google-deepmind/graph_nets
Graph Nets
graph_nets is a graph-structured deep learning framework and library for building message-passing neural networks. It provides tools for designing architectures that operate on nodes and edges to process and reason about data structured as graphs using TensorFlow.
The framework implements a message-passing paradigm for iterative information exchange between nodes. This approach enables the development of models that can reason about complex graph-structured inputs for tasks such as path-finding and sorting, or serve as a predictor for the future states and trajectories of physical systems.
Features
- Graph Message Passing Frameworks - Provides a framework for implementing message passing primitives to learn representations of nodes and edges in graphs.
- Geometric Deep Learning Frameworks - Provides a framework for applying deep learning to non-Euclidean data such as molecular graphs.
- Graph Neural Networks - Provides a library for building neural network architectures designed to process data represented as graphs.
- Message Passing Primitives - Implements differentiable compute patterns for aggregating and updating features across graph topologies.
- TensorFlow Graph Execution - Utilizes TensorFlow's computational graphs to execute mathematical operations on hardware accelerators.
- Graph Representations - Provides data structures that store graph topology and edge attributes for neural network processing.
- Fixed-Point Convergence - Processes graph information repeatedly until node and edge representations reach a stable state.
- Graph Reasoning Systems - Implements systems for reasoning over graph-structured data to perform tasks like path-finding and sorting.
- Iterative Prediction Refiners - Feeds model outputs back as inputs to iteratively refine predictions and simulate system evolution.
- Physical System Trajectories - Forecasts the future state of a physical system by iteratively feeding model predictions back as inputs.
- Graph Computation - Implements graph-based computation to perform complex tasks such as path-finding or sorting.
- Physical System Trajectory Predictors - Iteratively predicts the future state of physical systems by processing graph-based spatial data.
