Graph Representation Learning with Deep Embedding Approach
summary
Presents the structure2vec end-to-end graph embedding that maps node neighborhoods to vectors and is trained on supervised tasks; shows lower regression loss than convolution baselines and scales to a 2.3 million‑molecule screening dataset while trading off model size vs mean absolute error. Extends embeddings to temporal user–item graphs for better recurring‑event recommendation. Uses reinforcement learning for vertex cover to learn heuristics that train on 50–100 node graphs, generalize to thousands, achieve ~1.007 approximation and produce near‑optimal solutions in ~0.1s versus hundreds of seconds.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:17Graph applications
- 1:10Outline
- 3:06Review: Temporal Recommendation
- 3:46Review: Graph Combinatorial Optimization
- 7:18Experiment on RNA sequences
- 7:42Experiment on Molecules
- 8:54Interpretable results
- 9:39Extension to network analysis
- 10:07Experiment on network analysis
- 10:38Dynamic Graphs for Recommendation
- 11:36Experiment on recommendation
- 12:23Learning graph opt: Motivation
- 14:39Learning graph opt: RL background
- 15:41Learning graph opt: RL on graphs
- 17:08Learning graph opt: quantitative comparison
related talks
31mGraph Mining with Deep Learning: challenges and pitfalls
37mLike2Vec with neural network embeddings
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
38mContinuous Representation Of Molecules using Graph Variational Autoencoder
16mPreview: Graph NNs
30mGraph Machine Learning in All Its Glory!
39mUsing Graph Neural Networks to Embrace the Dependency Within Your Data
16mDeep Stream Dynamic Graph Analytics with Grapharis