人工智慧產業發展趨勢 - tri.org.t · Telefonica, BigML AI selects startups Deep Knowledge Ventures, AI votes on investments GE survival on software and AI Baidu, …

人工智慧產業發展趨勢蔣以仁

臺北醫學大學大數據科技及管理研究所 教授/所長

學士後大數據科技及管理學程 主任

何謂人工智慧︖運用科技所製造之智慧機器，尤其是智慧電腦程式。讓機器具備智慧能做人類能做的事。（JohnMcCarthy,1956）

嘗試模擬及重複人類思行緒為：思考、說話、感應、與推論

人工智慧進化

3

困境精確計算的不可行

Gödel‘s不完備定理

TuringTest

4

AI復甦DENDRAL– Feigenbaum

IntelligentAgent◦SearchEngine◦Recommendation◦BusinessIntelligence

5

6

人工智慧衝擊

經濟, 文化, 社會, … 無盡的崩解

勞工 - McKinsey 58%工作自動化

Martin Ford, Rise of the Robots

Elon Musk, artificial

intelligence... 潛藏的威脅

AI 驅動 Unprecedented Era

7

超越時空的壓縮新的破壞創新

極致收斂於不同的領域

全方位的連結擴充

指數加速的自動化– smart sensors and the 26 billion IoT devices by 2020

(11 trillion USD by 2025)

AoE: Evidence

8

Singapore self-driving Taxis September 2016

Norwegian Telenor AI and Big Data Lab

Telefonica, BigML AI selects startups

Deep Knowledge Ventures, AI votes on investments

GE survival on software and AI Baidu, AskADoctor, 520 diseases, refers specialists

Baidu, StockMasterpredicts market trends Controversy: AI bias

PredixCloud

今日工廠

明日工廠

商品的翻轉

HarvardBusinessReview

Big data —from the lab to the clinic and back

•Calliope A. Dendrou, Gil McVean & Lars Fugger, Neuroinflammation — using big data to inform clinicalpractice,NatureReviewsNeurology 12, 685–698 (2016)

12

Industrial Damage DetectionIndustrialdamagedetectionreferstodetectionofdifferentfaultsandfailuresincomplexindustrialsystems,structuraldamages,intrusionsinelectronicsecuritysystems,suspiciouseventsinvideosurveillance,abnormalenergyconsumption,etc.◦ Example:AircraftSafety

◦ AnomalousAircraft(Engine)/FleetUsage◦ Anomaliesinenginecombustiondata◦ Totalaircrafthealthandusagemanagement

KeyChallenges◦ Dataisextremelyhuge,noisyandunlabelled◦ Mostofapplicationsexhibittemporalbehaviour◦ Detectinganomalouseventstypicallyrequireimmediateintervention

13

異常偵測Detectingoutliersinaimagemonitoredovertime

Detectinganomalousregionswithinanimage

Usedin◦ mammographyimageanalysis◦ videosurveillance◦ satelliteimageanalysis

KeyChallenges◦ Detectingcollectiveanomalies◦ Datasetsareverylarge

Anomaly

50 100 150 200 250 300 350

50

100

150

200

250

14

異常偵測• N1 andN2 areregionsofnormalbehavior

• Pointso1 ando2 areanomalies

• PointsinregionO3areanomalies

X

Y

N1

N2

o1

o2

O3

15

重要挑戰• Definingarepresentativenormalregionischallenging• Theboundarybetweennormalandoutlyingbehavior isoftennotprecise

• Theexactnotionofanoutlierisdifferentfordifferentapplicationdomains

• Availabilityoflabeled datafortraining/validation• Maliciousadversaries• Datamightcontainnoise• Normalbehavior keepsevolving

16

AspectsofAnomalyDetectionProblem• Natureofinputdata• Availabilityofsupervision• Typeofanomaly:point,contextual,structural• Outputofanomalydetection• Evaluationofanomalydetectiontechniques

17

PointAnomalies• Anindividualdatainstanceisanomalousifitdeviatessignificantlyfromtherestofthedataset.

X

Y

N1

N2

o1

o2

O3

Anomaly

18

ContextualAnomaliesAnindividualdatainstanceisanomalouswithinacontextRequiresanotionofcontextAlsoreferredtoasconditionalanomalies

*Xiuyao Song,Mingxi Wu,ChristopherJermaine,SanjayRanka,ConditionalAnomalyDetection,IEEETransactionsonDataandKnowledgeEngineering,2006.

NormalAnomaly

19

CollectiveAnomaliesAcollectionofrelateddatainstancesisanomalousRequiresarelationshipamongdatainstances◦ SequentialData◦ SpatialData◦ GraphData

Theindividualinstanceswithinacollectiveanomalyarenotanomalousbythemselves

AnomalousSubsequence

20

類型Anomaly Detection

Contextual Anomaly Detection

Collective Anomaly Detection

Online Anomaly Detection

Distributed Anomaly Detection

Point Anomaly Detection

Classification BasedRule BasedNeural Networks BasedSVM Based

Nearest Neighbor BasedDensity BasedDistance Based

StatisticalParametricNon-parametric

Clustering Based OthersInformation Theory BasedSpectral Decomposition BasedVisualization Based

*AnomalyDetection– ASurvey,VarunChandola,Arindam Banerjee,andVipin Kumar,ACMComputingSurvey,2019

21

參考資料• AIassessesbreastcancerrisk30timesfaster

http://www.forbes.com/sites/janetwburns/2016/08/29/artificial-intelligence-can-help-doctors-assess-breast-cancer-risk-thirty-times-faster/#7b717af556e2

• GE,rebornasasoftwarestartupusingAIhttp://www.nytimes.com/2016/08/28/technology/ge-the-124-year-old-software-start-up.html?_r=0

• Worldleading2025ChinaAIindustryhttp://www.chinadaily.com.cn/business/tech/2016-08/27/content_26615174.htm

• GlobalAIMarket2015:127B;2016:165B;2018:200B• Audrey--NASA's New Self-Learning AI Could Save First

Respondershttp://motherboard.vice.com/read/this-nasa-ai-will-sense-danger-save-firefighters-and-learn-from-mistakes

• Voicerecognition3xfasterthantypinghttp://www.npr.org/sections/alltechconsidered/2016/08/24/491156218/voice-recognition-software-finally-beats-humans-at-typing-study-finds?utm_medium=RSS&utm_campaign=storiesfromnpr

參考資料• The world's first self-driving taxis will be picking up passengers

in Singapore in September 1http://www.cbc.ca/news/technology/driverless-taxi-nutonomy-1.3735375

• AI bias http://motherboard.vice.com/read/its-our-fault-that-ai-thinks-white-names-are-more-pleasant-than-black-names

• NorwegianTelcocreatesAIandBigDatalabhttps://www.telecomtvtracker.com/insights/telenor-supports-norwegian-entrepreneurship-and-artificial-intelligence-research-6448/

• TelefonicaandBigML usingAItoselectstartupshttps://www.telefonica.com/es/web/press-office/-/telefonica-open-future_-and-bigml-create-preseries-a-joint-venture-for-early-stage-investment

• DeepKnowledgeVenturesappointsAIlikeaboardmembertomakeinvestmentdecisionhttp://www.itbusiness.ca/blog/hong-kong-vc-firm-appoints-ai-to-board-of-directors/48815

• Satelliteimagesandmachinelearningcanmappovertyhttp://bit.ly/2bxEv3w

謝謝聆聽!QUESTIONSANDCOMMENTS

24