SLIDE 1
https://arxiv.org/abs/1711.00740
https://arxiv.org/abs/1711.00740 Li, Y., Tarlow, D., Brockschmidt, - - PowerPoint PPT Presentation
https://arxiv.org/abs/1711.00740 Li, Y., Tarlow, D., Brockschmidt, - - PowerPoint PPT Presentation
https://arxiv.org/abs/1711.00740 Li, Y., Tarlow, D., Brockschmidt, M., & Zemel, R. (2015). Gated graph sequence neural networks. arXiv preprint arXiv:1511.05493 . a a b b c c d d e e f f g g h h A A B B C C D D E E G
SLIDE 2
SLIDE 3
Li, Y., Tarlow, D., Brockschmidt, M., & Zemel, R. (2015). Gated graph sequence neural networks. arXiv preprint arXiv:1511.05493.
SLIDE 4
a b c d e f g h a b c d e f g h
SLIDE 5
B A E G D C F
Li et al (2015). Gated graph sequence neural networks.
B A E G D C F
SLIDE 6
E D F
Li et al (2015). Gated graph sequence neural networks.
B A E G D C F
SLIDE 7
Li et al (2015). Gated graph sequence neural networks.
SLIDE 8
Li et al (2015). Gated graph sequence neural networks.
- node selection
- node classification
- graph classification
SLIDE 9
https://arxiv.org/abs/1711.00740
SLIDE 10
SLIDE 11
SLIDE 12
SLIDE 13
SLIDE 14
SLIDE 15
SLIDE 16
SLIDE 17
B A E G D C F
SLIDE 18
Implementation
- Sparse TensorFlow implementation
- 16 edge types (forward + backward)
- ~900 nodes/graph ~8k edges/graph
Stats
- 55 graphs/s during training
- 219 graphs/s during testing
SLIDE 19
SLIDE 20
Accurac curacy y (%) Local cal Model del Avg vg BiRN RNN GGNN Seen Projects 15.8 73.5 82.1 Unseen Projects 13.8 59.7 68.6
3.8 type-correct alternative variables per slot (median 3, σ= 2.6)
SLIDE 21
bool string string
- ut string
var while null if return true null return false
SLIDE 22
bool string string
- ut string
var while null if return true null return false
SLIDE 23
SLIDE 24
SLIDE 25
SLIDE 26
SLIDE 27
SLIDE 28
SLIDE 29