Large-Scale Machine Learning at Verizon. Ashok N. Srivastava, Ph.D. Chief Data Scientist, Verizon

Transcription

1 Large-Scale Machine Learning at Verizon Ashok N. Srivastava, Ph.D. Chief Data Scientist, Verizon

2 A Transition in Roles Enormous data volumes Massive computing infrastructure Significant public benefit that touches daily lives

3 Organizations started to recognize they are operating with blind spots Factors supporting major decisions 1 in 3 To a little extent business leaders frequently make critical decisions without the information they need 79 % 62 % 53% 52 % don t have access to the information across their organization needed to do their jobs To a great extent Personal Experience Analytics Collective Experience

4 Mobile in the United States Consumers are Mobile Social Already Mobile Data Traffic Exploding Smartphone penetration 61% % CAGR 80% % Smartphone consumers using social 40%-45% growth projected per year will 74% Game will watch 45% Videos 58% of social media time 2,000 Terabytes per day 1 Terabyte (TB) = ~ 1,000 GB 21% will listen to 41% Music of 2013 Black Friday ecommerce 4 IN 10 social users bought after sharing on 200 TB metadata/day What site? What page? Last site? Next site? From where? What time? What app? From who? Sources: emarketer, Jumptap & comscore, Vision Critical, Verizon internal data

5 Revenue Streams due to Analytics 5

6 The Orion Cluster at Verizon VZ Management Center Vertical Solutions Web Services & APIs Advertising Managed Network Services Cybersecurity (unified network) Network Health Management M2M Reporting and Advanced Analytics BI Reporting & Dashboards Domain-Specific Rules Engine Anomaly & Pattern Detection Prediction Algorithms Recommendation Engine Insight Discovery Real-time & Batch Processing Big Data Infrastructure Hadoop Ecosystem Massive Parallel Processing RDBMS NoSQL DBMS Apache Projects Data Feeds VZW Network, Clickstream, Time, Location Data PIP and other Enterprise Data FiOS Other 3 rd party data Publicly available data

7 Orion today

8 Precision Market Insights Connecting marketers with consumers, improving engagement and driving audience response Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 8

9 New Challenges for Marketers The rise of Big Data & Mobile are complicating marketing activities Data Onslaught of 1 st and 3 rd party information Access to key information from channels Mobile Burgeoning channel for customer interactions The amount of data from mobile is exploding Mobile can be the solution Bridge Digital & Physical Direct customer relationship Context for cross-channel interactions Increased efficiency of customer communications Siloed data stores across internal groups Lack of standards for tracking and targeting

10 The Online/Mobile Ad Ecosystem AD NETWORKS $0.40 BRANDS $1.00 AGENCIES $0.10 PUBLISHERS $0.35 CONSUMER Buy wholesale inventory for advertisers Streamline multiple ad networks for publishers AD EXCHANGES $0.10 ENABLER (Targeting, Data provider) $0.05

11 Mobile + Big Data Solutions for Marketing App Usage Clickstream Location Demographics Married year old female in DC metro interested in tennis and luxury brands; recently browsed Tory Burch shoes at m.nordstrom.com and visits their store 3 times per month. Precision enables better 1:1 understanding of customers across physical and digital contexts to drive more relevant and personalized interactions

12 Case Study: Relevant Mobile Ads Drive A Major Sports League OBJECTIVE Engage a targeted audience to click on a mobile banner ad, inviting them to purchase and download the NFL Premium App. PRECISION MEETS THE CHALLENGE Precision can reach a broad audience across multiple properties and devices, accurately targeting specific mobile user segments. AUDIENCE Known interest in sports Previous basic app download Have basic app but not premium app Privacy RESULT Had 64% higher CTR compared to other mobile media properties Precision 0.38% Others 0.23%

13 Adaptive Architecture for Optimal Advertising Data Driven Customer Model Adjustment Mechanism Automated Decision Maker Observed Consumer Behavior 13

14 Automatic Profile Discovery 14

15 Large-scale Machine Learning with Applications to Connected Machines Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement.

16 Connected Ecosystem Market Size Healthcare By 2017, in the U.S., there will be Finance Energy Over 426 Million Ways to Interact with Customers Manufacturing Transportation 207 Million Smartphones Distribution Source: emarketer, August 2013 and Frost & Sullivan 2013 Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 16

17 Telematics / Fleet Management Today OEM & AFTERMARKET TELEMATICS COMMERCIAL TELEMATICS ENERGY SECURITY HEALTHCARE EDUCATION MANUFACTURING RETAIL FINANCIAL SERVICES Others CONSUMER TELEMATICS: O E M E M B E D D E D HUGHES ACQUISITION CONSUMER TELEMATICS: A F T E R M A R K E T I N F O T A I N M E N T S M A R T T R A N S P O R T A T I O N : T R A F F I C PARKING S M A R T T R A N S P O R T A T I O N : A I R M A R I T I M E COMMERCIAL T E LEMATICS: F LEET COMMERCIAL T E L E M A T I CS: W O R K F O R CE W O R K - O R D E R A S S E T M G T. S M A R T T R A N S P O R T A T I O N : CA R S H A R I N G R A I L PUBLIC T R A N S P O R T A S S E T M G T. M U N I T R A N S P O R T S M A R T B U I LDINGS R E T A I L Making Safety and Efficiency Real Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 17

18 Connected Machine Landscape (I) Car 250M cars on the road in the US, with about 20M new cars sold each year. Location analytics Comparative analytics across vehicle makes/models Prognostic and diagnostic vehicle health management (self-healing autonomous systems) Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 18

19 Connected Machine Landscape (II) Over 100M mobile subscribers currently on Verizon s network Car + Consumer Driver behavior (safety, remote management) Introduces uncertainty Need for personalization Autonomous control Consumer behavior (infotainment, advertising, etc.) Information retrieval Relevance Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 19

20 Connected Machine Landscape (III) Car + Consumer + Environment Environmental inputs (Traffic, weather, emergency services, etc.) Data correlation Distributive decision making and optimization 100s of TB of environmental data generated daily Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 20

21 Advanced Analytics for Connected Machines Verizon has core competency in advanced analytics areas including: Anomaly Detection Diagnostics Predictive Analytics Time Series Analysis and Forecasting Correlations and Association Analysis Optimization Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 21

22 Connected Machine Health Management Manual intervention or autonomous and selfhealing ( intelligent ) systems Domain rules and expert opinion Data and domain rules 4. Mitigation What actions to take? Predictive Analytics 1. Anomaly Detection Is something different? 3. Prognosis What will happen next and when? Real- and near-realtime monitoring and anomaly detection 2. Diagnosis What is happening? Domain rules and expert opinion Comprehensive and real-time view of big data relevant to diagnosis Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 22

23 Primary Source: Switches, routers and other machines Multiple Kernel Anomaly Detection Discrete Primary Source: Maintenance Reports Continuous Multiple Kernel Anomaly Detection (MKAD) Textual Primary Source: Network connectivity Networks Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 23

24 Optimization problem One class SVMs training algorithms require solving the quadratic problem Dual form Q 1 min 2 i j Ki, j i, j Subject to: i i 1 0,1, Linear equality constraint Control parameter 0 1 i, i l Bounds on design variables : Lagrange multipliers of the primal QP problem

25 Anomaly scores Decision boundary is determined only by margin and non-margin support vectors obtained by solving the QP problem,, f z, K h i i, z i k Data points with will be the support vectors Indicator 0 Sign of h: if negative outlier if positive - normal Value of h: degree of anomalousness

26 Increasing Degree of Anomalousness The Impact of Big Data Innovations Numeric Inputs Only Numeric and Text Inputs The addition of Numeric and Text Data in the new MKAD algorithm significantly improves (by as much as 7000 points) the ranking of the monitored anomalies. Execution time increased by approximately 5%. 26

27 Anomaly Detection using NASA MKAD Algorithm Unusual Auto Landing Configuration Reported Exceedance Level 3: Speed low at touchdown Level 2: Flaps at questionable setting at landing Aviation Safety Program Annual Review 2012 SSAT Project 27

28 Big Data and Telematics Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 28

29 Telematics F L E E T Monitor fleet location, condition and driver behavior W O R K F L O W Manage work and share updates both internally and externally W O R K F O R C E Align resource capacity with work and direct the right resource to the right location at the right time CLOUD ANALYTICS NETWORK I N V E N T O R Y Monitor the location and condition of cargo What Can We Build? Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 29

30 Connected Car Solutions After market data collectors (for all modern makes and models) Built-in data collectors for Mercedes and other brands Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 30

31 1 Billion Miles Dallas, TX United States Detroit, MI New York, NY Chicago, IL Houston, TX 58% City Driving, 42% Highway Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 31

32 Real-world MPG calculations are sampled from thousands of vehicles, not just a handful. Japanese OEM Vehicles U.S. OEM Vehicles Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 32

33 Driver behavior is analyzed at multiple levels. Driver behavior can be compared and contrasted to drivers from other OEM vehicles. Volkswagen Vehicles: mi/day Mitsubishi Vehicles: mi/day Honda Vehicles: mi/day Nissan Vehicles: mi/day Ford Vehicles: mi/day Toyota Vehicles: mi/day Cadillac (GM) Vehicles: mi/day Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 33

34 Driver behavior is analyzed at multiple levels. Older model years are driven at significantly lower speeds: More homogeneous driving across model years All Vehicles Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement. 34

35 Prediction Interval Estimation using Bootstrap Regression Ensemble Learning Approach We have proven that the empirical quantiles of the bootstrap prediction models can be used to consistently estimate the prediction intervals. S. Kumar and A. N. Srivastava, under submission to NIPS 2014.

36 Anomaly Detection Application S. Kumar and A. N. Srivastava, under review at NIPS 2014.

37 Acknowledgements + Notes The Entire Verizon Big Data and Analytics Team My former team at NASA Collaborators at Stanford We are hiring for junior and senior roles: Data Scientists Machine Learning experts Platform Engineers Analytics and visualization Application Developers Product Managers