8. 2019
●
2018
●
● ODSC East 2018, the Open Source Data Science
Project Award
● CEATEC Award
2017
●
● Japan-U.S. Innovation Awards Emerging Leader
Award
● FT ArcelorMittal Boldness in Business Awards
2016
● JEITA
● Forbes JAPAN’s CEO OF THE YEAR 2016
1
17. ü NN Python
→ NN
ü Define-by-Run NN
→ Python
→ NN
ü CuPy NumPy-like GPU
→ CPU/GPU agnostic
Speed up research and development of deep learning and its applications.
25. Backpropagation.
10
y = x1 * x2
ga * z * a ga ∇a z
x1 mul suby
x3
z
x2
z = y - x3
gx1 = gy * x2
gx2 = gy * x1
gz = 1
gy = gz
gx3 = -gz
gzgy
gx3
gx1
gx2
26. Multi Layer Perceptron; MLP
x Linear
W1 b1
h1 ReLU a1
Linear
W2 b2
h2
Soft
max
prob
Cross
Entropy
loss
t
11
1
2
27. 12
class MLP(chainer.Chain):
def __init__(self):
super(MLP, self).__init__()
with self.init_scope():
self.fc1 = L.Linear(None, 100)
self.fc2 = L.Linear(100, 10)
def forward(self, x):
a1 = F.relu(self.fc1(x))
h2 = self.fc2(a1)
return h2
model = MLP()
model = L.Classifier(model)
x Linear
W1 b1
h1 ReLU
a1 Linear
W2 b2
h2
40. DL (1)
DeepVariant ⇒ CNN
(Shanrong Zhao et al., Cloud Computing for Next-Generation Sequencing Data Analysis, 2017 )
(Poplin, Ryan et al., “A universal SNP and small-indel variant caller using deep neural networks.”, 2018 )
•
• DeepVariant (https://github.com/google/deepvariant)
Pileup image
41. DL (2)
AlphaFold
CASP (Critical Assessment of
techniques for protein Structure
Prediction)
2
(http://predictioncenter.org/casp13/doc/CAS
P13_Abstracts.pdf )
44. AI
• AI AI
• https://japan-medical-ai.github.io/medical-ai-course-materials/
• GitHub https://github.com/japan-medical-ai/medical-ai-course-materials
• AI AI https://www.japan-medical-ai.org/?page_id=26
•
• Python Google Colaboratory
1.
2.
3.
45. 1)
• Google Colaboratory Web
• Google
• GPU
1. https://colab.research.google.com/
2. Colaboratory ( )
3. GPU
4.
5.
Colab
62. ( )
C
N
O
1.0 0.0 0.0 6.0 1.0
atom type
0.0 1.0 0.0 7.0 1.0
0.0 0.0 1.0 8.0 1.0
charge
chirality
Man-made features
Molecular Graph Convolutions: Moving Beyond Fingerprints
Steven Kearnes, Kevin McCloskey, Marc Berndl, Vijay Pande, Patrick Riley arXiv:1603.00856
63. Graph Convolution: ( )
Graph Convolution
Han Altae-Tran, Bharath Ramsundar, Aneesh S. Pappu, & Vijay Pande (2017). Low Data Drug
Discovery with One-Shot Learning. ACS Cent. Sci., 3 (4)
64. Graph Readout: ( )
Han Altae-Tran, Bharath Ramsundar, Aneesh S. Pappu, & Vijay Pande (2017). Low Data Drug
Discovery with One-Shot Learning. ACS Cent. Sci., 3 (4)
75. Ours Average
(18 teams in total)
1st screening (TSA) 23 / 200 (11.5%) 69 / 3559 (1.9 %)
2nd screening (IC50) 1 5
We found one hit compound and
won one of Grand prize (IPAB )
79. BayesGrad
“BayesGrad: Explaining Predictions of Graph Convolutional Networks”
Akita et al., ICONIP 2018
https://arxiv.org/abs/1807.01985
https://github.com/pfnet-research/bayesgrad
Dropout Saliency
80. GWM
“Graph Warp Module: an Auxiliary Module for Boosting the Power of Graph
Neural Networks in Molecular Graph Analysis” Ishiguro et al.
https://arxiv.org/abs/1902.01020
https://github.com/pfnet-research/chainer-chemistry
Graph Convolution Neural Network
81. GraphNVP
“GraphNVP: An Invertible Flow Model for Generating Molecular Graphs”
Madhawa et al.
https://arxiv.org/abs/1905.11600
https://github.com/pfnet-research/graph-nvp
Flow