arXivSub Start free trial

IJCAI 2023 Papers with Code β€” Page 2

International Joint Conference on Artificial Intelligence Β· 241 papers

GLPocket: A Multi-Scale Representation Learning Approach for Protein Binding Site Prediction

Peiying Li (Shanghai Jiao Tong University), Lei Xu (Shanghai Jiao Tong University)

CodeRepresentation LearningDrug DiscoveryConvolutional Neural NetworkTransformerBiomedical Data

🎯 What it does: Designed and implemented a multi-scale network called GLPocket based on Lmser for precise prediction of protein binding sites;

Graph Neural Convection-Diffusion with Heterophily

Kai Zhao (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)

CodeClassificationGraph Neural NetworkDiffusion modelGraphOrdinary Differential Equation

🎯 What it does: Proposed a Graph Neural Convection-Diffusion (GNC-D) model based on the convection-diffusion equation (CDE), which enhances traditional diffusion models by introducing a learnable convection term to better model information flow in heterophilic graphs.

Graph-based Semi-supervised Local Clustering with Few Labeled Nodes

Zhaiming Shen, Sheng Li (University of Virginia)

CodeGraph Neural NetworkGraph

🎯 What it does: Propose a semi-supervised local clustering method CS-LCE based on compressed sensing, which can extract target clusters using only a small number of seed nodes.

Guide to Control: Offline Hierarchical Reinforcement Learning Using Subgoal Generation for Long-Horizon and Sparse-Reward Tasks

Wonchul Shin (Sungkyunkwan University), Yusung Kim (Sungkyunkwan University)

CodeRobotic IntelligenceReinforcement LearningAuto EncoderTabularSequential

🎯 What it does: Proposed an offline hierarchical reinforcement learning framework called Guider, which generates reachable subgoals through a high-level policy and learns a low-level policy to achieve these subgoals, thereby solving long-horizon sparse reward tasks without online interaction.

Handling Learnwares Developed from Heterogeneous Feature Spaces without Auxiliary Data

Peng Tan (Nanjing University), Zhi-Hua Zhou (Nanjing University)

CodeDomain AdaptationRecommendation SystemRepresentation LearningTabular

🎯 What it does: Construct a learnware market that can accommodate and reuse models from different feature spaces, achieving feature space alignment between models via the RKME specification without relying on any auxiliary data.

Hawkes Process Based on Controlled Differential Equations

Minju Jo (Yonsei University), Noseong Park (Yonsei University)

CodeTime SeriesElectronic Health RecordsStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: This paper proposes a Hawkes process model HP-CDE based on neural controlled differential equations (CDE) to continuously model irregular time series events and precisely compute the log-likelihood.

HDFormer: High-order Directed Transformer for 3D Human Pose Estimation

Hanyuan Chen (DAMO Academy, Alibaba Group), Xuansong Xie (DAMO Academy, Alibaba Group)

CodePose EstimationConvolutional Neural NetworkTransformerGraphSequential

🎯 What it does: Propose a high-order directed Transformer (HDFormer), which achieves direct mapping from 2D keypoints to 3D pose by modeling high-order attention (bone↔joint, super-bone↔joint) in skeletal structures.

Helpful Information Sharing for Partially Informed Planning Agents

Sarah Keren (Technion Israel Institute of Technology), Sara Bernardini (Royal Holloway University of London)

CodeOptimizationReinforcement LearningAgentic AIBenchmark

🎯 What it does: This paper proposes the Helpful Information Sharing (HIS) problem, where in a two-agent collaboration scenario, the auxiliary agent must provide the minimal critical information to the partially informed execution agent to ensure the goal can be achieved; two solution methods are presented: Lazy-BFS based on breadth-first search, and Tka translation, which compiles HIS into a single-agent classical planning problem solvable directly by existing classical planners;

Hierarchical Semantic Contrast for Weakly Supervised Semantic Segmentation

Yuanchen Wu (Shanghai University), Shaorong Xie (Shanghai University)

CodeSegmentationConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: Propose Hierarchical Semantic Contrast (HSC), which utilizes ROI, class, and pixel-level three-tier contrast in weakly supervised semantic segmentation, combined with cross-supervision and momentum prototype learning to improve CAM quality and final segmentation accuracy.

Hierarchical State Abstraction based on Structural Information Principles

Xianghua Zeng (Beihang University), Philip S. Yu (Didi Chuxing)

CodeComputational EfficiencyRepresentation LearningReinforcement LearningAuto EncoderImage

🎯 What it does: Proposes the SISA framework, an unsupervised adaptive hierarchical state abstraction method based on structural information principles, which enhances RL decision-making efficiency in environments with rich observations.

ICDA: Illumination-Coupled Domain Adaptation Framework for Unsupervised Nighttime Semantic Segmentation

Chenghao Dong (Beijing University of Posts and Telecommunications), Anlong Ming (Beijing University of Posts and Telecommunications)

CodeSegmentationDomain AdaptationTransformerGenerative Adversarial NetworkContrastive LearningImage

🎯 What it does: This paper proposes a Lighting-Coupled Domain Adaptation framework (ICDA), which addresses the challenges of illumination differences and dataset discrepancies in nighttime semantic segmentation by constructing day-night image pairs and leveraging semantic relevance.

Improve Video Representation with Temporal Adversarial Augmentation

Jinhao Duan (Drexel University), Kaidi Xu (Drexel University)

CodeRepresentation LearningGenerative Adversarial NetworkVideo

🎯 What it does: Propose Temporal Adversarial Augmentation (TA) and the TAF framework, which utilize temporal loss generated by CAM during video model fine-tuning for adversarial augmentation. This balances the model's attention distribution across different temporal segments of the video, thereby enhancing video representation and generalization capabilities.

Improving Heterogeneous Model Reuse by Density Estimation

Anke Tang (Wuhan University), Dacheng Tao (University of Sydney)

CodeFederated LearningSafty and PrivacyFlow-based ModelContrastive LearningImageBenchmark

🎯 What it does: Studied model reuse in federated learning using local models and local data density estimation, constructing a global model and proposing a cross-party cross-entropy calibration method;

Independent Feature Decomposition and Instance Alignment for Unsupervised Domain Adaptation

Qichen He (University of Electronic Science and Technology of China), Dongde Hou (Southwest University of Political Science & Law)

CodeDomain AdaptationFlow-based ModelContrastive LearningImageBenchmark

🎯 What it does: Proposes an independent feature decomposition and instance alignment framework (IndUDA) based on invertible flow, which maps features to a latent space and separates domain-invariant and domain-specific dimensions, achieving domain-invariant feature adaptation through domain-internal exchange and noise replacement;

Inferring Private Valuations from Behavioral Data in Bilateral Sequential Bargaining

Lvye Cui (Beijing Institute of Technology), Haoran Yu (Beijing Institute of Technology)

CodeRecurrent Neural NetworkTabularSequentialFinance Related

🎯 What it does: This paper proposes a Bayesian learning-based private valuation inference framework called BLUE, which infers sellers' private valuations of goods from bilateral sequential bargaining behaviors under the assumption of non-strictly dominant strategies.

Intent-aware Recommendation via Disentangled Graph Contrastive Learning

Yuling Wang (Beijing University of Posts and Telecommunications), Wei Wu (Meituan)

CodeRecommendation SystemExplainability and InterpretabilityGraph Neural NetworkContrastive LearningTabular

🎯 What it does: Propose an intent-aware recommendation model IDCL based on graph neural networks, which can simultaneously learn interpretable user intentions and their behavior distributions.

iRe2f: Rethinking Effective Refinement in Language Structure Prediction via Efficient Iterative Retrospecting and Reasoning

Zuchao Li (Wuhan University), Hai Zhao (Shanghai Jiao Tong University)

CodeComputational EfficiencyRepresentation LearningGraph Neural NetworkTextGraph

🎯 What it does: This paper proposes a lightweight iterative revision framework, IREβ€―F, which revises language structure predictions by leveraging retrospection and reasoning on graph representations without re-encoding.

Kβˆ— Search over Orbit Space for Top-k Planning

Michael Katz (IBM T.J. Watson Research Center), Junkyu Lee (IBM T.J. Watson Research Center)

CodeOptimizationBenchmark

🎯 What it does: Propose a new algorithm called OK* that performs K* search in the orbit space, achieving efficient solutions for Top-k planning problems.

Keep Skills in Mind: Understanding and Implementing Skills in Commonsense Question Answering

Meikai Bao (University of Science and Technology of China), Jun Zhou (University of Science and Technology of China)

CodeExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: This paper proposes a dynamic skill-aware closed-source commonsense question answering framework, DSCQA, which enhances the model's reasoning ability by extracting skill features from the training set and dynamically injecting skill information during decoding.

Latent Processes Identification From Multi-View Time Series

Zenan Huang (Zhejiang University), Nenggan Zheng (Zhejiang University)

CodeExplainability and InterpretabilityRepresentation LearningContrastive LearningTime Series

🎯 What it does: Propose a framework named MuLTI for identifying identifiable causal latent variables from multi-perspective time series, and merge the latent variables extracted from each perspective into a complete latent variable through optimal transport.

Learning Attention from Attention: Efficient Self-Refinement Transformer for Face Super-Resolution

Guanxin Li (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)

CodeSuper ResolutionTransformerImage

🎯 What it does: Proposed a self-refinement Transformer framework for facial image super-resolution tasks;

Learning Calibrated Uncertainties for Domain Shift: A Distributionally Robust Learning Approach

Haoxuan Wang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)

CodeDomain AdaptationImage

🎯 What it does: Propose an end-to-end framework based on Distributionally Robust Learning (DRL), which models differences between source and target domains using a differentiable density ratio estimator. This enables learning calibrated confidence in domain drift scenarios, and the confidence is utilized for Unsupervised Domain Adaptation (UDA) and cross-domain Semi-Supervised Learning (SSL).

Learning Efficient Truthful Mechanisms for Trading Networks

Takayuki Osogami (IBM Research Technion Israel Institute Of Technology), Elisheva S. Shamash

CodeOptimizationTabularFinance Related

🎯 What it does: This paper proposes a mechanism learning method for transaction networks, leveraging the Groves mechanism under Bayesian settings to achieve DSIC, efficiency, expected weak budget balance, and expected individual rationality, addressing the infeasibility of simultaneously satisfying these four properties in such networks.

Learning Gaussian Mixture Representations for Tensor Time Series Forecasting

Jiewen Deng (Southern University of Science and Technology), Xuan Song (Southern University of Science and Technology)

CodeRepresentation LearningTime Series

🎯 What it does: Proposed a tensor time series prediction framework GMRL based on Gaussian Mixture Models.

Learning Object Consistency and Interaction in Image Generation from Scene Graphs

Yangkang Zhang (Zhejiang University), Lingyun Sun (Zhejiang University)

CodeGenerationGraph Neural NetworkTransformerAuto EncoderImageGraph

🎯 What it does: This paper proposes the LOCI method, aiming to solve the problem of objects being overlooked and insufficient interaction between objects in scene graph to image generation through a consistency module and an interaction module.

Learning Prototype Classifiers for Long-Tailed Recognition

Saurabh Sharma (University of California Santa Barbara), Ambuj Singh (University of California Santa Barbara)

CodeClassificationImage

🎯 What it does: This paper proposes a classifier based on learnable prototypes, which directly discriminates samples in the representation space using Euclidean distance, thereby eliminating the bias caused by the soft max classifier in long-tailed data due to the relationship between weight norm and sample count.

Learning Survival Distribution with Implicit Survival Function

Yu Ling (Fudan University), Bo Yan (Fudan University)

CodeBiomedical Data

🎯 What it does: Propose an Implicit Survival Function (ISF) that directly estimates the conditional hazard rate using implicit neural representations and time positional encoding, then obtains the survival distribution through numerical integration.

Learning to Learn from Corrupted Data for Few-Shot Learning

Yuexuan An (Southeast University), Hui Xue (Southeast University)

CodeClassificationData-Centric LearningMeta LearningConvolutional Neural NetworkRecurrent Neural NetworkTransformerImage

🎯 What it does: Designed a peer collaborative learning framework that uses dual modules to jointly assess sample importance, enhancing the robustness of few-shot learning models against polluted data.

Learning to Speak from Text: Zero-Shot Multilingual Text-to-Speech with Unsupervised Text Pretraining

Takaaki Saeki (University of Tokyo), Hiroshi Saruwatari (University of Tokyo)

CodeGenerationData SynthesisTransformerGenerative Adversarial NetworkContrastive LearningTextAudio

🎯 What it does: A zero-shot multilingual text-to-speech (TTS) system is studied, which achieves cross-lingual phoneme and prosody transfer through unsupervised multilingual pre-training using only text, enabling speech synthesis for unseen languages.

Learning When to Use Automatic Tabulation in Constraint Model Reformulation

Carlo Cena (University of Bologna), Felix Ulrich-Oltean (University of York)

CodeClassificationOptimizationTabular

🎯 What it does: Explored how to use machine learning to predict when to apply automatic tabulation techniques in constraint models to improve the runtime of SAT/CP solvers.

Less Learn Shortcut: Analyzing and Mitigating Learning of Spurious Feature-Label Correlation

Yanrui Du (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)

CodeClassificationExplainability and InterpretabilityData-Centric LearningSupervised Fine-TuningText

🎯 What it does: This paper investigates the 'word-label shortcut' phenomenon in natural language tasks caused by dataset bias in deep learning models, and proposes a task-agnostic training strategy called LessLearn-Shortcut (LLS) to reduce the model's over-reliance on biased words;

Levin Tree Search with Context Models

Laurent Orseau (Google DeepMind), Levi H. S. Lelis (University of Alberta)

CodeOptimizationReinforcement LearningMixture of ExpertsBenchmark

🎯 What it does: Propose using a parameterized context model (based on online compression of context trees) instead of neural networks as the strategy for Levin Tree Search (LTS), and prove that the LTS loss function is convex under this model, enabling training of the strategy via convex optimization algorithms. Subsequent experiments were conducted on Sokoban, The Witness, 24-Sliding Tile Puzzle (STP), and Rubik’s Cube.

LGI-GT: Graph Transformers with Local and Global Operators Interleaving

Shuo Yin (Ocean University of China), Guoqiang Zhong (Ocean University of China)

CodeRepresentation LearningDrug DiscoveryGraph Neural NetworkTransformerGraph

🎯 What it does: Designed an interleaved local-global operation Graph Transformer (LGI-GT) and proposed an attention-based local information enhancement module (EELA).

Linguistic More: Taking a Further Step toward Efficient and Accurate Scene Text Recognition

Boqiang Zhang (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)

CodeRecognitionComputational EfficiencyTransformerImage

🎯 What it does: This paper proposes the Linguistic Perception Vision model (LPV), which enhances scene text recognition accuracy by integrating linguistic information into visual models through Cascade Position Attention (CPA) and Global Linguistic Reconstruction Module (GLRM).

LION: Label Disambiguation for Semi-supervised Facial Expression Recognition with Progressive Negative Learning

Zhongjing Du (Sichuan University), Yan Wang (Sichuan University)

CodeRecognitionImage

🎯 What it does: Propose a semi-supervised deep facial expression recognition framework called LION, combining a label disambiguation module and a progressive negative learning module.

Manifold-Aware Self-Training for Unsupervised Domain Adaptation on Regressing 6D Object Pose

Yichen Zhang (South China University of Technology), Kui Jia (South China University of Technology)

CodePose EstimationDomain AdaptationImage

🎯 What it does: Propose a 6D pose estimation method for unsupervised domain adaptation called MAST, which employs self-supervised manifold regularization and self-training strategies to decompose the regression task into coarse classification and fine-grained regression.

MAT: Mixed-Strategy Game of Adversarial Training in Fine-tuning

Zhehua Zhong (Hangzhou Dianzi University), Zhen Wang (Hangzhou Dianzi University)

CodeRepresentation LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: Propose a mixed strategy adversarial training algorithm (MAT), introducing mixed strategy game theory during fine-tuning of pre-trained language models to improve model generalization and robustness.

Minimally Supervised Contextual Inference from Human Mobility: An Iterative Collaborative Distillation Framework

Jiayun Zhang (University of California, San Diego), Jingbo Shang (University of California, San Diego)

CodeClassificationKnowledge DistillationConvolutional Neural NetworkTime Series

🎯 What it does: This paper proposes a minimally supervised mobile context reasoning framework called STCOLAB, which learns contextual information from human mobility data with extremely few labels by utilizing alternating training and collaborative distillation of spatial and temporal modules.

MM-PCQA: Multi-Modal Learning for No-reference Point Cloud Quality Assessment

Zicheng Zhang (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)

CodeConvolutional Neural NetworkGraph Neural NetworkImageMultimodalityPoint Cloud

🎯 What it does: This paper proposes a no-reference multimodal point cloud quality assessment method called MM-PCQA, which combines features from point cloud local sub-models and 2D projections;

Modeling Moral Choices in Social Dilemmas with Multi-Agent Reinforcement Learning

Elizaveta Tennant (University College London), Mirco Musolesi (University College London)

CodeOptimizationReinforcement LearningTabular

🎯 What it does: In three classic social dilemma games (Iterated Prisoner's Dilemma, Volunteer's Dilemma, Stag Hunt), Q-learning is used to learn moral reward functions based on different ethical theories (utilitarianism, deontology, virtue equality, virtue benevolence, hybrid virtue). The behaviors and social outcomes of these moral agents, as well as selfish agents and traditional game theory agents, are systematically evaluated in two-player interactions.

MolHF: A Hierarchical Normalizing Flow for Molecular Graph Generation

Yiheng Zhu (Zhejiang University), Jian Wu (Zhejiang University)

CodeDrug DiscoveryGraph Neural NetworkFlow-based ModelGraphBiomedical Data

🎯 What it does: Propose a hierarchical normalization flow model called MolHF that generates molecular graphs using a coarse-to-fine approach.

Multi-Agent Intention Recognition and Progression

Michael Dann (RMIT University), John Thangarajah (RMIT University)

CodeOptimizationReinforcement Learning

🎯 What it does: This paper proposes a multi-agent intent recognition and scheduling framework named Iβ€―GR, which utilizes online goal recognition (based on KL divergence in reinforcement learning) to infer the current goals of other agents in the same scenario, and embeds the inferred results into an MCTS intent scheduler to achieve prediction of other agents' behaviors and optimization of self-actions.

Multi-objective Optimization-based Selection for Quality-Diversity by Non-surrounded-dominated Sorting

Ren-Jian Wang (Nanjing University), Qiang Fu (Tencent)

CodeOptimizationReinforcement LearningBenchmark

🎯 What it does: Proposed a new parent selection method called Non-Encircled Dominance Sorting (NSS) and integrated it into QD algorithms such as MAP-Elites; experimental results verified its superiority on multiple benchmark tasks.

Multi-view Contrastive Learning Hypergraph Neural Network for Drug-Microbe-Disease Association Prediction

Luotao Liu (Huazhong Agricultural University), Wen Zhang (Huazhong Agricultural University)

CodeDrug DiscoveryGraph Neural NetworkContrastive LearningBiomedical Data

🎯 What it does: This paper constructs a hypergraph of drug-microbe-disease tripartite associations and employs a multi-view contrastive learning hypergraph neural network (MCHNN) for association prediction.

MultiPar-T: Multiparty-Transformer for Capturing Contingent Behaviors in Group Conversations

Dong Won Lee (Massachusetts Institute of Technology), Hae Won Park (Massachusetts Institute of Technology)

CodeClassificationRecurrent Neural NetworkTransformerMultimodality

🎯 What it does: Studied a multi-participant Transformer called MultiPar-T for capturing follower behaviors in group conversations, and achieved interactive modeling for multiple participants.

Musical Voice Separation as Link Prediction: Modeling a Musical Perception Task as a Multi-Trajectory Tracking Problem

Emmanouil Karystinaios (Johannes Kepler University Linz), Gerhard Widmer (Johannes Kepler University Linz)

CodeConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkText

🎯 What it does: This paper models the sound separation problem as a multi-target tracking (MTT) problem, using graph neural networks to embed each note and determining adjacent notes within the same voice through link prediction to recover complete voice trajectories.

Negative Flux Aggregation to Estimate Feature Attributions

Xin Li (Wayne State University), Dongxiao Zhu (Wayne State University)

CodeExplainability and InterpretabilityConvolutional Neural NetworkImage

🎯 What it does: Proposes the Negative Flux Aggregation (NeFLAG) method based on vector divergence and flux for DNN explanation without benchmarks or path integrals;

NerCo: A Contrastive Learning Based Two-Stage Chinese NER Method

Zai Zhang (Xi'an Jiaotong University), Qinghua Zheng (Xi'an Jiaotong University)

CodeRecognitionRepresentation LearningTransformerSupervised Fine-TuningContrastive LearningText

🎯 What it does: This paper proposes a two-stage Chinese Named Entity Recognition framework called NerCo, first aggregating representations of words with the same entity type through contrastive learning, and then performing fine-tuning based on traditional sequence labeling.

NeuPSL: Neural Probabilistic Soft Logic

Connor Pryor (University of California Santa Barbara), Lise Getoor (University of California Santa Barbara)

CodeClassificationOptimizationComputational EfficiencyImageGraph

🎯 What it does: Propose the NeuPSL neuro-symbolic framework, integrating outputs of deep neural networks with probabilistic soft logic (PSL) to achieve differentiable joint reasoning and learning.

Neural Capacitated Clustering

Jonas K. Falkner (University of Hildesheim), Lars Schmidt-Thieme (University of Hildesheim)

CodeOptimizationGraph Neural NetworkTransformerSupervised Fine-TuningBenchmark

🎯 What it does: Proposed a deep learning-based neural capacitated clustering method (Neural Capacitated Clustering, NCC), which uses graph neural networks to learn point-to-cluster assignment probabilities and dynamically satisfies capacity constraints during the iterative k-means process.

Neuro-Symbolic Class Expression Learning

Caglar Demir (Paderborn University), Axel-Cyrille Ngonga Ngomo (Paderborn University)

CodeRepresentation LearningConvolutional Neural NetworkReinforcement LearningGraphBiomedical DataBenchmark

🎯 What it does: Propose a neuro-symbolic class expression learning method called DRILL, modeling description logic class expression learning as a reinforcement learning problem, using deep Q-learning to drive the search for faster convergence to the target expression.

Neuro-Symbolic Learning of Answer Set Programs from Raw Data

Daniel Cunnington (IBM Research Europe), Alessandra Russo (Imperial College London)

CodeExplainability and InterpretabilityData-Centric LearningConvolutional Neural NetworkRecurrent Neural NetworkImage

🎯 What it does: Propose an end-to-end neuro-symbolic learning framework NSIL that can simultaneously train a general neural network and an Answer Set Programming (ASP) knowledge base using only raw data labels, learning the perceptual and inferential mapping from raw inputs to target labels.

ODEE: A One-Stage Object Detection Framework for Overlapping and Nested Event Extraction

Jinzhong Ning (Dalian University of Technology), Hongfei Lin (Dalian University of Technology)

CodeObject DetectionTransformerLarge Language ModelTextBiomedical DataFinance Related

🎯 What it does: Propose a one-stage object detection-based framework called ODEE, which utilizes vertex-based marking and trigger/argument span and type prediction to directly extract overlapping and nested events.

On Discovering Interesting Combinatorial Integer Sequences

Martin SvatoΕ‘ (Czech Technical University in Prague), OndΕ™ej KuΕΎelka (Czech Technical University in Prague)

CodeGenerationData SynthesisSequential

🎯 What it does: Automatically generate integer sequences with combinatorial explanations and build a database containing over 26,000 unique sequences.

On Optimal Strategies for Wordle and General Guessing Games

Michael Cunanan (University of New South Wales), Michael Thielscher (University of New South Wales)

CodeOptimizationText

🎯 What it does: This paper proposes a general framework for solving optimal strategies in guessing games without performing full search, and implements and verifies it using Wordle as an example.

On the Paradox of Learning to Reason from Data

Honghua Zhang (University of California, Los Angeles), Guy Van den Broeck (University of California, Los Angeles)

CodeData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: This paper trains BERT (and T5) models in a restricted logical reasoning problem space (SimpleLogic) and evaluates their reasoning ability across different data distributions; it finds that models achieve nearly 100% accuracy on training distributions but perform extremely poorly on other distributions, indicating that models have not truly learned reasoning but instead exploit statistical features in the data.

On the Reuse Bias in Off-Policy Reinforcement Learning

Chengyang Ying (Tsinghua University), Jun Zhu (Tsinghua University)

CodeReinforcement Learning

🎯 What it does: This paper studies the estimation bias caused by reusing data in offline replay buffers and introduces the concept of Reuse Bias.

One Model for All Domains: Collaborative Domain-Prefix Tuning for Cross-Domain NER

Xiang Chen (Zhejiang University), Huajun Chen (Zhejiang University)

CodeRecognitionDomain AdaptationTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: Propose a cross-domain named entity recognition (NER) method, CP-NER, which redefines NER as a text-to-text generation task and adapts to different domains by freezing a pre-trained language model (T5) combined with collaborative domain prefix fine-tuning.

One Model, Any CSP: Graph Neural Networks as Fast Global Search Heuristics for Constraint Satisfaction

Jan TΓΆnshoff (RWTH Aachen University), Martin Grohe (RWTH Aachen University)

CodeOptimizationGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: Designed a general graph neural network architecture ANYCSP, which can be unsupervised trained to become a search heuristic for any constraint satisfaction problem (CSP).

Online Harmonizing Gradient Descent for Imbalanced Data Streams One-Pass Classification

Han Zhou (Chongqing University), Yuyu Huang (Chongqing University)

CodeClassificationOptimizationTabularSequential

🎯 What it does: Propose an online harmonized gradient descent (OHGD) algorithm for one-pass streaming data class imbalance classification;

Open Anomalous Trajectory Recognition via Probabilistic Metric Learning

Qiang Gao (Southwestern University of Finance and Economics), Fan Zhou (University of Electronic Science and Technology of China)

CodeAnomaly DetectionKnowledge DistillationAuto EncoderContrastive LearningTime SeriesSequential

🎯 What it does: Propose the open abnormal trajectory identification task and design the ATROM model

Open-world Semi-supervised Novel Class Discovery

Jiaming Liu (University of Electronic Science and Technology of China), Junming Shao (University of Electronic Science and Technology of China)

CodeClassificationContrastive LearningImage

🎯 What it does: Propose a semi-supervised open-world new category discovery method called OpenNCD, which can simultaneously identify known categories and discover unknown categories in unlabeled data.

OptIForest: Optimal Isolation Forest for Anomaly Detection

Haolong Xiang (Macquarie University), Xiaolong Xu (Nanjing University of Information Science and Technology)

CodeAnomaly DetectionTabularBenchmark

🎯 What it does: Proposed OptIForest, combining theoretical analysis and practical implementation, which achieves efficient and robust anomaly detection through the optimal isolation tree design based on the best branching factor e

Optimal Decision Tree Policies for Markov Decision Processes

DaniΓ«l Vos, Sicco Verwer (Delft University of Technology)

CodeReinforcement LearningTabularBenchmark

🎯 What it does: Proposes OMDT, a framework based on mixed-integer linear programming (MILP), which directly solves the optimal decision tree strategy for Markov decision processes (MDPs) under given tree size constraints.

Orientation-Independent Chinese Text Recognition in Scene Images

Haiyang Yu (Fudan University), Xiangyang Xue (Fudan University)

CodeRecognitionConvolutional Neural NetworkTransformerAuto EncoderImageBenchmark

🎯 What it does: A method for direction-agnostic recognition of Chinese text in scenes is studied by separating content and directional information.

Overlooked Implications of the Reconstruction Loss for VAE Disentanglement

Nathan Michlo (University of Witwatersrand), Steven James (University of Witwatersrand)

CodeGenerationRepresentation LearningConvolutional Neural NetworkAuto EncoderImage

🎯 What it does: This paper analyzes the impact of reconstruction loss on perceptual data distance when VAEs learn separable representations. It constructs an adversarial dataset to demonstrate the non-separability caused by traditional pixel-level loss, and subsequently proposes a spatially aware reconstruction loss to restore separability.

Part Aware Contrastive Learning for Self-Supervised Action Recognition

Yilei Hua (Wuhan University of Science and Technology), Shiqian Wu (University of North Carolina at Charlotte)

CodeRecognitionRecurrent Neural NetworkGraph Neural NetworkTransformerContrastive LearningGraph

🎯 What it does: Designed and implemented a self-supervised contrastive learning framework called SkeAttnCLR for 3D skeleton action recognition, enhancing feature representation by integrating global and local contrastive learning.

PasCore: A Chinese Overlapping Relation Extraction Model Based on Global Pointer Annotation Strategy

Peng Wang (Southeast University), Wei Li (Beijing Institute of Computer Technology and Application)

CodeTransformerLarge Language ModelText

🎯 What it does: Proposes PasCore, a Chinese overlapping relation extraction model based on a global pointer annotation strategy, which sequentially performs three stages: relation prediction, head entity annotation, and tail entity annotation.

PathLAD+: An Improved Exact Algorithm for Subgraph Isomorphism Problem

Yiyuan Wang (Northeast Normal University), Qingwei Lin (Microsoft Research)

CodeOptimizationComputational EfficiencyImageMesh

🎯 What it does: Propose an improved exact algorithm called PathLAD+, which addresses the subgraph isomorphism problem through three novel heuristics: probe search, matching sorting based on probe information, and adaptive propagation.

PED-ANOVA: Efficiently Quantifying Hyperparameter Importance in Arbitrary Subspaces

Shuhei Watanabe (University of Freiburg), Frank Hutter (University of Freiburg)

CodeHyperparameter SearchImageBenchmark

🎯 What it does: Propose a fast local f-ANOVA method based on Pearson divergence (PED-ANOVA), which can efficiently compute hyperparameter importance in any subspace (local space);

PowerBEV: A Powerful Yet Lightweight Framework for Instance Prediction in Bird’s-Eye View

Peizheng Li (Mercedes-Benz AG), Juergen Gall (University of Bonn)

CodeObject DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkFlow-based ModelOptical FlowImage

🎯 What it does: Proposed an end-to-end BEV prediction framework called PowerBEV, achieving instance prediction through only two outputs: semantic segmentation and backward clustering flow.

PPAT: Progressive Graph Pairwise Attention Network for Event Causality Identification

Zhenyu Liu (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)

CodeClassificationGraph Neural NetworkTransformerText

🎯 What it does: Proposed the Progressive Graph Pairwise Attention Network (PPAT) for event causality recognition.

Prompt Federated Learning for Weather Forecasting: Toward Foundation Models on Meteorological Data

Shengchao Chen (University of Technology Sydney), Jing Jiang (University of Technology Sydney)

CodeFederated LearningGraph Neural NetworkTransformerPrompt EngineeringTime Series

🎯 What it does: Proposed MetePFL, a framework for weather prediction on multi-region heterogeneous meteorological time series data, combining a pre-trained Transformer base model with federated Prompt learning;

Pyramid Diffusion Models for Low-light Image Enhancement

Dewei Zhou (Zhejiang University), Yi Yang (Zhejiang University)

CodeRestorationDiffusion modelImage

🎯 What it does: Propose a low-light image enhancement method called PyDiff based on diffusion models, achieving faster and higher quality restoration through pyramid diffusion and a global corrector.

Quantifying Consistency and Information Loss for Causal Abstraction Learning

Fabio Massimo Zennaro (University of Warwick), Theodoros Damoulas (University of Warwick)

CodeExplainability and InterpretabilityRepresentation LearningGraphBiomedical Data

🎯 What it does: This paper proposes an intervention-based abstract approximation metric, combining consistency (IC) with information loss (IIL, ISIL, ISC) to quantify differences between causal models at different hierarchical structural levels, and provides corresponding evaluation and learning algorithms;

RaMLP: Vision MLP via Region-aware Mixing

Shenqi Lai (InsightFace.ai), Kaipeng Zhang (Shanghai AI Laboratory)

CodeClassificationObject DetectionSegmentationImage

🎯 What it does: Designed a Region-aware MLP (RaMLP) visual backbone, achieving efficient visual representations with variable sizes through the Region-aware Mixing (RaM) module, directly applicable to image classification and dense prediction tasks.

RaSa: Relation and Sensitivity Aware Representation Learning for Text-based Person Search

Yang Bai (Soochow University), Min Zhang (Soochow University)

CodeRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality

🎯 What it does: Propose a representation learning method RaSa for person re-identification in text retrieval, which includes two tasks: relation-aware learning (RA) and sensitivity-aware learning (SA).

Recognizable Information Bottleneck

Yilin Lyu (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)

CodeRepresentation LearningImage

🎯 What it does: The paper proposes the Recognizable Information Bottleneck (RIB), which enhances model generalization performance by constraining the recognizability of representations.

Relation-enhanced DETR for Component Detection in Graphic Design Reverse Engineering

Xixuan Hao (University of Hong Kong), Chin-Yew Lin (Microsoft Research)

CodeObject DetectionTransformerImage

🎯 What it does: This paper addresses reverse engineering in graphic design, studying the component detection task;

Revisiting the Evaluation of Deep Learning-Based Compiler Testing

Yongqiang Tian (University of Waterloo), Shing-Chi Cheung (Hong Kong University of Science and Technology)

CodeTextBenchmark

🎯 What it does: This paper proposes Kitten, a language-agnostic, mutation-based program generator designed to fairly evaluate deep learning-driven compiler testing tools.

RFENet: Towards Reciprocal Feature Evolution for Glass Segmentation

Ke Fan (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)

CodeSegmentationConvolutional Neural NetworkImage

🎯 What it does: Propose a recursive feature evolution network called RFENet, specifically designed for the segmentation of glass-like objects.

Robust Image Ordinal Regression with Controllable Image Generation

Yi Cheng (Zhejiang University), Jian Wu (Zhejiang University)

CodeClassificationGenerationData SynthesisConvolutional Neural NetworkTransformerAuto EncoderImage

🎯 What it does: In image ordinal regression tasks, the controllable image generation framework CIG is proposed, utilizing generated synthetic samples to address class imbalance and class overlap issues;

Robust Reinforcement Learning via Progressive Task Sequence

Yike Li (Beijing Jiaotong University), Jiqiang Liu (Beijing Jiaotong University)

CodeReinforcement LearningImageSequential

🎯 What it does: Proposed a robust reinforcement learning framework in the form of max-expectation, and designed the Dynamic Robust RL (DRRL) framework, which generates and sorts evolutionary task sequences through genetic algorithms to achieve dynamic multi-task learning, thereby enhancing robustness and training stability.

RuleMatch: Matching Abstract Rules for Semi-supervised Learning of Human Standard Intelligence Tests

Yunlong Xu (Shanghai Jiao Tong University), Ru-Yuan Zhang (Shanghai Jiao Tong University)

CodeClassificationExplainability and InterpretabilityData-Centric LearningImageBenchmark

🎯 What it does: Propose the RuleMatch semi-supervised learning framework, which utilizes rule-level data augmentation and rule consistency loss to train deep models on the RPM problem.

SAD: Semi-Supervised Anomaly Detection on Dynamic Graphs

Sheng Tian (Ant Group), Liang Chen (Sun Yat-sen University)

CodeAnomaly DetectionGraph Neural NetworkContrastive LearningGraphFinance Related

🎯 What it does: Propose a semi-supervised anomaly detection framework named SAD for detecting node anomalies in dynamic graphs.

Safe Reinforcement Learning via Probabilistic Logic Shields

Wen-Chi Yang (KU Leuven), Luc De Raedt (KU Leuven)

CodeConvolutional Neural NetworkReinforcement Learning

🎯 What it does: Proposed a differentiable safe reinforcement learning method called Probabilistic Logic Policy Gradient (PLPG), achieving safety constraints in continuous action spaces by combining probabilistic logic programming with policy gradients.

Safety Verification and Universal Invariants for Relational Action Bases

Silvio Ghilardi (Universita degli Studi di Milano), Andrey Rivkin (Technical University of Denmark)

CodeSafty and Privacy

🎯 What it does: Proposed a new Relational Action Basis (RAB) framework for modeling and verifying relational dynamic systems with arithmetic and universal quantification constraints.

Sample Efficient Model-free Reinforcement Learning from LTL Specifications with Optimality Guarantees

Daqian Shao (University of Oxford), Marta Kwiatkowska (University of Oxford)

CodeOptimizationReinforcement Learning

🎯 What it does: Propose a model-free reinforcement learning framework that learns optimal policies on unknown Markov Decision Processes (MDPs) using Linear Temporal Logic (LTL) specifications, with theoretical optimality guarantees;

SAT-Based PAC Learning of Description Logic Concepts

Balder ten Cate (University of Amsterdam), Carsten Lutz (Leipzig University)

CodeBenchmark

🎯 What it does: This paper proposes a description logic concept learning framework based on bounded fitting, implements a SAT-driven system called SPELL, and provides theoretical guarantees for PAC learning; meanwhile, it proves that traditional most specific/general fitting and refinement methods lack sample efficiency.

Scalable Optimal Margin Distribution Machine

Yilin Wang (Huazhong University of Science and Technology), Hai Jin (Huazhong University of Science and Technology)

CodeOptimizationComputational EfficiencyImageText

🎯 What it does: This paper proposes a scalable optimal margin distribution machine (SODM), which significantly improves training speed while maintaining generalization performance through a novel partitioning strategy and accelerated SVRG under a linear kernel.

Scalable Verification of Strategy Logic through Three-Valued Abstraction

Francesco Belardinelli (Imperial College London), Aniello Murano (University of Naples Federico II)

CodeGraph

🎯 What it does: Propose a three-valued semantics for strategy logic, design a three-valued abstraction method based on this semantics, and implement and verify it within the MCMAS framework.

ScriptWorld: Text Based Environment for Learning Procedural Knowledge

Abhinav Joshi (Indian Institute of Technology Kanpur), Ashutosh Modi (Indian Institute of Technology Kanpur)

CodeTransformerLarge Language ModelReinforcement LearningText

🎯 What it does: Developed the ScriptWorld text-based game environment and trained RL agents based on script knowledge within it.

Self-Supervised Neuron Segmentation with Multi-Agent Reinforcement Learning

Yinda Chen (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)

CodeSegmentationTransformerReinforcement LearningAuto EncoderBiomedical Data

🎯 What it does: Designed and implemented a self-supervised masked image model based on multi-agent reinforcement learning (Decision-based MIM) to automatically identify optimal mask ratios and strategies. During pre-training, a Vision Transformer was trained with HOG and MSE reconstruction objectives, and a UNETR decoder was added for neuron instance segmentation.

Semantic-Aware Generation of Multi-View Portrait Drawings

Biao Ma (Hangzhou Dianzi University), Gang Xu (Hangzhou Dianzi University)

CodeGenerationNeural Radiance FieldGenerative Adversarial NetworkImage

🎯 What it does: Designed a semantic-aware generative model SAGE for generating multi-view human face sketches or paintings;

Semi-supervised Domain Adaptation via Prototype-based Multi-level Learning

Xinyang Huang (Beijing University of Posts and Telecommunications), Wenkai Chen (Beijing University of Posts and Telecommunications)

CodeClassificationDomain AdaptationConvolutional Neural NetworkImageBenchmark

🎯 What it does: Propose the Prototype-based Multi-level Learning (ProML) framework to fully utilize the limited target label samples in semi-supervised domain adaptation, enhancing model transfer performance through prototype aggregation, cross-domain alignment, and batch-level dual consistency.

Sequential Attention Source Identification Based on Feature Representation

Dongpeng Hou (Northwestern Polytechnical University), Xuelong Li (Northwestern Polytechnical University)

CodeAnomaly DetectionRepresentation LearningRecurrent Neural NetworkGraph Neural NetworkGraphTime Series

🎯 What it does: Study the problem of multi-source rumor source identification, proposing a sequence-to-sequence framework called TGASI based on time series attention and graph attention, which uses infection snapshot sequences to locate the propagation source.

Sequential Recommendation with Probabilistic Logical Reasoning

Huanhuan Yuan (Soochow University), Lei Zhao (Soochow University)

CodeRecommendation SystemRecurrent Neural NetworkTransformerSequential

🎯 What it does: Integrate deep learning with symbolic learning, proposing the SR-PLR framework, which employs Beta distribution for probabilistic logical reasoning in sequence recommendation, and generates recommendations by concatenating logical embeddings with feature embeddings.

SeRO: Self-Supervised Reinforcement Learning for Recovery from Out-of-Distribution Situations

Chan Kim (Seoul National University), Seong-Woo Kim (Seoul National University)

CodeRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a self-supervised reinforcement learning method called SeRO to restore reliable robot behaviors in discrete distribution (OOD) states.

Singularformer: Learning to Decompose Self-Attention to Linearize the Complexity of Transformer

Yifan Wu (Central South University), Min Li (Central South University)

CodeComputational EfficiencyRepresentation LearningTransformerImageTextBiomedical Data

🎯 What it does: Propose Singularformer, which utilizes a neural network to learn the singular value decomposition (SVD) of the attention matrix, achieving linear complexity and low memory consumption in self-attention;

SLViT: Scale-Wise Language-Guided Vision Transformer for Referring Image Segmentation

Shuyi Ouyang (Zhejiang University), Lanfen Lin (Zhejiang University)

CodeSegmentationTransformerLarge Language ModelVision Language ModelMultimodality

🎯 What it does: This paper proposes a Transformer-based referential image segmentation framework named SLViT, integrating visual-language encoders and cross-scale enhancement modules to precisely segment targets in images based on language expressions.