AI/ML is a Means to Digital Transformation, Not an End Itself
1 like302 views
The document discusses the role of AI and machine learning in digital transformation, emphasizing that they are tools for achieving business value rather than ends in themselves. It highlights challenges in big data analytics, including data collection and processing issues, and presents use cases for product quality control and demand forecasting. Additionally, it illustrates how integrating solutions like cloud data lakes can enhance analytics efficiency and provide competitive advantages in the market.
![Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential
http://www.bespinglobal.com
16
To Do Product Quality Control with AI & ML using Data, Identify Main Steps
Assembling Process
Test results
Performance
Test Result
Performance
Issues & Improve
ment Reports
Claim by Customers
1. Line Claim
2. Field Claim
Parts data
(Specifications)
[Not Accessible]
Parts data (Specificati
ons)
Parts & Component Performance Test ClaimAssembling
Manufacturing
Sub parts
(Outsourcing)
Manafacturing
Main Parts
(In-house Processing)
1 2 3 4
Main Processes for Product Quality Control in Use Case 1](https://image.slidesharecdn.com/session04aimlbigdatasemianr-180921061204/85/AI-ML-is-a-Means-to-Digital-Transformation-Not-an-End-Itself-16-320.jpg)
![Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential
http://www.bespinglobal.com
25
Sentiment Detection from Amazon Product Review (API Example)
library("aws.comprehend")
for(i in 1:n) {
y <- detect_sentiment(REview_Store$d_content[i])[2]
y1 <- detect_sentiment(REview_Store$d_content[i])[3]
y2 <- detect_sentiment(REview_Store$d_content[i])[4]
y3 <- detect_sentiment(REview_Store$d_content[i])[5]
y4 <- detect_sentiment(REview_Store$d_content[i])[6]
w1[i] <- as.character(y[1,1])
w2[i] <- as.numeric(round(y1[1,1], 3))
w3[i] <- as.numeric(round(y2[1,1], 3))
w4[i] <- as.numeric(round(y3[1,1], 3))
w5[i] <- as.numeric(round(y4[1,1], 3))
}
Sentiment_detected Sentiment_score_Mixed Sentiment_score_Negative Sentiment_score_Neutral Sentiment_score_Positive d_content
MIXED 0.557 0.074 0.012 0.358 a great phone, can do more than I understand. Although a couple pof features n
POSITIVE 0.003 0.001 0.01 0.985 Terrific!
NEUTRAL 0.154 0.032 0.471 0.343 MetroPCS initially had issues getting it added to my account, but I woke up the n
POSITIVE 0.001 0 0 0.999
Love this phone
Its amazing
Have been an iPhone fan and user for long long time .This phone
NEGATIVE 0.02 0.966 0.008 0.006 I bought this phone in March and it eas delivered on March 20, and today the cam
POSITIVE 0.002 0 0.001 0.996 Whats to say, so far so good the phone have been working perfectly, its my little
POSITIVE 0.002 0 0.003 0.994 An excellent phone.
POSITIVE 0.001 0.001 0.012 0.986 I really like my unlocked S9 phone.
MIXED 0.78 0.098 0.003 0.119 I like the phone but this one died. The other 2 are working great. The exchange i
Amazon Comprehend](https://image.slidesharecdn.com/session04aimlbigdatasemianr-180921061204/85/AI-ML-is-a-Means-to-Digital-Transformation-Not-an-End-Itself-25-320.jpg)
AI/ML is a Means to Digital Transformation, Not an End Itself
- 2. AI/ML is a Means to Digital Transformation, Not an End Itself 김왕식 매니저
- 3. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 3 Big Data = Big Value $$$ Big DataIoT
- 4. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 4 Big Data = Big Value? 수집 저장 분석처리 INSIGHTS 데이터 수집 이슈 느린 처리 및 다양한 장애 ML에 대한 경험 부족
- 5. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 5 Big Data = Big Value? 수집 저장 분석처리 데이터 수집 이슈 느린 처리 및 다양한 장애 ML에 대한 경험 부족 INSIGHTS
- 6. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 6 Big Data = Big Value? 수집 저장 분석처리 데이터 수집 이슈 느린 처리 및 다양한 장애 ML에 대한 경험 부족 INSIGHTS
- 7. AGENDA 1. Adoption Challenges & Digital Transformation 2. Use Case 1: Product Quality Control 3. Use Case 2: Demand Forecasting 4. Summary & Next Step
- 9. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 9 Envisioning New World Source: AWS re:INVENT Digital Transformation (2017) Old World New World Traditional Business Model + Cloud ITTraditional Business Model
- 10. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 10 Future is Digital Business Digital Business Model + Cloud IT Source: AWS re:INVENT Digital Transformation (2017) Old World New World Traditional Business Model Maximum Value
- 11. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 11 Source: AWS Summit 2016 The Why, How and What of Digital Business Transformation in the Cloud Fundamental Principles in Digital Business The faster you get new ideas to market, The better you collect, analyze and predict data, The quicker you react to customer feedback,
- 12. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 12 Digital Transformation is Competitive Advantage Market Competition is War Digitalized Business Traditional BusinessVS
- 13. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 13 Key Components in Digital Business Model = Digital Business Big Data Analytics (AI/ML) Cloud Applications with Business Process Digital Transformation
- 14. Use Case 1: Product Quality Control
- 15. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 15 Global Use Cases in Manufacturing Industry IoT Predictive Maintenance Inventory Demand Forecasting (S&OP) Manufacturing Process Optimization Product Quality Control Raw Materials Optimal Price Optimal Timing Big Data Strategy I II III IV Data & Analytics Machine Learning Use Case 1 Process optimization Product quality control Anomaly detection Defect tracking Forecasting of outputs Simulation & testing … Use Case 2
- 16. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 16 To Do Product Quality Control with AI & ML using Data, Identify Main Steps Assembling Process Test results Performance Test Result Performance Issues & Improve ment Reports Claim by Customers 1. Line Claim 2. Field Claim Parts data (Specifications) [Not Accessible] Parts data (Specificati ons) Parts & Component Performance Test ClaimAssembling Manufacturing Sub parts (Outsourcing) Manafacturing Main Parts (In-house Processing) 1 2 3 4 Main Processes for Product Quality Control in Use Case 1
- 17. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 17 Before Assessing As-Is Analytics, Understand Analytic Lifecycle Latency (Time to Answear) Ingest Collect & ingest data from multiple sources Store Store data in a format that is easy to be changeable and be accessed Process & Analyze Transform data from raw form into actionable form Consume & Visualize Consume & deploy the insights (ML/AI) into (application) production environment to maximize business value
- 18. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 18 Assessment Outcome: Why Data Lake? Data Lake Today ERP data are captured into their own processes: Fragmented Information & Analytics Lifecycle Tomorrow Individual Processes feed into a single data lake: cross-referenced insights Spreadsheets Spreadsheets Spreadsheets Spreadsheets Transition From Fragmented Data Sources to Data Lake
- 19. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 19 Data Transformation: 가공 테스트 결과 값 변화 이후 각 수치에 대한 분포 확인 Histogram of 가공 테스트 데이터 변환 및 가공을 통해서 미세한 차이도 감지 (within 0.10mm) Normalization Index: 최대값과 최소값을 이용 (within group among 17 different tests) Easy & Efficient Workflow Management with Data Lake
- 20. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 20 Data tell you a Story? 가공 테스트 결과와 조립 결과의 관계 그래프 조립 테스트 (Normalization Index) Normalization Index: 최대값과 최소값을 이용 가공 테스트 (Normalization Index) Pre-defined specifications on parts and components still create systematic variations for outcome, which leads to differences in product quality
- 21. Use Case 2: Demand Forecasting
- 22. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 22 수요예측 분석 배경 급변하는 시장 트렌드와 글로벌화 된 시장 경쟁에 선제적으로 대응하기 위한 핵심 정보로서의 수요예측 및 빅데이터 빠르고 의미 있는 Business Insight 제공을 위한 Cloud 기반 Big Data 분석 과제 BIG DATA Analytics
- 23. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 23 수요예측 분석 절차 Data Scientist에게 최적화 된 AWS 머신러닝 서비스를 활용하여 분석을 효과적으로 진행 1 2 3 4 빅데이터 전략 / 수집 Data Science 모델 개발 최적 모델 활용 계획요구 사항 / 현업 미팅 데이터 수집 및 분석은 사전 준비, 데이터 수집, 데이터 가공, 분석 단계로 진행 ML 구현 플랫폼 AWS SageMaker 빠른 학습 속도, 대용량 데이터에 최적화된 SageMaker 데이터셋 스트리밍을 통한 저렴한 학습 비용 싱글 패스로 빠른 학습 아주 큰 데이터셋에 대한 학습이 가능 다양한 문제에 대한 빌트인 ML Algorithm 제공 AWS AI API 서비스 다양한 종류의 AI 서비스 제공
- 24. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 24 Amazon Comprehend
- 25. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 25 Sentiment Detection from Amazon Product Review (API Example) library("aws.comprehend") for(i in 1:n) { y <- detect_sentiment(REview_Store$d_content[i])[2] y1 <- detect_sentiment(REview_Store$d_content[i])[3] y2 <- detect_sentiment(REview_Store$d_content[i])[4] y3 <- detect_sentiment(REview_Store$d_content[i])[5] y4 <- detect_sentiment(REview_Store$d_content[i])[6] w1[i] <- as.character(y[1,1]) w2[i] <- as.numeric(round(y1[1,1], 3)) w3[i] <- as.numeric(round(y2[1,1], 3)) w4[i] <- as.numeric(round(y3[1,1], 3)) w5[i] <- as.numeric(round(y4[1,1], 3)) } Sentiment_detected Sentiment_score_Mixed Sentiment_score_Negative Sentiment_score_Neutral Sentiment_score_Positive d_content MIXED 0.557 0.074 0.012 0.358 a great phone, can do more than I understand. Although a couple pof features n POSITIVE 0.003 0.001 0.01 0.985 Terrific! NEUTRAL 0.154 0.032 0.471 0.343 MetroPCS initially had issues getting it added to my account, but I woke up the n POSITIVE 0.001 0 0 0.999 Love this phone Its amazing Have been an iPhone fan and user for long long time .This phone NEGATIVE 0.02 0.966 0.008 0.006 I bought this phone in March and it eas delivered on March 20, and today the cam POSITIVE 0.002 0 0.001 0.996 Whats to say, so far so good the phone have been working perfectly, its my little POSITIVE 0.002 0 0.003 0.994 An excellent phone. POSITIVE 0.001 0.001 0.012 0.986 I really like my unlocked S9 phone. MIXED 0.78 0.098 0.003 0.119 I like the phone but this one died. The other 2 are working great. The exchange i Amazon Comprehend
- 26. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 26 AWS Architecture DB Data Source Weekly Batch Sentiment Values from Reviews Positive | Negative | Neutral | Mixed Collect Store Process Consume Answers Featurization Model Inference & Visualization Storage Search Term Result 1. 빠르고 정확한 감정분석 서비스 2. 완전 관리형 머신러닝 플랫폼 SageMaker AWS 클라우드 및 머신 러닝 분석 플랫폼의 지속적 혁신을 활용 사용 된 핵심 서비스
- 27. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 27 • 감정분석 스코어 총량(Amazon Product Review & Best Buy)과 핸드폰 판매량의 관계 감정분석 종합 그래프 시간 축 (월 단위) 0 10월8월6월4월 6월4월2월12월 8월 12월10월 2월 4월 6월 2016 2017 2018 8월 Normalization Index: 최대값과 최소값을 이용 Summary: #1 베스트 바이 보다는 아마존이 출시 이후의 최대 판매량 예측의 선행지표로서 역할 가능 #2 각각의 스토어 마다 다른 유형의 소비자가 존재 #3 시간에 따른 변화의 모습이 상이 함 유저가 남긴 글을 수치화 => 월간 감정의 총량 계산 => 기간 별 산출된 값들의 최대값/최소값을 변환 => 시각화 핸드폰 판매량
- 28. Next Step
- 29. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 29 Challenges identified • Big Data = Big Value? Data Doesn’t Connect Limited ETL Processes Difficult to Scale Difficult to run Experiment Lacks Advanced Analytics Low Utilization Difficult to Collect Unstructured & New Types of Data Capture & Ingest Store Analyze & VisualizeProcess INSIGHTS Ingestion Problem Time to get Insight Limited Analytics Data
- 30. Copyright © 2018 BESPIN GLOBAL Co., Ltd. All rights reserved | Confidential http://www.bespinglobal.com 30 Analytic Lifecycle is Accelerated in (Cloud) Data Lake ERP Insight Transaction CRM Insight Insight + Big Data as an Asset Public Data + Social Data + Global Trend
- 31. THANK YOU.