IJCAI 2025 Papers — Page 11
International Joint Conference on Artificial Intelligence · 1014 papers
What Can We Learn From MIMO Graph Convolutions?
Andreas Roth (TU Dortmund University), Thomas Liebig (TU Dortmund University)
Graph Neural NetworkGraph
🎯 What it does: This paper derives the essence of multi-channel (MIMO) graph convolution (MIMO-GC) and proposes a localized MIMO graph convolution (LMGC) framework based on this derivation; subsequently, it theoretically proves that LMGC is injective under a single computational graph and generates linearly independent representations when using multiple computational graphs, and experimentally verifies its performance on graph-level and node-level tasks.
What Makes You Special? Contrastive Heuristics Based on Qualified Dominance
Rasmus G. Tollund (Aalborg University), Alvaro Torralba (Aalborg University)
OptimizationBenchmark
🎯 What it does: Propose contrastive heuristics and qualified dominance methods for cost-optimal planning;
Where and How to Enhance: Discovering Bit-Width Contribution for Mixed Precision Quantization
Haidong Kang, Shangce Gao (University of Toyama)
Computational EfficiencyNeural Architecture SearchConvolutional Neural NetworkImage
🎯 What it does: Research and improve the mid-bitwidth selection process in mixed-precision quantization (MPQ), proposing a bitwidth contribution evaluation method based on Shapley values (SMPQ), and using Monte-Carlo sampling approximation to address the α value misjudgment problem in traditional gradient-driven methods.
Where and When: Predict Next POI and Its Explicit Timestamp in Sequential Recommendation
Yuanbo Xu (Jilin University), En Wang (Jilin University)
Recommendation SystemRecurrent Neural NetworkTransformerTime SeriesSequential
🎯 What it does: Propose a multi-task learning framework called TAPT, which can simultaneously predict the next POI a user will visit and the corresponding timestamp.
Where Does This Data Come From? Enhanced Source Inference Attacks in Federated Learning
Haiyang Chen (Nanjing University of Information Science and Technology), Hongsheng Hu (University of Newcastle)
Federated LearningAdversarial AttackImage
🎯 What it does: Propose and implement proactively enhanced source information inference attacks (ESIAs) in federated learning, amplifying client differences through gradient ascent and data augmentation to accurately identify sample origins.
Why the Agent Made that Decision: Contrastive Explanation Learning for Reinforcement Learning
Rui Zuo (Syracuse University), Qinru Qiu (Syracuse University)
Explainability and InterpretabilityConvolutional Neural NetworkReinforcement LearningContrastive LearningImage
🎯 What it does: This paper proposes VisionMask, an interpretable method for reinforcement learning (RL) agents, which generates sparse saliency masks for each action through self-supervised contrastive learning, explaining why an agent chooses a particular action over others.
Wisdom from Diversity: Bias Mitigation Through Hybrid Human-LLM Crowds
Axel Abels (Universit e Libre de Bruxelles), Tom Lenaerts (Universit e Libre de Bruxelles)
Explainability and InterpretabilityLarge Language ModelMixture of ExpertsText
🎯 What it does: Analyze the responses of large language models (LLMs) to bias-inducing headlines, comparing their bias and accuracy with humans; subsequently explore methods to mitigate bias through aggregating multiple models (LLM ensembles) and hybrid human-AI ensembles.
Witnesses for Answer Sets of Basic Logic Programs
Yisong Wang (State Key Laboratory of Public Big Data), Thomas Eiter (TU Wien)
Explainability and Interpretability
🎯 What it does: This paper addresses basic logic programs with abstract constraint atoms (c-atoms), proposing a new definition of minimal reduct and constructing an extensible CA-resolution proof. It further defines α-witness and β-witness to explain the reasons for atomic occurrences in SPT-answer sets.
Wrapped Partial Label Dimensionality Reduction via Dependence Maximization
Xiang-Ru Yu (Southeast University), Min-Ling Zhang (Southeast University)
OptimizationRepresentation LearningData-Centric LearningTabular
🎯 What it does: Propose a unified hierarchical partial label dimensionality reduction method called WPLDR, which simultaneously addresses feature dimensionality reduction and label deblurring.
X-KAN: Optimizing Local Kolmogorov-Arnold Networks via Evolutionary Rule-Based Machine Learning
Hiroki Shiraishi (Yokohama National University), Masaya Nakata (Yokohama National University)
OptimizationBenchmark
🎯 What it does: Propose X-KAN, a partitioned function approximation method that integrates the Kolmogorov-Arnold network (KAN) with the XCSF rule system, using local KAN models in rule consequents for approximation and employing evolutionary algorithms to adaptively partition rule antecedents;
Zero-shot Federated Unlearning via Transforming from Data-Dependent to Personalized Model-Centric
Wenhan Wu (Wuhan University), Dazhao Cheng (Wuhan University)
Federated LearningSafty and PrivacyKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: Proposed the ZeroFU framework, achieving zero-shot, data-free model-centralized client forgetting in federated learning.
Zero-shot Generalist Graph Anomaly Detection with Unified Neighborhood Prompts
Chaoxi Niu (University of Technology Sydney), Guansong Pang (Singapore Management University)
Anomaly DetectionGraph Neural NetworkPrompt EngineeringContrastive LearningGraph
🎯 What it does: Proposes UNPrompt, a zero-shot general graph anomaly detection framework that can detect abnormal nodes in any other arbitrary graph without any fine-tuning or labels, trained only on a single graph dataset.
Zero-Shot Machine Unlearning with Proxy Adversarial Data Generation
Huiqiang Chen (City University of Macau), Wanlei Zhou
Safty and PrivacyAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes a zero-shot machine unlearning framework ZS-PAG, which can remove the influence of specified samples from the original model under the condition that only the samples to be deleted are available.
Α Descent-based Method on the Duality Gap for Solving Zero-sum Games
Michail Fasoulakis (Royal Holloway, University of London), Christodoulos Santorinaios (Athens University of Economics and Business)
Optimization
🎯 What it does: Proposes a steepest descent method targeting the dual gap function in zero-sum games to solve approximate Nash equilibria.