Graph neural network

Graph neural networks (GNN) are specialized <a href="/facts/Artificial_neural_network/6V1jMlkx">artificial neural networks</a> that are designed for tasks whose inputs are <a href="/facts/Graph_(abstract_data_type)/b8qcddXa">graphs</a>.
One prominent example is molecular drug design. Each input sample is a graph representation of a molecule, where atoms form the nodes and chemical bonds between atoms form the edges. In addition to the graph representation, the input also includes known chemical properties for each of the atoms. Dataset samples may thus differ in length, reflecting the varying numbers of atoms in molecules, and the varying number of bonds between them. The task is to predict the efficacy of a given molecule for a specific medical application, like eliminating <a href="/facts/Escherichia_coli/nR4Gvl56">E. coli</a> bacteria.
The key design element of GNNs is the use of pairwise message passing, such that graph nodes iteratively update their representations by exchanging information with their neighbors. Several GNN architectures have been proposed, which implement different flavors of message passing, started by recursive or convolutional constructive approaches. As of 2022, it is an open question whether it is possible to define GNN architectures "going beyond" message passing, or instead every GNN can be built on message passing over suitably defined graphs.

In the more general subject of "geometric <a href="/facts/Deep_learning/JLuwD3ea">deep learning</a>", certain existing neural network architectures can be interpreted as GNNs operating on suitably defined graphs. A <a href="/facts/Convolutional_neural_network/kHEdnmGU">convolutional neural network</a> layer, in the context of <a href="/facts/Computer_vision/Tl2Yyk66">computer vision</a>, can be considered a GNN applied to graphs whose nodes are <a href="/facts/Pixel/UzwBvXBV">pixels</a> and only adjacent pixels are connected by edges in the graph. A <a href="/facts/Transformer_(machine_learning_model)/cDbjx6a8">transformer</a> layer, in <a href="/facts/Natural_language_processing/1hjMKsSN">natural language processing</a>, can be considered a GNN applied to <a href="/facts/Complete_graph/AYD6hUqI">complete graphs</a> whose nodes are <a href="/facts/Words/JzXV6kWh">words</a> or tokens in a passage of <a href="/facts/Natural_language/B4gLS7Kd">natural language</a> text.
Relevant application domains for GNNs include <a href="/facts/Natural_Language_Processing/1hjMKsSN">natural language processing</a>, <a href="/facts/Social_networks/3NkYaPIr">social networks</a>, <a href="/facts/Citation_graph/Ps8u3uFD">citation networks</a>, <a href="/facts/Molecular_biology/YgqU35ub">molecular biology</a>, chemistry, <a href="/facts/Physics/a7aH2y08">physics</a> and <a href="/facts/NP-hard/mQeNPv4R">NP-hard</a> <a href="/facts/Combinatorial_optimization/PTcloS79">combinatorial optimization</a> problems.
<a href="/facts/Open_source/7X9rCdz4">Open source</a> <a href="/facts/Library_(computing)/xRzL3P7q">libraries</a> implementing GNNs include PyTorch Geometric (<a href="/facts/PyTorch/4Gv1t13B">PyTorch</a>), TensorFlow GNN (<a href="/facts/TensorFlow/kDwfBeNd">TensorFlow</a>), Deep Graph Library (framework agnostic), jraph (<a href="/facts/Google_JAX/MnxmwXqc">Google JAX</a>), and GraphNeuralNetworks.jl/GeometricFlux.jl (<a href="/facts/Julia_(programming_language)/AoB0PJ9C">Julia</a>, <a href="/facts/Flux_(machine-learning_framework)/k5zTnDKh">Flux</a>).

Graph neural network open-in-new

Graph neural network