Deep Learning with Graph Convolutional Networks: An Overview and Latest Applications in Computational Intelligence

International Journal of Intelligent Systems, Journal Year: 2023, Volume and Issue: 2023, P. 1 - 28

Published: Feb. 28, 2023

Convolutional neural networks (CNNs) have received widespread attention due to their powerful modeling capabilities and been successfully applied in natural language processing, image recognition, other fields. On the hand, traditional CNN can only deal with Euclidean spatial data. In contrast, many real-life scenarios, such as transportation networks, social reference so on, exist graph The creation of convolution operators pooling is at heart migrating data analysis processing. With advancement Internet technology, network (GCN), an innovative technology artificial intelligence (AI), has more attention. GCN widely used different fields intelligent recommender system, knowledge-based graph, areas excellent characteristics processing non-European At same time, communication also embraced AI recent years, serves brain future realizes comprehensive grid. Many complex problems be abstracted graph-based optimization solved by GCN, thus overcoming limitations methods. This survey briefly describes definition machine learning, introduces types summarizes application various research fields, analyzes status, gives direction.

Language: Английский

---

Structure-Aware Positional Transformer for Visible-Infrared Person Re-Identification

IEEE Transactions on Image Processing, Journal Year: 2022, Volume and Issue: 31, P. 2352 - 2364

Published: Jan. 1, 2022

Visible-infrared person re-identification (VI-ReID) is a cross-modality retrieval problem, which aims at matching the same pedestrian between visible and infrared cameras. Due to existence of pose variation, occlusion, huge visual differences two modalities, previous studies mainly focus on learning image-level shared features. Since they usually learn global representation or extract uniformly divided part features, these methods are sensitive misalignments. In this paper, we propose structure-aware positional transformer (SPOT) network semantic-aware sharable modality features by utilizing structural information. It consists main components: attended structure (ASR) transformer-based interaction (TPI). Specifically, ASR models modality-invariant feature for each dynamically selects discriminative appearance regions under guidance TPI mines part-level position relations with With weighted combination TPI, proposed SPOT explores rich contextual information, effectively reducing difference enhancing robustness against Extensive experiments indicate that superior state-of-the-art cross-modal datasets. Notably, Rank-1/mAP value SYSU-MM01 dataset has improved 8.43%/6.80%.

Language: Английский

---

Learnable graph convolutional network and feature fusion for multi-view learning

Information Fusion, Journal Year: 2023, Volume and Issue: 95, P. 109 - 119

Published: Feb. 16, 2023

Language: Английский

---

Deep learning for Alzheimer's disease diagnosis: A survey

Artificial Intelligence in Medicine, Journal Year: 2022, Volume and Issue: 130, P. 102332 - 102332

Published: June 12, 2022

Language: Английский

---

Adaptive Feature Projection With Distribution Alignment for Deep Incomplete Multi-View Clustering

IEEE Transactions on Image Processing, Journal Year: 2023, Volume and Issue: 32, P. 1354 - 1366

Published: Jan. 1, 2023

Incomplete multi-view clustering (IMVC) analysis, where some views of data usually have missing data, has attracted increasing attention. However, existing IMVC methods still two issues: (1) they pay much attention to imputing or recovering the without considering fact that imputed values might be inaccurate due unknown label information, (2) common features multiple are always learned from complete while ignoring feature distribution discrepancy between and incomplete data. To address these issues, we propose an imputation-free deep method consider alignment in learning. Concretely, proposed learns for each view by autoencoders utilizes adaptive projection avoid imputation All available projected into a space, cluster information is explored maximizing mutual achieved minimizing mean discrepancy. Additionally, design new loss learning make it applicable mini-batch optimization. Extensive experiments demonstrate our achieves comparable superior performance compared with state-of-the-art methods.

Language: Английский

---

Variational gated autoencoder-based feature extraction model for inferring disease-miRNA associations based on multiview features

Neural Networks, Journal Year: 2023, Volume and Issue: 165, P. 491 - 505

Published: June 5, 2023

Language: Английский

---

Multiplex Graph Representation Learning Via Dual Correlation Reduction

IEEE Transactions on Knowledge and Data Engineering, Journal Year: 2023, Volume and Issue: 35(12), P. 12814 - 12827

Published: April 26, 2023

Recently, with the superior capacity for analyzing multiplex graph data, self-supervised representation learning (SMGRL) has received much interest. However, existing SMGRL methods are still limited by following issues: (i) they generally ignore noisy information within each and common among different graphs, thus weakening effectiveness of SMGRL, (ii) conduct negative sample encoding complex pretext tasks contrastive learning, efficiency SMGRL. To solve these issues, in this work, we propose a new framework to effective efficient Specifically, proposed method investigates intra-graph inter-graph decorrelation losses, respectively, reducing impact capturing achieve effectiveness. Moreover, does not need samples designs simple task, efficiency. We further theoretically justify that our achieves maximal mutual instead directly conducting actually minimizes bottleneck, which guarantees In addition, an extension semi-supervised scenarios is fit case few labels provided reality. Extensive experimental results verify respect various downstream tasks.

Language: Английский

---

Alzheimer’s Disease Detection Using Deep Learning on Neuroimaging: A Systematic Review

Machine Learning and Knowledge Extraction, Journal Year: 2024, Volume and Issue: 6(1), P. 464 - 505

Published: Feb. 21, 2024

Alzheimer’s disease (AD) is a pressing global issue, demanding effective diagnostic approaches. This systematic review surveys the recent literature (2018 onwards) to illuminate current landscape of AD detection via deep learning. Focusing on neuroimaging, this study explores single- and multi-modality investigations, delving into biomarkers, features, preprocessing techniques. Various models, including convolutional neural networks (CNNs), recurrent (RNNs), generative are evaluated for their performance. Challenges such as limited datasets training procedures persist. Emphasis placed need differentiate from similar brain patterns, necessitating discriminative feature representations. highlights learning’s potential limitations in detection, underscoring dataset importance. Future directions involve benchmark platform development streamlined comparisons. In conclusion, while learning holds promise accurate refining models methods crucial tackle challenges enhance precision.

Language: Английский

---

A New Brain Network Construction Paradigm for Brain Disorder via Diffusion-Based Graph Contrastive Learning

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2024, Volume and Issue: 46(12), P. 10389 - 10403

Published: Aug. 13, 2024

Brain network analysis plays an increasingly important role in studying brain function and the exploring of disease mechanisms. However, existing construction tools have some limitations, including dependency on empirical users, weak consistency repeated experiments time-consuming processes. In this work, a diffusion-based pipeline, DGCL is designed for end-to-end networks. Initially, region-aware module (BRAM) precisely determines spatial locations regions by diffusion process, avoiding subjective parameter selection. Subsequently, employs graph contrastive learning to optimize connections eliminating individual differences redundant unrelated diseases, thereby enhancing networks within same group. Finally, node-graph loss classification jointly constrain process model obtain reconstructed network, which then used analyze connections. Validation two datasets, ADNI ABIDE, demonstrates that surpasses traditional methods other deep models predicting development stages. Significantly, proposed improves efficiency generalization construction. summary, can be served as universal scheme, effectively identify through generative paradigms has potential provide interpretability support neuroscience research.

Language: Английский

---

Alzheimer’s disease diagnosis from single and multimodal data using machine and deep learning models: Achievements and future directions

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 255, P. 124780 - 124780

Published: July 14, 2024

Language: Английский

---