Article
Mathematics
Jose M. Cuevas-Munoz, Nicolas E. Garcia-Pedrajas
Summary: Multi-label classification is an increasingly popular data mining task, and the ML-kNN approach has shown promising results. However, current methods using a single k value for all labels may not consider the different label distributions. In this paper, we propose a novel approach of predicting each label using a different k value, optimizing the best k for each label. Our approach outperforms two different tested ML-kNN implementations in a large set of 40 real-world multi-label problems.
Article
Computer Science, Artificial Intelligence
Ming-Kun Xie, Sheng-Jun Huang
Summary: The proposed approach tackles partial multi-label learning by recovering ground-truth information and identifying noisy labels simultaneously. By formalizing two objectives in a unified framework with trace norm and l(1) norm regularizers, the multi-label classifier and noisy label identifier are jointly optimized, incorporating label correlation exploitation and feature-induced noise model. Extensive experiments on multiple real-world tasks demonstrate the effectiveness of the approach.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Sichao Qiu, Mengyi Wang, Yuanlin Yang, Guoxian Yu, Jun Wang, Zhongmin Yan, Carlotta Domeniconi, Maozu Guo
Summary: Multi-Instance Multi-Label Learning (MIML) is a learning model that can handle complex objects. Current solutions focus on a single type of data and assume an independent and identical distribution. To better handle MIML objects linked with objects of other types, we propose a heterogeneous network embedding and meta learning based approach (MetaMIML). Experiments demonstrate that MetaMIML achieves significantly better performance than state-of-the-art algorithms.
INFORMATION FUSION
(2023)
Article
Computer Science, Artificial Intelligence
Amelia Zafra, Eva Gibaja
Summary: This work presents new neighbor-based approaches for solving the multi-instance multi-label (MIML) problem. The performance of these methods is studied using different configurations, and the results show that the problem transformation applied and the distance function used impact the performance. Most of the proposed algorithms outperformed the state-of-art MIMLkNN algorithm. These findings demonstrate the relevance and capabilities of neighbor-based approaches in MIML learning. Furthermore, all the algorithms developed in this paper have been included in the MIML library for future comparisons.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Yongjian Zhong, Bo Du, Chang Xu
Summary: This study presents a new method for reweighting examples in multi-label classification problems by upgrading classical weight functions through considering instance complexities. Experimental results demonstrate the significance of this algorithm in multi-label classification problems.
Article
Computer Science, Artificial Intelligence
Yibin Wang, Weijie Zheng, Yusheng Cheng, Dawei Zhao
Summary: Different from single-label learning, multi-label learning with rich semantic information requires label embedding to capture the inherent intelligence of the label space. However, due to the incompleteness of the label space, label data recovery becomes crucial. This paper proposes a two-level label recovery mechanism for multi-label classification, which effectively addresses missing labels in incomplete datasets and improves classification performance by capturing instance and label correlations in the recovered label space.
APPLIED SOFT COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Nicolas Garcia-Pedrajas, Jose M. Cuevas-Munoz, Juan A. Romero del Castillo, Aida de Haro-Garcia
Summary: Multi-label classification is an important task in data mining, and instance selection is a key approach to improve its efficiency and accuracy. In this paper, a novel instance selection method is proposed, which allows each instance to be selectively used for different labels. Experimental results on multiple datasets demonstrate the effectiveness and competitiveness of the proposed method compared to other existing approaches.
INFORMATION FUSION
(2023)
Article
Mathematics
Runxin Li, Jiaxing Du, Jiaman Ding, Lianyin Jia, Yinong Chen, Zhenhong Shang
Summary: In this paper, a new approach called SMDR-IC is proposed for semi-supervised multi-label dimensionality reduction learning. The proposed method incorporates label propagation mechanism and investigates instance correlations using the k-nearest neighbor technique. Experimental results show that SMDR-IC outperforms other related methods on public multi-label datasets.
Article
Computer Science, Artificial Intelligence
Alvaro Belmonte, Amelia Zafra, Eva Gibaja
Summary: The MIML library is a Java software tool for developing, testing, and comparing classification algorithms for MIML learning. It includes 43 algorithms, provides specific data management and partitioning formats, supports various evaluation methods, and allows algorithms to be executed through xml configuration files.
Article
Computer Science, Artificial Intelligence
Yang Yang, Zhao-Yang Fu, De-Chuan Zhan, Zhi-Bin Liu, Yuan Jiang
Summary: This study proposes a novel deep network model for Multi-modal Multi-instance Multi-label (M3) learning, which effectively utilizes label correlation and multi-modal learning to process unlabeled instances. The model achieves better label prediction and exploits label correlation simultaneously.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2021)
Article
Computer Science, Information Systems
Wanting Ji, Ruili Wang
Summary: A novel multi-instance multi-label dual learning approach (MIMLDL) is proposed for video captioning, utilizing an encoder-decoder-reconstructor structure to generate video captions by minimizing the semantic gap between raw videos and generated captions. The approach consists of caption generation and video reconstruction modules, fine-tuned by a dual learning mechanism to reproduce visual sequences and improve the accuracy of video captioning.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yuling Fan, Jinghua Liu, Peizhong Liu, Yongzhao Du, Weiyao Lan, Shunxiang Wu
Summary: The study proposes a structured subspace manifold learning approach for multi-label feature selection, including uncovering latent subspace, exploring instance correlations and label correlations, and introducing l2,1-norm and sparse regularization. Experimental results demonstrate the superiority of the method under various metrics.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Haikun Li, Min Fang, Hang Li, Peng Wang
Summary: In multi-label learning, training data often contains noisy and redundant instances. Traditional classification methods with raw data can be memory-intensive and have lower performance. We propose a novel approach called CO-GCNN, which uses co-occurrence and generalized condensed nearest neighbor to reduce the number of instances in multi-label learning. Experimental results on six benchmark datasets demonstrate the effectiveness of the proposed CO-GCNN method in improving classification performance.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Biochemical Research Methods
Jing Liu, Xinghua Tang, Shuanglong Cui, Xiao Guan
Summary: In this paper, a novel MIML learning framework MK-EnMIMLNN is proposed for rice protein function prediction. This framework combines the residue couple model method and the pseudo amino acid composition method with Principal Component Analysis, and utilizes a multiple kernel fusion function neural network. Experimental results show that the hybrid feature extraction method is better than the single feature extraction method, and the MK-EnMIMLNN algorithm outperforms most classic MIML learning algorithms in rice protein function prediction.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Xuewei Li, Hongjun Wu, Mengzhu Li, Hongzhe Liu
Summary: This article introduces a framework called MALL-CNN, which is a Multi-Attention Label Relation Learning Convolutional Neural Network, for solving the problem of multi-label video classification. The framework establishes correspondences between labels and videos using attention mechanisms and captures label co-occurrence through graph learning methods. Experimental results demonstrate that MALL-CNN outperforms other models using only frame features in the context of multi-modal features and ensemble models.
PATTERN RECOGNITION LETTERS
(2022)