Label-Guided Masking and Imputation Network for Incomplete Multi-View Multi-Label Classification

计算机科学 人工智能 模式识别(心理学) 遮罩(插图) 插补(统计学) 数据挖掘 稳健性(进化) 电子邮件 缺少数据 信号处理 人工神经网络 算法 特征提取 数据建模 图像处理
作者
J Y Chen,Mengqing Wang,Tong Guo,Xiaohuan Lu,Youliang Tian
出处
期刊:IEEE Transactions on Circuits and Systems for Video Technology [Institute of Electrical and Electronics Engineers]
卷期号:36 (10): 15141-15158
标识
DOI:10.1109/tcsvt.2026.3707408
摘要

In recent years, multiview multilabel classification has gained attention because of its ability to assign multiple labels to each instance using information from different views. However, real-world applications often face challenges caused by sensor malfunctions, communication errors, or human annotation mistakes, which can result in missing views or labels. Despite progress in the field, two key limitations persist: first, the cross-view data collection process may introduce irrelevant features, which reduces the discriminative power of the model; second, traditional zero-filling imputation fails to fully leverage the complex relationships between views and feature-label dependencies, which may affect imputation and classification performance. To address these issues holistically, we propose a label-guided masking and Imputation (LGMI) network that leverages label supervision to enhance both information selection and recovery processes. Our model introduces two key components: supervision-driven adaptive dynamic masking (SAM) and a label-guided imputation learning (GIL) module. SAM uses supervised label information to dynamically identify important semantic regions and filter out irrelevant features, enabling the model to focus on the most relevant data and increase stability in complex environments. GIL aggregates available view features to infer missing content, ensuring high-fidelity reconstruction within the embedding space. This reconstruction is then integrated with the original features into a unified, label-guided semantic alignment framework, which leverages the underlying semantic structures to maintain consistency and increase the robustness of the reconstructed representations within the semantic manifold. Extensive experiments performed on five datasets show that our approach outperforms several leading methods.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
shengji发布了新的文献求助10
刚刚
shangguang的应助被zzl采纳,获得10
1秒前
1秒前
2秒前
freak发布了新的文献求助10
2秒前
桃子汽水完成签到 ,获得积分10
3秒前
英俊的铭的应助被小王采纳,获得10
3秒前
夏安发布了新的文献求助10
3秒前
若梦发布了新的文献求助10
4秒前
CDD完成签到,获得积分10
4秒前
虚拟的信封完成签到,获得积分10
4秒前
4秒前
红霉素软膏完成签到,获得积分10
4秒前
贤君发布了新的文献求助20
5秒前
5秒前
隐世求开完成签到,获得积分10
5秒前
5秒前
乔治发布了新的文献求助10
5秒前
5秒前
6秒前
MM的应助被之星君采纳,获得10
6秒前
小阮完成签到,获得积分10
6秒前
7秒前
顺利的水瑶完成签到,获得积分10
7秒前
7秒前
Savage完成签到,获得积分10
8秒前
jhlllllll完成签到,获得积分20
8秒前
8秒前
8秒前
8秒前
9秒前
萤火虫发布了新的文献求助10
9秒前
9秒前
芽芽乐完成签到,获得积分10
9秒前
英姑的应助被杨九斤Jenney采纳,获得10
9秒前
南一发布了新的文献求助10
10秒前
泽ze发布了新的文献求助50
10秒前
11秒前
吱吱吱发布了新的文献求助10
11秒前
chenoo完成签到,获得积分10
11秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
What Else Can I Tell You?: A Pragmatic Study of English Rhetorical Questions as Discursive and Argumentative Acts 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8164445
求助须知:如何正确求助?哪些是违规求助? 9558283
关于积分的说明 21744871
捐赠科研通 7754979
什么是DOI,文献DOI怎么找? 3428582
关于科研通互助平台的介绍 2553119
邀请新用户注册赠送积分活动 2473142