دانلود کتاب شبکه‌های عصبی گراف: ملزومات و موارد استفاده

The paradigm of neural networks (NNs) has gained surging popularity these days as it has turned out to be a powerful technology for unearthing hidden patterns in massive datasets. Undoubtedly, the neural network concept is transformative, with worldwide researchers unearthed and expounded a series of innovations and disruptions. One noteworthy advancement is forming different deep neural network (DNN) types. Convolutional neural networks (CNNs) are prominent DNNs for tasks related to computer vision (CV). Recurrent neural networks (RNNs) and their advanced variants are DNNs that automate natural language processing (NLP) tasks such as speech recognition, translation, and generation. The DNNs mainly deal with Euclidean data. However, non-Euclidean data comprising complex item descriptions and interactions is typically coded in graphs. Specific applications need to tackle non-Euclidean data. This is the key motivation for adopting the graph neural networks (GNN) phenomenon, a transformative power that will inspire and intrigue you.
GNNs are a powerful deep learning (DL) model that works with graph-structured data. A graph consists of nodes representing individual data points and edges representing the relationships or connections between those nodes. There are a few GNN variants, such as graph convolutional networks (GCN), graph attentional networks (GAT), and graph recurrent networks (GRN). GCNs can learn from the graph structure by deeply looking at a node and its neighbours—GCNs aggregate information from the neighbouring nodes to update the node’s representation accordingly. Applying CNN to graph data is computationally challenging as the underlying topology is arbitrary and complicated. There is no spatial locality and an unfixed node ordering. GATs introduce the hugely popular attention mechanism that allows the model to focus on specific neighbours when aggregating information. That is, some neighbours contribute more to the task at hand. GRNs combine the principles of RNNs with graph structures. There are spatial and spectral-based GNNs.
The Use Cases of GNNs: There are advanced use cases such as node classification, link prediction, graph clustering, visualization and classification, anomaly detection, recommendation systems, NLP, adversarial attack prevention, and computer vision. GNNs contribute to forecasting traffic speed, volume, or the density of roads in traffic networks. Chemists can use GNNs to research the graph structure of molecules or compounds. GNNs contribute to problems such as programme verification and reasoning, social influence prediction, electrical health records (EHR) modelling, and brain networks.
This book has incorporated several changes to demystify the enabling GNN technologies and tools. Various industrial applications, businesses, and use cases are meticulously delineated to empower our readers. AI and graph theory researchers will get details on the latest trends and transitions being unearthed in the GNN space, ensuring that you are up-to-date and knowledgeable. A few practical chapters are stuffed in this book to explain how GNNs are powerful and pioneering across industry verticals. This book is a guide for AI engineers and data scientists. The latest developments of generative AI, agentic AI, and physical AI are to be significantly emboldened by the distinct improvisations articulated and accentuated in the GNN space.