© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Sung Kim, Xingjian Shi
홍콩과기대 (HKUST)
모두를 위한 MXNET
본 강연에서 다룰 내용
●MXNET
●Linear Regression, Logistic Regression
●Deep Neural Net (DNN)
●CNN
●RNN
●MXNET in AWS
● Open Source deep learning framework
● Most scalable and flexible framework
● Partner with AWS
http://www.allthingsdistributed.com/2016/11/mxnet-default-framework-deep-learning-aws.html
https://aws.amazon.com/mxnet/
Why MXNET?
http://mxnet.io/architecture/program_model.html
Imperative vs Symbolic Programs
● Imperative-style programs perform computation as you run them
● In symbolic programs, when program is executed, no computation
occurs.
● Instead, these operations generate a computation graph
http://mxnet.io/architecture/program_model.html
Imperative Programs Are More Flexible
● Imperative-style programs perform computation as you run them
http://mxnet.io/architecture/program_model.html
Symbolic Programs Are More Efficient
http://mxnet.io/architecture/program_model.html
Running code in GPU
http://mxnet.io/architecture/program_model.html
NN Layers
(Linear regression) Hypothesis
x=2.5, y=?
(Linear regression) Hypothesis
Simplified hypothesis
Logistic Regression
http://cs231n.github.io/neural-networks-1/
Logistic Regression
Logistic regression
https://nbviewer.jupyter.org/github/sxjscience/DeepLearningZeroToAll/blob/master/aws-2017-seoul/mxnet-logistic_regression_diabetes.ipynb
Forward Neural Net
http://cs231n.github.io/neural-networks-1/
Forward NN Code (Demo)
CNN
http://parse.ele.tue.nl/cluster/2/CNNArchitecture.jpg
Convolution layer and max pooling
Simple convolution layer
Image: 1,3,3,1 image, Filter: 2,2,1,1, Stride: 1x1, Padding: VALID
1
1
1
1
[[[[1.]],[[1.]]],
[[[1.]],[[1.]]]]
shape=(2,2,1,1)
Max Pooling
Deep CNN
Image credit: http://personal.ie.cuhk.edu.hk/~ccloy/project_target_code/index.html
CNN
Inception-V3
RNN
1) Using native implementation
rnn = mx.rnn.LSTMCell (num_hidden=2, prefix="lstm1_")
1) Using CUDNN V5.1
rnn = mx.rnn.FusedRNNCell (num_hidden=2, prefix="lstm1_")
Unfolding to n sequences
Hidden_size=2
sequence_length=5
Sequence to sequence model
Sequence to Sequence
MXNET on AWS
Instance: g2.2x-large (K520)
Task: CIFAR-10 with ResNet-50
AWS Deep Learning AMI
Up to~40k CUDA cores
Apache MXNet
TensorFlow /Theano / Caffe
Torch / Keras
Pre-configured CUDA drivers,
MKL
Anaconda, Python3
Ubuntu and Amazon Linux
One-Click GPU or CPU
Deep Learning
+ CloudFormation template
+ Container Image
Summary
● MXNET
● Linear Regression, Logistic Regression
● Deep Neural Net (DNN)
● CNN
● RNN
● MXNET in AWS
본 강연이 끝난 후…
MXNet Resources:
• MXNet Blog Post | AWS Endorsement
• Read up on MXNet and Learn More: mxnet.io
• MXNet Github Repo
• MXNet Recommender Systems Talk | Leo Dirac
Developer Resources:
• Deep Learning AMI |Amazon Linux
• Deep Learning AMI | Ubuntu
• CloudFormation Template Instructions
• Deep Learning Benchmark
• MXNet on Lambda
• MXNet on ECS/Docker
• MXNet on Raspberry Pi | Image Detector using Inception Network
Thank you!
함께 해주셔서 감사합니다!
https://www.awssummit.kr
AWS Summit 모바일 앱을 통해 지금 세션 평가에
참여하시면, 행사 후 기념품을 드립니다.
#AWSSummitKR 해시태그로 소셜 미디어에 여러분
의 행사 소감을 올려주세요.
발표 자료 및 녹화 동영상은 AWS Korea 공식 소셜 채널
로 공유될 예정입니다.
여러분의 피드백을 기다립니다!

모두를 위한 MxNET - AWS Summit Seoul 2017