Koss 6 a17_deepmachinelearning_mariocho_r10

49
딥러닝의 이해와 응용 (Deep Machine Learning KOSS Lab. 2Mario Cho ( 조만석) [email protected]

Transcript of Koss 6 a17_deepmachinelearning_mariocho_r10

Page 1: Koss 6 a17_deepmachinelearning_mariocho_r10

딥러닝의이해와응용

(Deep Machine LearningKOSS Lab. 2기

Mario Cho (조만석)

[email protected]

Page 2: Koss 6 a17_deepmachinelearning_mariocho_r10

Contents

• 기계학습 (NNs)

• Neural Networks

• 딥러닝기술

• 활용방안

Page 3: Koss 6 a17_deepmachinelearning_mariocho_r10

Mario ChoDevelopment Experience◆ Image Recognition using Neural Network◆ Bio-Medical Data Processing◆ Human Brain Mapping on High Performance

Computing◆ Medical Image Reconstruction

(Computer Tomography) ◆Enterprise System◆Open Source Software Developer

Open Source Software Developer◆ OPNFV (NFV&SDN) & OpenStack◆ Machine Learning (TensorFlow)

Book◆ Unix V6 Kernel Korea Open Source Software Lab.

Mario [email protected]

Page 4: Koss 6 a17_deepmachinelearning_mariocho_r10

Today’s information

* http://www.cray.com/Assets/Images/urika/edge/analytics-infographic.html

Page 5: Koss 6 a17_deepmachinelearning_mariocho_r10

The Future of Jobs

“The Fourth Industrial Revolution, which includes developments in previously disjointed fields such as artificial intelligence & machine-learning, robotics, nanotechnology, 3-D printing, and genetics & biotechnology, will cause widespread disruption not only to business models but also to labor market over the next five years, with enormous change predicted in the skill sets needed to thrive in the new landscape.”

Page 6: Koss 6 a17_deepmachinelearning_mariocho_r10

What is the Machine Learning ?• Field of Computer Science that evolved from the study of pattern recognition and computational learning theory into Artificial Intelligence.

• Its goal is to give computers the ability to learn without being explicitly programmed.

• For this purpose, Machine Learning uses mathematical / statistical techniques to construct models from a set of observed data rather than have specific set of instructions entered by the user that define the model for that set of data.

Page 7: Koss 6 a17_deepmachinelearning_mariocho_r10

Required of New type of Computing understand information, to learn, to reason, and act upon it

Page 8: Koss 6 a17_deepmachinelearning_mariocho_r10

Human Intelligence

Page 9: Koss 6 a17_deepmachinelearning_mariocho_r10

Brain Map

Page 10: Koss 6 a17_deepmachinelearning_mariocho_r10

Neuron

Page 11: Koss 6 a17_deepmachinelearning_mariocho_r10

hippocampus

Page 12: Koss 6 a17_deepmachinelearning_mariocho_r10

Neural network vs Learning networkNeural Network Deep Learning Network

Page 13: Koss 6 a17_deepmachinelearning_mariocho_r10

Neural Network

W1

W2

W3

f(x)

1.4

-2.5

-0.06

Page 14: Koss 6 a17_deepmachinelearning_mariocho_r10

Neural Network

2.7

-8.6

0.002

f(x)

1.4

-2.5

-0.06

x = -0.06×2.7 + 2.5×8.6 + 1.4×0.002 = 21.34

Page 15: Koss 6 a17_deepmachinelearning_mariocho_r10

Neural Network : x1 XNOR x2

+1

x1

x2

+1

a1(2)

a2(2)

a1(3)

-30

20

20

10

-20

-10

-1010

-20x1 x2 a1(2) a2(2) a1(3)

0 0 0 1 10 1 0 0 01 0 0 0 01 1 1 0 0

Output10

Page 16: Koss 6 a17_deepmachinelearning_mariocho_r10

Multi-layer Neural Networks

Train this layer first

Page 17: Koss 6 a17_deepmachinelearning_mariocho_r10

Multi-layer Neural Networks

Train this layer firstthen this layer

then this layerthen this layer

finally this layer

Page 18: Koss 6 a17_deepmachinelearning_mariocho_r10

Deep learning - CNN

Page 19: Koss 6 a17_deepmachinelearning_mariocho_r10

Traditional learning vs Deep Machine Learning

Eiffel Tower

Eiffel Tower

RAW data

RAW data

Deep Learning Network

FeatureExtraction

Vectored Classification

Traditional Learning

Deep Learning

Page 20: Koss 6 a17_deepmachinelearning_mariocho_r10

Image recognition in Google Map

* Source: Oriol Vinyals – Research Scientist at Google Brain

Page 21: Koss 6 a17_deepmachinelearning_mariocho_r10

Deep Learning Hello World == MNIST

Page 22: Koss 6 a17_deepmachinelearning_mariocho_r10

MNIST (predict number of image)

Page 23: Koss 6 a17_deepmachinelearning_mariocho_r10

CNN (convolution neural network) training

Page 24: Koss 6 a17_deepmachinelearning_mariocho_r10

MNIST

Page 25: Koss 6 a17_deepmachinelearning_mariocho_r10

Hierarchical Representation of Deep Learning

* Source: : Honglak Lee and colleagues (2011) as published in “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks”.

Page 26: Koss 6 a17_deepmachinelearning_mariocho_r10

* Source: : Honglak Lee and colleagues (2011) as published in “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks”.

Deep Learning (object parts)

Page 27: Koss 6 a17_deepmachinelearning_mariocho_r10

Face extraction method

Page 28: Koss 6 a17_deepmachinelearning_mariocho_r10

Face recognition data- sets?

Page 29: Koss 6 a17_deepmachinelearning_mariocho_r10

Human-Level Face Recognition

• Convolutional neural networks based face recognition system is dominant

• 99.15% face verification accuracy on LFW dataset in DeepID2 (2014)� Beyond human-level recognition

Source: Taigman et al. DeepFace: Closing the Gap to Human-Level Performance in Face Verification, CVPR’14

Page 30: Koss 6 a17_deepmachinelearning_mariocho_r10

Image Recognition

* Source: Oriol Vinyals – Research Scientist at Google Brain

Page 31: Koss 6 a17_deepmachinelearning_mariocho_r10

Object Classification and Detection

Page 32: Koss 6 a17_deepmachinelearning_mariocho_r10

Scene Parsing

[Farabet et al. ICML 2012, PAMI 2013]

Page 33: Koss 6 a17_deepmachinelearning_mariocho_r10

How to the Object recognition ?

Page 34: Koss 6 a17_deepmachinelearning_mariocho_r10

ILSVRC (Image-net Large Scale Visual Recognition Challenge)

Page 35: Koss 6 a17_deepmachinelearning_mariocho_r10

Language Generating

* Source: Oriol Vinyals – Research Scientist at Google Brain

Page 36: Koss 6 a17_deepmachinelearning_mariocho_r10

Image Caption Generation

Page 37: Koss 6 a17_deepmachinelearning_mariocho_r10

Human-Level Object Recognition

• ImageNet• Large-Scale Visual Recognition Challenge�Image Classification / Localization�1.2M labeled images, 1000 classes�Convolutional Neural Networks (CNNs)has been dominating the contest since..� 2012 non-CNN: 26.2% (top-5 error)� 2012: (Hinton, AlexNet)15.3%� 2013: (Clarifai) 11.2%� 2014: (Google, GoogLeNet) 6.7%� 2015: (Google) 4.9%� Beyond human-level performance

Page 38: Koss 6 a17_deepmachinelearning_mariocho_r10

Scene Parsing

[Farabet et al. ICML 2012, PAMI 2013]

Page 39: Koss 6 a17_deepmachinelearning_mariocho_r10

Auto pilot car

Page 40: Koss 6 a17_deepmachinelearning_mariocho_r10

Tic Tac Toc

Page 41: Koss 6 a17_deepmachinelearning_mariocho_r10

AlphaGo

RAW data

Layer1Policy network & value network

Selection

Expansions

Evaluation

Backup

~ Layer13

Page 42: Koss 6 a17_deepmachinelearning_mariocho_r10

Automatic Colorization of Black and White Images

Page 43: Koss 6 a17_deepmachinelearning_mariocho_r10

Generate sounds on old-movies

Page 44: Koss 6 a17_deepmachinelearning_mariocho_r10

Automatic Machine Translation• Automatic Translation of Text.

• Automatic Translation of Images.

Page 45: Koss 6 a17_deepmachinelearning_mariocho_r10

Automatic Handwriting Generation

Page 46: Koss 6 a17_deepmachinelearning_mariocho_r10

Text Generation

Page 47: Koss 6 a17_deepmachinelearning_mariocho_r10

create stylized images from rough sketches.

Page 48: Koss 6 a17_deepmachinelearning_mariocho_r10

Inspirer Humanity

Page 49: Koss 6 a17_deepmachinelearning_mariocho_r10

Thanks you!

Q&A