_KrlGF발표자료_AI

Similar documents
KAKAO AI REPORT Vol.01

Ch 1 머신러닝 개요.pptx

<4D F736F F D20B1E2C8B9BDC3B8AEC1EE2DC0E5C7F5>

김기남_ATDC2016_160620_[키노트].key

<4D F736F F D20C3D6BDC C0CCBDB4202D20BAB9BBE7BABB>

2 min 응용 말하기 01 I set my alarm for It goes off. 03 It doesn t go off. 04 I sleep in. 05 I make my bed. 06 I brush my teeth. 07 I take a shower.

Stage 2 First Phonics

<4D F736F F D20C3D6BDC C0CCBDB4202D20BAB9BBE7BABB>

DIY 챗봇 - LangCon

(JBE Vol. 23, No. 2, March 2018) (Special Paper) 23 2, (JBE Vol. 23, No. 2, March 2018) ISSN

2 : (Seungsoo Lee et al.: Generating a Reflectance Image from a Low-Light Image Using Convolutional Neural Network) (Regular Paper) 24 4, (JBE

OP_Journalism

Hi-MO 애프터케어 시스템 편 5. 오비맥주 카스 카스 후레쉬 테이블 맥주는 천연식품이다 편 처음 스타일 그대로, 부탁 케어~ Hi-MO 애프터케어 시스템 지속적인 모발 관리로 끝까지 스타일이 유지되도록 독보적이다! 근데 그거 아세요? 맥주도 인공첨가물이

본문01

Data Industry White Paper

PowerPoint 프레젠테이션

레이아웃 1

김경재 안현철 지능정보연구제 17 권제 4 호 2011 년 12 월

DocsPin_Korean.pages

Macaron Cooker Manual 1.0.key

기획 1 서울공대생에게 물었다 글 재료공학부 1, 이윤구 재료공학부 1, 김유리 전기정보공학부 1, 전세환 편집 재료공학부 3, 오수봉 이번 서울공대생에게 물었다! 코너는 특별히 설문조사 형식으로 진행해 보려고 해 요. 설문조사에는 서울대학교 공대 재학생 121명, 비

지난 10월 6일과 12일

APOGEE Insight_KR_Base_3P11

Connection 8 22 UniSQLConnection / / 9 3 UniSQL OID SET

1

도비라

삼성SDI_SR국문_최종

Software Requirrment Analysis를 위한 정보 검색 기술의 응용

VZ94-한글매뉴얼

4. #include <stdio.h> #include <stdlib.h> int main() { functiona(); } void functiona() { printf("hihi\n"); } warning: conflicting types for functiona

±èÇö¿í Ãâ·Â

Á¶´öÈñ_0304_final.hwp

(JBE Vol. 23, No. 2, March 2018) (Special Paper) 23 2, (JBE Vol. 23, No. 2, March 2018) ISSN

4 : (Hyo-Jin Cho et al.: Audio High-Band Coding based on Autoencoder with Side Information) (Special Paper) 24 3, (JBE Vol. 24, No. 3, May 2019

Page 2 of 5 아니다 means to not be, and is therefore the opposite of 이다. While English simply turns words like to be or to exist negative by adding not,

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Jul.; 29(7),

<5B4E DBDBAB8B6C6AEC4DCC5D9C3F720BFF9B0A3B5BFC7E2BAB8B0ED5F35C8A32838BFF9292E687770>


정보기술응용학회 발표

untitled

(JBE Vol. 24, No. 2, March 2019) (Special Paper) 24 2, (JBE Vol. 24, No. 2, March 2019) ISSN

ePapyrus PDF Document

Microsoft PowerPoint - ch03ysk2012.ppt [호환 모드]

강의10

AT_GraduateProgram.key

강의지침서 작성 양식

Microsoft Word - Westpac Korean Handouts.doc

DBPIA-NURIMEDIA

example code are examined in this stage The low pressure pressurizer reactor trip module of the Plant Protection System was programmed as subject for

04_오픈지엘API.key

Page 2 of 6 Here are the rules for conjugating Whether (or not) and If when using a Descriptive Verb. The only difference here from Action Verbs is wh

지능정보연구제 16 권제 1 호 2010 년 3 월 (pp.71~92),.,.,., Support Vector Machines,,., KOSPI200.,. * 지능정보연구제 16 권제 1 호 2010 년 3 월

Microsoft PowerPoint - 실습소개와 AI_ML_DL_배포용.pptx

3 Gas Champion : MBB : IBM BCS PO : 2 BBc : : /45

RNN & NLP Application

Intra_DW_Ch4.PDF

[한반도]한국의 ICT 현주소(송부)

Social Network

Orcad Capture 9.x

IPAK 윤리강령 나는 _ 한국IT전문가협회 회원으로서 긍지와 보람을 느끼며 정보시스템 활용하 자. 나는 _동료, 단체 및 국가 나아가 인류사회에 대하여 철저한 책임 의식을 가진 다. 나는 _ 활용자에 대하여 그 편익을 증진시키는데 최선을 다한다. 나는 _ 동료에 대해

한국외국어대학교 세계와 만나는 가장 빠른 길 한국외대는 진리( 眞 理 ), 평화( 平 和 ), 창조( 創 造 )의 창학 정신을 바탕으로 국가와 세계 발전에 기여할 수 있는 잠재력을 지닌 인재를 선발하고자 노력하고 있습니다. 자주적 탐구인 합리적 사고 폭넓은 지식 정심대

Slide 1

퇴좈저널36호-4차-T.ps, page Preflight (2)

<C0CCBCBCBFB52DC1A4B4EBBFF82DBCAEBBE7B3EDB9AE2D D382E687770>

4 CD Construct Special Model VI 2 nd Order Model VI 2 Note: Hands-on 1, 2 RC 1 RLC mass-spring-damper 2 2 ζ ω n (rad/sec) 2 ( ζ < 1), 1 (ζ = 1), ( ) 1

사회통계포럼

08SW

1_2•• pdf(••••).pdf

Something that can be seen, touched or otherwise sensed

슬라이드 1

PowerPoint 프레젠테이션

야쿠르트2010 3월 - 최종

°í¼®ÁÖ Ãâ·Â

방송공학회논문지 제18권 제2호

6주차.key

2016년 트렌드 책목차를 활용한 시장 예측.numbers

SchoolNet튜토리얼.PDF

PowerPoint 프레젠테이션

서론 34 2

May 2014 BROWN Education Webzine vol.3 감사합니다. 그리고 고맙습니다. 목차 From Editor 당신에게 소중한 사람은 누구인가요? Guidance 우리 아이 좋은 점 칭찬하기 고맙다고 말해주세요 Homeschool [TIP] Famil

¹Ìµå¹Ì3Â÷Àμâ

PCServerMgmt7

ePapyrus PDF Document

가장 낮은 증가세를 기록했다. 분기당 400만 명을 유치하긴 했지만, 경쟁업체로 부상하고 있는 인스타그램(Instagram)은 분기당 500만 명을 유치했다. 분명히 트위터는 어려운 시기를 겪고 있고, 기업가의 입장에서는 이를 극복할 수 있는 방안 모색이 필요한 시점이

소식지도 나름대로 정체성을 가지게 되는 시점이 된 거 같네요. 마흔 여덟번이나 계속된 회사 소식지를 가까이 하면서 소통의 좋은 점을 배우기도 했고 해상직원들의 소탈하고 소박한 목소리에 세속에 찌든 내 몸과 마음을 씻기도 했습니다. 참 고마운 일이지요 사람과 마찬가지로


PowerPoint Presentation

DW 개요.PDF

Vol.259 C O N T E N T S M O N T H L Y P U B L I C F I N A N C E F O R U M


Microsoft PowerPoint - 3.공영DBM_최동욱_본부장-중소기업의_실용주의_CRM

Microsoft Word - 조병호

pdf 16..

클라우드컴퓨팅확산에따른국내경제시사점 클라우드컴퓨팅확산에따른국내경제시사점 * 1) IT,,,, Salesforce.com SaaS (, ), PaaS ( ), IaaS (, IT ), IT, SW ICT, ICT IT ICT,, ICT, *, (TEL)

融合先验信息到三维重建 组会报 告[2]

ETL_project_best_practice1.ppt

광운소식-68호F

I care - Do you?

1

VOL /2 Technical SmartPlant Materials - Document Management SmartPlant Materials에서 기본적인 Document를 관리하고자 할 때 필요한 세팅, 파일 업로드 방법 그리고 Path Type인 Ph

Transcription:

AI 의과거와현재그리고내일 AI is the New Electricity 2017.09.15

AI! 2

Near Future of Super Intelligence? *source l http://www.motherjones.com/media/2013/05/robots-artificial-intelligence-jobs-automation 3

4

I think that we live in a world where just as electricity transformed almost everything almost 100 years ago, Today I actually have a hard time thinking of an industry that I don't think AI will transform in the next several years 100 AI. 5

Bill Gates to college grads I expect AI to create breakthroughs that makes people better learners. Get a job in AI, but don t forget inequity around you. AI. AI,. 6

AI is a technology that gets so close to everything we care about. It s going to carry the values that matter to our lives, be it the ethics, the bias, the justice, or the access. AI.,,,,. 7

( Positive & Negative ) vs, vs vs Fake/Garbage vs 24hr Everything 8

1. AI 2. AI? - 3. State - of - Arts 4.

AI Breakthrough *source l http://www.slideshare.net/luma921/deep-learning-the-past-present-and-future-of-artificial-intelligence 10

AI *source l https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligence-machine-learning-deep-learning-ai/ 11

Level - 1. If ~ then ~ 12

Level - 2. "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E. by Tom M. Mitchell Classification Regression Clustering *source l https://www.slideshare.net/terrytaewoongum/machine-learning-54531674 13

Level - 3. Deep Neural Networks (DNN) DNN, AI *source l http://www.cs.toronto.edu/~ranzato/files/ranzato_cnn_stanford2015.pdf 14

AI 1997 2011 2016 Next 15

1. AI 2. AI? - 3. State - of - Arts 4.

AI AI Answer : it's a girl is brushing her hair. *source l https://www.captionbot.ai/ (MS ) 17

?? 18

? 19

AI Option Pricing. AI. 20

? 21

? Darpa Robot Challenge 2015 22

AI AI 23

AI? AI 24

AI = ( )?? [ ] [ ] [ ]. 25

F 26

(Machine Learning)? (T, Task) (P, Performance measure) (E, Experience). 27

? AI,! 28

? ( ) (Supervised Learning) >> >> SL >> ( Google -> Kaggle, DR ) (Unsupervised Learning) >> >> ( : ) >> ) Clustering (Reinforcement Learning) >> >> >> ) 29

-? *Video l 1m40s 30

(Machine Learning) (Artificial Neural Network) (Deep Learning) 31

AI Neural Information Processing Systems International Conference on Machine Learning 2012 AI breakthrough 2012 ImageNet 32

AI - Y. LeCun IBM, Google, MS G. Hinton Google Y. Bengio Facebook Andrew Ng Baidu (ex) "1990 2000 (Neural Network)... by Yoshua Bengio 33

AI - (ICML, NIPS) 34

AI? - 35

(ANN) (Neuron) (Synapse) 36

ANN 28 x 28 = 784 0~9 = 10 Labels Output 0 0.1 1 0.05 2 0.3 3 0.2 28 x 28 0 0 0 1 1 1 0 0 1 1 1 1 0 1 1 0 0 1 1 1 0 0 0 1 1 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 4 0.41 5 0.26 6 0.11 7 0.84 8 0.17 9 0.45 37

ANN Output Labels Input 0.1 0 0 0.05 1 0 0.3 2 0 28 x 28 = 784 0.2 0.41 3 0 4 0 0.26 5 0 0.11 6 0 0.84 7 1 0.17 8 0 RGB 28 x 28 x 3 = 2,352 0.45 9 0 Label Output Input Weight 38

ANN

Train : 94.6%, Test : 87.7% 40

ANN ANN? 28 x 28 = 784? Input Data Parameter Linear Combination Output Parameter 41

DNN (Deep Neural Network)? Hidden Layer 2 Deep Layer 2012 SuperVision 65 6000, 6 3000 42

DNN? Linear Equation Combination Output Parameter 43

Deep Learning (1) - Overfitting Data Over Fitting Data Fitting 44

Deep Learning (2) - Vanishing Gradient Signal 45

Deep Learning (3) - Local Minimum 46

Deep? Not Deep ImageNet Label Layer Deep,! 47

Deep? Performance Amount of data *source l Andrew Ng Deep Learning Lecture Note 48

DNN (Deep Neural Network)? 2012? Nvidia GPU Tesla 10 100 (miles) Computer Processing Power GPU *source l http://qz.com/694520/tesla-has-780-million-miles-of-driving-data-and-adds-another-million-every-10-hours/ (Tesla ) 49

Nvidia GPU Computing Power Nvidia - 11 50

Google Cloud Machine Learning 1,000 GPU 25 51

Google I/O 17 Chip ( TPU ) 52

Data? Critical Mass Deep Learning 2004 Caltech 101 10K Images 2005-2010 Pascal VOC 2K 30K objects 2010-2015 Image Net 10M 15M images Image source: http://www.vision.caltech.edu/ Image source: http://doi.ieeecomputersociety.org/ http://www.image-net.org/ 53

Data? A B C Deep Learning Neural Networks Log. Regression Deep learning K-Nearest neighbors Support vector machines Boosting Artificial neural networks Bayesian networks Sparse dictionary learning Regression forest 54

1. AI 2. AI? - 3. State - of - Arts 4.

Quality Data Labeled Data ( Supervised Learning ) 56

Input Output Application Home Features Ad, User Info Image Audio English Image, Radar, Sign Price Click Ad? (0/1) Object( cat, dog) Text trasnscipt Chiness People, Car Real Estate Online Ad Photo Tagging (name) Speech Recognition Translation Autonomous Car 57

TTS AI *source l http://campaign.happybean.naver.com/yooinna_audiobook 58

DeepMind WaveNet Google DeepMind TTS 1/1000 TTS WaveNet *source l https://deepmind.com/blog/wavenet-generative-model-raw-audio/ 59

Music Generation Daddy s Car, a pop song in the style of The Beatles. *source l http://www.theverge.com/2016/9/26/13055938/ai-pop-song-daddys-car-sony *Video l 3m00s 60

Recognizing Pain ARTIFICIAL INTELLIGENCE COULD END ANIMAL SUFFERING BY RECOGNIZING PAIN *source l http://www.newsweek.com/artificial-intelligence-sheep-pain-emotion-cambridge-618085 61

사진으로 나이 알아 내기 *source l https://how-old.net/ (by MS제공 ) 62

*source l https://github.com/alexjc/neural-doodle 63

이미지 설명하기 *source l http://cs.stanford.edu/people/karpathy/sfmltalk.pdf 64

Breast Cancer Detection Google uses machine learning to detect breast cancer better than pathologists *source l http://siliconangle.com/blog/2017/03/05/google-uses-machine-learning-better-detect-breast-cancer-pathologists/ 65

Image Generation (1) [ ] AI [ ] [ ] AI [ ]? 66

Image Generation (2) - GAN GAN(Generative Adversarial Network) g-net : d-net : *source l NIPS 2016. 67

Image Generation - 16 x 16 64 x 64 (x4) Augment *source l https://github.com/david-gpu/srez 68

1. AI 2. AI? - 3. State - of - Arts 4.

GAFA (Google, Amazon, Facebook, Apple) AI 70

AI 71

AI in 2017 Tech giants acquired 34 AI startups in Q1 2017 AI 17 1 Amazon $ 19M buy harvest.ai (AI-security) 17 2 Ford $1B buy Argo ( ) 17 5 Apple $200M buy Lattice Data ( ) 17 5 Cisco $125M buy Mindmeld ( ) > AI *source l https://venturebeat.com/2017/05/28/tech-giants-acquired-34-ai-startups-in-q1-2017/ 72

AI 73

AI? 1 AI AI 3 Domain API, 2 Data Garbage In -> Garbage Out Big Data 4 Computing Power Trial & Error

AI 1. ( SL ) 2. ( ) AI. 75

Tesla Real Time - Data Processing 76

AI 2017 3. https://brunch.co.kr/@kakao-it/ KAKAO AI REPORT import kakao.ai.dataset.daisy import kakao.ai.image import kakao.ai.classifier import mxnet as mx def Conv(data, num_filter, kernel=(1, 1), stride=(1, 1), pad=(0, 0), name=none, suffix=''): conv = mx.sym.convolution(data=data, num_filter=num_filter, kernel=kernel, stride=stride, pad=pad, no_bias=true, name='%s%s_conv2d' %(name, suffix)) bn = mx.sym.batchnorm(data=conv, name='%s%s_batchnorm' %(name, suffix), fix_gamma=true) act = mx.sym.activation(data=bn, act_type='relu', name='%s%s_relu' %(name, suffix)) return act def Inception7A(data, num_1x1, num_3x3_red, num_3x3_1, num_3x3_2, num_5x5_red, num_5x5, pool, proj, name): tower_1x1 = Conv(data, num_1x1, name=('%s_conv' % name)) tower_5x5 = Conv(data, num_5x5_red, name=('%s_tower' % name), suffix='_conv') tower_5x5 = Conv(tower_5x5, num_5x5, kernel=(5, 5), pad=(2, 2), name=('%s_tower' % name), suffix='_conv_1') tower_3x3 = Conv(data, num_3x3_red, name=('%s_tower_1' % name), suffix='_conv') tower_3x3 = Conv(tower_3x3, num_3x3_1, kernel=(3, 3), pad=(1, 1), name=('%s_tower_1' % name), suffix='_conv_1') tower_3x3 = Conv(tower_3x3, num_3x3_2, kernel=(3, 3), pad=(1, 1), name=('%s_tower_1' % name), suffix='_conv_2') pooling = mx.sym.pooling(data=data, kernel=(3, 3), stride=(1, 1), pad=(1, 1), pool_type=pool, name=('%s_pool_%s_pool' % (pool, name))) cproj = Conv(pooling, proj, name=('%s_tower_2' % name), suffix='_conv') concat = mx.sym.concat(*[tower_1x1, tower_5x5, tower_3x3, cproj], name='ch_concat_%s_chconcat' % name) return concat def Inception7B(data, num_3x3, num_d3x3_red, num_d3x3_1, num_d3x3_2, pool, name): tower_3x3 = Conv(data, num_3x3, kernel=(3, 3), pad=(0, 0), stride=(2, 2), name=('%s_conv' % name)) tower_d3x3 = Conv(data, num_d3x3_red, name=('%s_tower' % name), suffix='_conv') tower_d3x3 = Conv(tower_d3x3, num_d3x3_1, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name=('%s_tower' % name), suffix='_conv_1') Vol.01 2017.03 AI CODE int FaceAlignment::track(unsigned char *src, int width, int height, int rotation) { m_img.img.wrap(height, width, src); m_img.gt_enabled = false; for (int i = 0; i < m_face_cnt; i++) { } if (m_data[i].enabled == false) // pass if it is not valid continue; // connect image to data DataInfo *data = &m_data[i]; data->img_ptr = &m_img; // compute bounding box of current face MyRect<float> bbox = FAUtil::compute_bbox(data->current_pts); data->fd = bbox; // put mean shape to the face in image data->current_norm_pts = mean_shape; data->q = FAUtil::compute_similarity(data->current_norm_pts, data->current_pts); data->inv_q = FAUtil::compute_inverse_similarity(data->q); Vol.02 2017.04 data->current_pts = FAUtil::apply_similarity_transform(data->current_norm_pts, data->q); data->initial_pts = data->current_pts; // DO align align(data); AI CODE vector<pairif_t> c_dmf_topk( ) { Map<const VectorXf> vv(v, M), AA(A, N); Map<const MatrixXf> XX(X, N, M); vector<pair<int, float>> result; RowVectorXf dot = hidden_layers[0].transpose() * vv; for(int j=1; j < (int)hidden_layers.size(); ++j) dot = dot * hidden_layers[j]; dot = dot.array().max(0.0f); VectorXf ex = fast_softmax(xx, dot, AA); get_fast_knn(ex, k, result); return result; } vector<pairsf_t> prediction_by_vector(const float* v, int k) { auto ret = pool_.enqueue([](const float* _v, vector<factortype>& hidden_layers, const float* sfxb, const float* sfx, int N, int M, int _k){ return c_dmf_topk(_v, hidden_layers, sfxb, sfx, N, M, _k); }, v, hidden_layers_, (const float*)softmax.data(), k); auto val = move(ret.get()); vector<pairsf_t> result = as_result(val); return result; } Vol.03 2017.05 tower_d3x3 = Conv(tower_d3x3, num_d3x3_2, kernel=(3, 3), pad=(0, 0), stride=(2, 2), name=('%s_tower' % name), suffix='_conv_2') pooling = mx.symbol.pooling(data=data, kernel=(3, 3), stride=(2, 2), pad=(0,0), pool_type="max", name=('max_pool_%s_pool' % name)) concat = mx.sym.concat(*[tower_3x3, tower_d3x3, pooling], name='ch_concat_%s_chconcat' % name) return concat } // validate aligned face return validate(src, width, height, rotation); 77

:, noah.jung@kakaocorp.com Question

End of Document