June 3rd, 2013BigML Inc, 2013Challenges toMake Machine Learning EasyACM San Francisco Bay Area Professional ChapterFrancis...
June 3rd, 2013BigML Inc, 2013 2Expert: Published papers at KDD, ICML, NIPS, etc ordeveloped own ML algorithms used at larg...
June 3rd, 2013BigML Inc, 2013 3Data, dataeverywhereA special report on managing informationWhy make ML easy?In the age of ...
June 3rd, 2013BigML Inc, 2013However, Machine Learning isCOMPLEX:‣tools are complicated and donot scale well‣solutions are...
June 3rd, 2013BigML Inc, 2013 5Why make ML easy?
June 3rd, 2013BigML Inc, 2013 6Why make ML easy?
April, 2013BigML Inc, 2013 7BigMLA cloud-based service that makesMachine Learning SIMPLE$ bigmler --train customer2012.csv...
June 3rd, 2013BigML Inc, 2013 8AgendaBigML web-based interface (10-15 min)Questions (10-15 min)$ bigmler --train customer2...
June 3rd, 2013BigML Inc, 2013 9How it works
June 3rd, 2013BigML Inc, 2013 10BigML Resourcescsv, arff, xlshttps, s3, azure, odataSources local and remoteDatasetsStream...
June 3rd, 2013BigML Inc, 2013 11BigML API
June 3rd, 2013BigML Inc, 2013 123,500+ users35,000+ modelsBigML
June 3rd, 2013BigML Inc, 2013 13FREE subscription?mail your username to:acm@bigml.com
June 3rd, 2013BigML Inc, 2013 14Challenges#1 Machine Learning breadth and depth#2 User Diversity#3 Simplicity#4 Scalabilit...
June 3rd, 2013BigML Inc, 2013 15...or you can deal with that!#1 Supervised learning#2 Unsupervised learning#3 Semi-supervi...
June 3rd, 2013BigML Inc, 2013 16...or you can deal with that!#1 machine learning breadth and depth
June 3rd, 2013BigML Inc, 2013 17Phrase a problem as an ML taskThe stages of an ML applicationData WranglingFeature Enginee...
June 3rd, 2013BigML Inc, 2013 18ProblemsTechniquesApplicationsClassificationRegressionClusteringDensity EstimationManifold...
June 3rd, 2013BigML Inc, 2013 19Understandingthe pastPredicting thefutureWhy Trees first?
June 3rd, 2013BigML Inc, 2013 20Why Trees?
June 3rd, 2013BigML Inc, 2013 21A Machine Learning application requires more tasks (thatare even more important) than just...
June 3rd, 2013BigML Inc, 2013 22ExpertsAficionadosPractitionerNewbiesAbsolute beginners#2 user diversityHow to prioritize ...
June 3rd, 2013BigML Inc, 2013 23Time-to-productivity++Expertise#2 user diversity
June 3rd, 2013BigML Inc, 2013 24#2 user diversityMBs PBsMBs PBsActual sizeSizeMost users believe their data is much bigger...
June 3rd, 2013BigML Inc, 2013 25NumberofJobs++Size of Job#2 user diversity
June 3rd, 2013BigML Inc, 2013 26#3 simplicity
June 3rd, 2013BigML Inc, 2013 27“Any fool can make somethingcomplicated. It takes a genius tomake it simple.”― Woody Guthr...
June 3rd, 2013BigML Inc, 2013 28‣install‣configure‣use‣train‣understand‣test‣pre-evaluate‣measure impact‣deploy‣scale‣acce...
June 3rd, 2013BigML Inc, 2013 29#4 scalabilityNCONCURRENTJOBSfrom1 CUSTOMER1 JOBfrom1 USERN JOBSfromM CUSTOMERS
June 3rd, 2013BigML Inc, 2013 30Infrastructure
June 3rd, 2013BigML Inc, 2013 31#5 measuring machinelearning impact
June 3rd, 2013BigML Inc, 2013 32Measuring “actual” impact is complex and goesbeyond traditional performance evaluation.Ima...
June 3rd, 2013BigML Inc, 2013 33Kiri Wagstaff, Machine Learning that Matters, ICML, 2012The stages of an ML research progr...
June 3rd, 2013BigML Inc, 2013 34Phrase a problem as an ML taskData WranglingLearn from DataThe stages of an ML application...
June 3rd, 2013BigML Inc, 2013 35#6 pricing
June 3rd, 2013BigML Inc, 2013 36#6 pricing
June 3rd, 2013BigML Inc, 2013 37Pre-pay-as-you-go
June 3rd, 2013BigML Inc, 2013 38Subscriptions
June 3rd, 2013BigML Inc, 2013 39...or you can deal with that!BigML 1-click modelYou can dealwith this...Machine Learning m...
June 3rd, 2013BigML Inc, 2013 40BigML 1-click modelYou can dealwith this......or you can deal with that!Machine Learning m...
June 3rd, 2013BigML Inc, 2013 41Ease-of-use++2013Machine Learning made easy?
June 3rd, 2013BigML Inc, 2013 42Ease-of-use++2013 2014 2015 2016 2017 2018Machine Learning made Easy!!!
June 3rd, 2013BigML Inc, 2013 43Questions
June 3rd, 2013BigML Inc, 2013 44Unknown Modelf : X -> YExample: ideal credit approval formulaModelsMExample: set of candid...
Upcoming SlideShare
Loading in …5
×

A few Challenges to Make Machine Learning Easy

816 views

Published on

Dr. Francisco J Martin: In the age of data, Machine Learning is the key component to make data-driven decisions, develop smart applications, and build predictive analytics. However, Machine Learning is complex. The current tools are complicated and do not scale well. Most solutions are costly, easily involving hundreds of thousands of dollars and substantial resources. Additionally, experts with industry experience are very scarce. BigML is building a scalable, cloud-based service that makes Machine Learning easy or, at least, lowers the barriers that most developers and business folks face to learn from data. In this talk, I will first demo BigML and then describe the efforts, highlight some of the key findings, and discuss some of the challenges from a technical, user, and business perspective, related to developing a Machine Learning service for the masses.

Published in: Technology, Education
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
816
On SlideShare
0
From Embeds
0
Number of Embeds
5
Actions
Shares
0
Downloads
10
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

A few Challenges to Make Machine Learning Easy

  1. 1. June 3rd, 2013BigML Inc, 2013Challenges toMake Machine Learning EasyACM San Francisco Bay Area Professional ChapterFrancisco J Martin, Ph.D.BigML Co-founder & CEOeBay Whitman Campus
  2. 2. June 3rd, 2013BigML Inc, 2013 2Expert: Published papers at KDD, ICML, NIPS, etc ordeveloped own ML algorithms used at large scale.Sampling the AudienceAficionado: Understands pros/cons of differenttechniques and/or can tweak algorithms as needed.Newbie: Just taking Coursera ML class or reading anintroductory book to ML.Absolute beginner: ML sounds like science fictionPractitioner: Very familiar with ML packages (Weka,Scikit, R, etc).
  3. 3. June 3rd, 2013BigML Inc, 2013 3Data, dataeverywhereA special report on managing informationWhy make ML easy?In the age of data, MachineLearning is the key component to:‣ make data-driven decisions‣ develop smart applications‣ build predictive analytics
  4. 4. June 3rd, 2013BigML Inc, 2013However, Machine Learning isCOMPLEX:‣tools are complicated and donot scale well‣solutions are costly‣e x p e r t s w i t h i n d u s t r yexperience are scarce4Why make ML easy?http://ttic.uchicago.edu/~samory/
  5. 5. June 3rd, 2013BigML Inc, 2013 5Why make ML easy?
  6. 6. June 3rd, 2013BigML Inc, 2013 6Why make ML easy?
  7. 7. April, 2013BigML Inc, 2013 7BigMLA cloud-based service that makesMachine Learning SIMPLE$ bigmler --train customer2012.csv --test new_customers.csv --objective churn>>> from bigml.api import BigML>>> api = BigML()>>> source = api.create_source("s3://bigml-public/csv/sales.csv")>>> dataset = api.create_dataset(source)>>> model = api.create_model(dataset)$ curl https://bigml.io/model?$BIGML_AUTH -X POST -H "content-type: application/json" -d {"dataset": "dataset/50ca447b3b56356ae0000029"}
  8. 8. June 3rd, 2013BigML Inc, 2013 8AgendaBigML web-based interface (10-15 min)Questions (10-15 min)$ bigmler --train customer2012.csv --test new_customers.csv --objective churn>>> from bigml.api import BigML>>> api = BigML()>>> source = api.create_source("s3://bigml-public/csv/iris.csv")>>> dataset = api.create_dataset(source)>>> model = api.create_model(dataset)$ curl https://bigml.io/dataset?$BIGML_AUTH -X POST -H "content-type: application/json" -d {"source": "source/50ca447b3b56356ae0000029"}BigML API, API Bindings, BigMLer (5 min)Challenges (10-15 min)#1 Machine Learning Breadth and Depth#2 User Diversity#3 Simplicity#4 Scalability#5 Measuring Impact#6 Pricing
  9. 9. June 3rd, 2013BigML Inc, 2013 9How it works
  10. 10. June 3rd, 2013BigML Inc, 2013 10BigML Resourcescsv, arff, xlshttps, s3, azure, odataSources local and remoteDatasetsStream histogramsStatisticsModelsInteractiveCompoundable Random Decision ForestsActionable: exportable to rules, code, pmmlPredictionsForm-based PredictionsQuestion by QuestionLocal predictionsEvaluationsClassificationRegressionComparison
  11. 11. June 3rd, 2013BigML Inc, 2013 11BigML API
  12. 12. June 3rd, 2013BigML Inc, 2013 123,500+ users35,000+ modelsBigML
  13. 13. June 3rd, 2013BigML Inc, 2013 13FREE subscription?mail your username to:acm@bigml.com
  14. 14. June 3rd, 2013BigML Inc, 2013 14Challenges#1 Machine Learning breadth and depth#2 User Diversity#3 Simplicity#4 Scalability#5 Measuring Machine Learning Impact#6 Pricing
  15. 15. June 3rd, 2013BigML Inc, 2013 15...or you can deal with that!#1 Supervised learning#2 Unsupervised learning#3 Semi-supervised learning#4 Reinforcement learning#5 Learning to Learn#1 machine learning breadth and depth
  16. 16. June 3rd, 2013BigML Inc, 2013 16...or you can deal with that!#1 machine learning breadth and depth
  17. 17. June 3rd, 2013BigML Inc, 2013 17Phrase a problem as an ML taskThe stages of an ML applicationData WranglingFeature EngineeringLearn from DataPre-evaluateMeasure Impact
  18. 18. June 3rd, 2013BigML Inc, 2013 18ProblemsTechniquesApplicationsClassificationRegressionClusteringDensity EstimationManifold learningActive learningetc.Just solving a couple ofproblems and using a fewtechniques thousands ofapplications can be developedchurn prevention, date matching, decision making, diagnostics, frauddetection, detecting tumors, detecting investment opportunities, humanbody pose estimation, pedestrian tracking, predictive analytics,recommendation systems, risk analysis, spam detection, etc#1 machine learning breadth and depth
  19. 19. June 3rd, 2013BigML Inc, 2013 19Understandingthe pastPredicting thefutureWhy Trees first?
  20. 20. June 3rd, 2013BigML Inc, 2013 20Why Trees?
  21. 21. June 3rd, 2013BigML Inc, 2013 21A Machine Learning application requires more tasks (thatare even more important) than just learning from data.Just solving one problem more will enable a huge number ofapplications more.What problem(s) to tackle next and which techniques touse?#1 machine learning breadth and depth
  22. 22. June 3rd, 2013BigML Inc, 2013 22ExpertsAficionadosPractitionerNewbiesAbsolute beginners#2 user diversityHow to prioritize what to buildnext? More features for theexpert or simplifying more forthe newbies?
  23. 23. June 3rd, 2013BigML Inc, 2013 23Time-to-productivity++Expertise#2 user diversity
  24. 24. June 3rd, 2013BigML Inc, 2013 24#2 user diversityMBs PBsMBs PBsActual sizeSizeMost users believe their data is much bigger thanit really is
  25. 25. June 3rd, 2013BigML Inc, 2013 25NumberofJobs++Size of Job#2 user diversity
  26. 26. June 3rd, 2013BigML Inc, 2013 26#3 simplicity
  27. 27. June 3rd, 2013BigML Inc, 2013 27“Any fool can make somethingcomplicated. It takes a genius tomake it simple.”― Woody Guthrie#3 simplicity
  28. 28. June 3rd, 2013BigML Inc, 2013 28‣install‣configure‣use‣train‣understand‣test‣pre-evaluate‣measure impact‣deploy‣scale‣access programmatically (API)#3 simplicitySimple means much more than a easy-to-use interface
  29. 29. June 3rd, 2013BigML Inc, 2013 29#4 scalabilityNCONCURRENTJOBSfrom1 CUSTOMER1 JOBfrom1 USERN JOBSfromM CUSTOMERS
  30. 30. June 3rd, 2013BigML Inc, 2013 30Infrastructure
  31. 31. June 3rd, 2013BigML Inc, 2013 31#5 measuring machinelearning impact
  32. 32. June 3rd, 2013BigML Inc, 2013 32Measuring “actual” impact is complex and goesbeyond traditional performance evaluation.Imagine that an algorithm predicts that user Alice is goingto buy a Magic Potion.‣ But Magic Potions are out of stock.‣ Should we blame‣ the algorithm for the “false positive” prediction?‣ the data scientist for not including that feature?‣ operations for running out of stock on things thatcustomers want to buy?#5 measuring machine learning impact
  33. 33. June 3rd, 2013BigML Inc, 2013 33Kiri Wagstaff, Machine Learning that Matters, ICML, 2012The stages of an ML research programVery inspirational!!!
  34. 34. June 3rd, 2013BigML Inc, 2013 34Phrase a problem as an ML taskData WranglingLearn from DataThe stages of an ML applicationFeature EngineeringPre-evaluateMeasure Impact !!!!!
  35. 35. June 3rd, 2013BigML Inc, 2013 35#6 pricing
  36. 36. June 3rd, 2013BigML Inc, 2013 36#6 pricing
  37. 37. June 3rd, 2013BigML Inc, 2013 37Pre-pay-as-you-go
  38. 38. June 3rd, 2013BigML Inc, 2013 38Subscriptions
  39. 39. June 3rd, 2013BigML Inc, 2013 39...or you can deal with that!BigML 1-click modelYou can dealwith this...Machine Learning made easy?
  40. 40. June 3rd, 2013BigML Inc, 2013 40BigML 1-click modelYou can dealwith this......or you can deal with that!Machine Learning made easy?
  41. 41. June 3rd, 2013BigML Inc, 2013 41Ease-of-use++2013Machine Learning made easy?
  42. 42. June 3rd, 2013BigML Inc, 2013 42Ease-of-use++2013 2014 2015 2016 2017 2018Machine Learning made Easy!!!
  43. 43. June 3rd, 2013BigML Inc, 2013 43Questions
  44. 44. June 3rd, 2013BigML Inc, 2013 44Unknown Modelf : X -> YExample: ideal credit approval formulaModelsMExample: set of candidatecredit approval formulasLearning from DataLearningAlgorithmBased on Learning from Data by Y. Abu-Mostafa, M. Magdon-Ismail and H. LinFinal Modelg ~ fExample: learned creditapproval formulaTraining Examples(x1, l1), (x2, l2), ..., (xN, lN)Example: historical records of credit customersx1xNlabelf1 f2 fn

×