Deep Graph Library: Deep Graph learning at scale
summary
DGL 0.6 is an open-source graph neural network library built on PyTorch/MXNet/TensorFlow with domain packages like DGL-LifeSci and DGL-KG, implements optimized C++ graph operators, and exposes single-call message-passing APIs. Distributed training partitions graphs and uses minibatch neighbor sampling; it scales to 10 million nodes and 3.2 billion edges and achieves 15.3× speedup on 16 machines while reporting no accuracy degradation.
AI-written summary · about
chapters (13)
- 0:00Introduction to DGL
- 1:11Graph learning applications
- 5:49Graph learning problem types
- 8:49Limitations of traditional methods
- 9:11Graph neural networks overview
- 12:30Message passing in GNNs
- 13:55Relational graph convolution
- 15:22Scaling GNNs with mini batches
- 18:59Neighbor sampling for large graphs
- 19:22Why DGL is needed
- 20:49DGL overview and features
- 23:09DGL distributed training
- 25:24Conclusion and resources
related talks
31mDGL Operator: Distributed Graph Neural Network Training with DGL and K8s
14mP3: Distributed Deep Graph Learning at Scale
16mDeep Stream Dynamic Graph Analytics with Grapharis
19mNeuGraph: Parallel Deep Neural Network Computation on Large Graphs
15mMGG: Accelerating Graph Neural Networks with Fine-Grained Intra-Kernel Communication...
29mGraph Neural Networks + LLM on
15mPL4XGL: A Programming Language Approach to Explainable Graph Learning
50mDeep Learning with Go