity.
Adjective Scale Probe: Can Language Models Encode Formal Semantics Information?
Wei Liu (Zhejiang University), Nai Ding (University of Chicago)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper constructs the 'Adjective Scale Probe (ASP)' diagnostic dataset to test the understanding ability of language models regarding the semantics of adjective degrees.
ADMoE: Anomaly Detection with Mixture-of-Experts from Noisy Labels
Yue Zhao (Carnegie Mellon University), Ahmed Awadallah (Microsoft)
Anomaly DetectionMixture of ExpertsTabular
🎯 What it does: The ADMoE framework is proposed, utilizing a Mixture-of-Experts structure to enable end-to-end learning of anomaly detection models in a weakly supervised environment with multi-source noisy labels; it also provides a model-agnostic plugin approach that can be directly embedded into any neural network-based anomaly detection method.
Adversarial Alignment for Source Free Object Detection
Qiaosong Chu (Tsinghua University), Xiu Li (Tsinghua University)
Object DetectionDomain AdaptationKnowledge DistillationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a source-free object detection (SFOD) method A2 SFOD, which uses detection variance to adaptively divide target domain samples into two categories: source-similar and source-dissimilar. Then, through adversarial learning and the Mean Teacher structure, the feature spaces of the two categories are aligned, ultimately generating high-quality pseudo-labels for fine-tuning.
Adversarial Robust Deep Reinforcement Learning Requires Redefining Robustness
Ezgi Korkmaz
Adversarial AttackReinforcement LearningVideo
🎯 What it does: This study investigates the robustness of deep reinforcement learning policies against imperceptible perturbations and proposes a method to detect policy decision boundaries through high-sensitivity directions that are independent of the policy.
Adversarial Self-Attention for Language Understanding
Hongqiu Wu (Shanghai Jiao Tong University), Min Zhang (Soochow University)
OptimizationAdversarial AttackTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes an Adversarial Self-Attention (ASA) mechanism, which incorporates a learnable adversarial mask into the self-attention layer of the Transformer, thereby suppressing the model's over-reliance on keywords and enhancing the generalization and robustness of pre-training and fine-tuning.
Adversarial Weight Perturbation Improves Generalization in Graph Neural Networks
Yihan Wu (University of Pittsburgh), Heng Huang (CISPA Helmholtz Center for Information Security)
ClassificationAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper studies and proposes a new weight perturbation method WT-AWP to enhance the generalization and robustness of graph neural networks in non-i.i.d. node classification tasks.
Adversarial Word Dilution as Text Data Augmentation in Low-Resource Regime
Junfan Chen (Beihang University), Yongyi Mao (University of Ottawa)
ClassificationOptimizationAdversarial AttackTransformerSupervised Fine-TuningText
🎯 What it does: A text data augmentation method based on adversarial word dilution, AWD, is proposed, which generates hard positive samples by diluting positive word vectors;
AEC-GAN: Adversarial Error Correction GANs for Auto-Regressive Long Time-Series Generation
Lei Wang (Tsinghua University), Jian Li (Tsinghua University)
GenerationData SynthesisGenerative Adversarial NetworkTime Series
🎯 What it does: A conditional generative adversarial network named AEC-GAN is proposed, specifically designed to generate high-quality time series data of arbitrary lengths;
Aesthetically Relevant Image Captioning
Zhipeng Zhong (Shenzhen University), Guoping Qiu (University of Nottingham)
GenerationRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a framework that combines Image Aesthetic Quality Assessment (AQA) and Image Aesthetic Description (IAC), and introduces the Aesthetic Relevance Score (ARS) for the first time to measure the degree to which text describes the aesthetics of images.
AIO-P: Expanding Neural Performance Predictors beyond Image Classification
Keith G. Mills (University of Alberta), Shangling Jui (Huawei Technologies)
Object DetectionSegmentationPose EstimationGraph Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A global performance predictor AIO-P has been constructed, which can be pre-trained on multiple tasks and search spaces, and transferred to unseen tasks and architectures.
Alignment-Enriched Tuning for Patch-Level Pre-trained Document Image Models
Lei Wang (Singapore Management University), Hui Liu (Beijing Forestry University)
ClassificationRecognitionTransformerSupervised Fine-TuningContrastive LearningImageTextMultimodality
🎯 What it does: A new architecture for Alignment Enhanced Fine-tuning of pre-trained document image models (AETNet) is proposed, utilizing additional visual and textual Transformers and alignment loss to improve downstream task performance.
Almost Cost-Free Communication in Federated Best Arm Identification
Srinivas Reddy Kota (National University of Singapore), Vincent Y. F. Tan (National University of Singapore)
Recommendation SystemOptimizationFederated LearningTabular
🎯 What it does: This paper proposes an algorithm FEDELIM within the framework of federated learning multi-armed bandits, aimed at identifying local optimal arms and global optimal arms while keeping communication costs in check.
AlphaRoute: Large-Scale Coordinated Route Planning via Monte Carlo Tree Search
Guiyang Luo (Beijing University of Posts and Telecommunications), Jinglin Li (Beijing University of Posts and Telecommunications)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes AlphaRoute, which utilizes a regional-level game model, graph attention reinforcement learning, and Monte Carlo tree search to achieve large-scale road network collaborative path planning.
Alternating Layered Variational Quantum Circuits Can Be Classically Optimized Efficiently Using Classical Shadows
Afrad Basheer (University of Technology Sydney), Sanjiang Li (University of Technology Sydney)
OptimizationComputational EfficiencyAuto EncoderPhysics Related
🎯 What it does: The ALSO algorithm is proposed, which efficiently trains alternating layer variational quantum circuits on classical computers using classical shadow techniques.
Amodal Instance Segmentation via Prior-Guided Expansion
Junjie Chen (Shanghai Jiao Tong University), Liqing Zhang (SenseTime)
Object DetectionSegmentationConvolutional Neural NetworkOptical FlowImage
🎯 What it does: A prior-guided expansion framework based on a memory bank is proposed, utilizing regression and optical flow transformations to perform box-level and pixel-level expansions on the original visible boxes and masks, thereby achieving more complete invisible instance segmentation.
AMOM: Adaptive Masking over Masking for Conditional Masked Language Model
Yisheng Xiao (Soochow University), Min Zhang (Microsoft Research Asia)
GenerationTransformerLarge Language ModelText
🎯 What it does: This paper proposes an adaptive masking strategy AMOM to improve the Conditional Masked Language Model (CMLM) for enhancing the quality and inference speed of non-autoregressive text generation.
An Adaptive Layer to Leverage Both Domain and Task Specific Information from Scarce Data
Gaël Guibon (Telecom Paris), Chloé Clavel (Telecom Paris)
ClassificationDomain AdaptationTransformerSupervised Fine-TuningTextAudio
🎯 What it does: In the customer service dialogue environment lacking large-scale labeled data, the author proposes a two-step semi-supervised fine-tuning method—Task-Adaptive Fine-Tuning (TAFT). It first uses unlabeled speaker role information (customer/customer service) as a simple task for adaptation, then freezes the original model and only fine-tunes an additional lightweight adaptive layer to complete dual-task predictions of customer satisfaction and problem status.
An Efficient Algorithm for Fair Multi-Agent Multi-Armed Bandit with Low Regret
Matthew Jones (Northeastern University), Thy Nguyen (Northeastern University)
OptimizationReinforcement Learning from Human FeedbackTabular
🎯 What it does: This paper proposes an efficient algorithm that utilizes UCB and the logarithmic concavity of the Nash social welfare function to achieve low returns in the multi-agent multi-armed bandit problem, reaching an upper bound of approximately O~(√(NKT)+NK), and provides a non-efficient algorithm that achieves this upper bound.
An Efficient Deep Reinforcement Learning Algorithm for Solving Imperfect Information Extensive-Form Games
Linjian Meng (Nanjing University), Yang Gao (Nanjing University)
Reinforcement Learning
🎯 What it does: Proposes the FTRL-based ORD-DGF algorithm and its deep implementation Deep FTRL-ORW for efficiently learning Nash equilibria in large incomplete information extensive-form games.
An Ensemble Distillation Framework for Sentence Embeddings with Multilingual Round-Trip Translation
Tianyu Zong (University of Chinese Academy of Sciences), Likun Zhang (University of Chinese Academy of Sciences)
Knowledge DistillationRepresentation LearningData-Centric LearningTransformerContrastive LearningText
🎯 What it does: This paper proposes an unsupervised contrastive learning framework based on multilingual back-translation (RTT) data augmentation to generate more semantically expressive sentence embeddings.
An Equivalence Analysis of Binary Quantification Methods
Alberto Castaño (University of Oviedo), Juan José del Coz (University of Oviedo)
ClassificationOptimizationTabular
🎯 What it does: This paper proves through theoretical derivation and experimental validation that various binary quantization algorithms (AC, PAC, HDy, ORD, SORD, MM, etc.) are equivalent under the prior probability shift assumption, and based on this, proposes the QUANTy algorithm with a richer representation; it also provides a unified distribution matching framework.
An Extreme-Adaptive Time Series Prediction Model Based on Probability-Enhanced LSTM Neural Networks
Yanhong Li (Santa Clara University), David C. Anastasiu (Santa Clara Valley Water District)
Recurrent Neural NetworkGaussian SplattingMultimodalityTime Series
🎯 What it does: A highly adaptive time series forecasting model NEC+ is proposed, which learns the prediction functions for both extreme events and normal events simultaneously, and dynamically selects them using a classifier;
An Improved Algorithm for Online Min-Sum Set Cover
Marcin Bienkowski (University of Wroclaw), Marcin Mucha (University of Warsaw)
Optimization
🎯 What it does: The paper studies the Online Min-Sum Set Cover problem, proposing new randomized algorithms and corresponding deterministic algorithms, and provides a competitive ratio analysis compared to traditional static optimal solutions.
An Improved Approximation Algorithm for Wage Determination and Online Task Allocation in Crowd-Sourcing
Yuya Hikima (NTT Corporation), Taichi Asami (NTT Corporation)
OptimizationTabular
🎯 What it does: This paper proposes a joint optimization algorithm for wage determination and online task allocation on crowdsourcing platforms, aiming to maximize platform revenue.
An Operator Theoretic Approach for Analyzing Sequence Neural Networks
Ilan Naiman (Ben Gurion University), Omri Azencot (Ben Gurion University)
ClassificationExplainability and InterpretabilityRecurrent Neural NetworkTextTime SeriesBiomedical DataElectrocardiogramOrdinary Differential Equation
🎯 What it does: A KANN method based on Koopman theory is proposed for the global analysis of the hidden state dynamics of sequential neural networks and to reveal semantic information.
Analogical Inference Enhanced Knowledge Graph Embedding
Zhen Yao (Zhejiang University), Huajun Chen (Huawei Technologies Co., Ltd)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Enhancing the missing link prediction capability of knowledge graph embedding models through analogical reasoning
Analyzing and Improving the Use of the FastMap Embedding in Pathfinding Tasks
Reza Mashayekhi (University of Alberta), Nathan R. Sturtevant (University of Alberta)
OptimizationTabularBenchmark
🎯 What it does: This study investigates the effect of FastMap (L1) embedding in path search and proposes a method that combines Differential Heuristic (DH) with FastMap and improves the pivot selection strategy (HE) to enhance the performance of A* and MM* algorithms.
Anomaly Segmentation for High-Resolution Remote Sensing Images Based on Pixel Descriptors
Jingtao Li (Wuhan University), Yanfei Zhong (Wuhan University)
SegmentationAnomaly DetectionConvolutional Neural NetworkImageAgriculture Related
🎯 What it does: This paper proposes a high-resolution remote sensing image anomaly segmentation model (ASD) based on pixel descriptors, achieving precise localization of anomalous pixels by learning compact, diverse, and feature-rich pixel descriptors in the feature space.
Anytime User Engagement Prediction in Information Cascades for Arbitrary Observation Periods
Akshay Aravamudan (Florida Institute of Technology), Georgios C. Anagnostopoulos (Florida Institute of Technology)
Recommendation SystemGraph Neural NetworkGraphTime Series
🎯 What it does: A single model DANTE is proposed to predict whether users participate in information cascades within any observation period and prediction window.
Approval-Based Voting with Mixed Goods
Xinhang Lu (University of New South Wales), Warut Suksompong (National University of Singapore)
🎯 What it does: This study investigates a voting model for approval voting with mixed divisible and indivisible items, unifying multi-winner voting and cake division.
Approximating Full Conformal Prediction at Scale via Influence Functions
Javier Abad Martinez (ETH Zurich), Giovanni Cherubin (Microsoft Research)
ClassificationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkImageTabular
🎯 What it does: This paper proposes a method (ACP) that uses Influence Functions to approximate Full Conformal Prediction. It achieves a rapid estimation of the impact of adding or removing training points on the model through first-order approximation, avoiding the high computational cost of complete retraining.
Approximations for Indivisible Concave Allocations with Applications to Nash Welfare Maximization
Nathaniel Kell (Denison University), Kevin Sun (Elon University)
Optimization
🎯 What it does: This study investigates the allocation problem of non-separable projects under the concave value model, providing multiplication and addition approximation algorithms based on local curvature and extending them to smooth asymmetric Nash welfare maximization and piecewise linear utility.
Are Transformers Effective for Time Series Forecasting?
Ailing Zeng (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)
TransformerTime Series
🎯 What it does: This paper questions the effectiveness of Transformer in long-term time series forecasting (LTSF) and proposes a single-layer linear model LTSF-Linear as a direct multi-step (DMS) forecasting baseline. Experiments show that it often significantly outperforms existing Transformer solutions on nine public datasets.