SlideShare a Scribd company logo
1 of 13
10/15/11	
  




It’s	
  a	
  great	
  pleasure	
  and	
  privilege	
  to	
  be	
  here	
  today	
  at	
  the	
  ACM	
  Data	
  Mining	
  Camp	
  in	
  
San	
  Jose.	
  We	
  are	
  delighted	
  to	
  once	
  again	
  par�cipate	
  in	
  this	
  event	
  as	
  a	
  Gold	
  Sponsor.	
  
Many	
  thanks	
  for	
  invi�ng	
  us	
  back	
  this	
  year!	
  	
  




                                                                                                                                                   1	
  
10/15/11	
  




As	
  the	
  name	
  of	
  our	
  product	
  implies,	
  we	
  are	
  all	
  about	
  promo�ng	
  Bayesian	
  networks	
  as	
  
a	
  framework	
  and	
  BayesiaLab	
  as	
  a	
  so�ware	
  tool.	
  Of	
  all	
  the	
  possible	
  mo�va�ons	
  for	
  
using	
  Bayesian	
  network,	
  such	
  as	
  knowledge	
  discovery	
  in	
  high-­‐dimensional	
  domains,	
  I	
  
want	
  to	
  focus	
  on	
  another,	
  o�en	
  neglected	
  topic,	
  namely	
  causality.	
  But	
  before	
  we	
  get	
  
to	
  that	
  par�cular	
  point,	
  please	
  allow	
  me	
  to	
  recap	
  some	
  of	
  the	
  dominant	
  headlines	
  in	
  
our	
  industry.	
  




                                                                                                                                             2	
  
10/15/11	
  




I	
  don’t	
  need	
  to	
  tell	
  you	
  that	
  big	
  data	
  is	
  probably	
  the	
  single	
  most	
  frequently	
  used	
  
buzzword	
  in	
  our	
  industry.	
  As	
  data	
  miners	
  we	
  are	
  (and	
  should	
  be)	
  delighted	
  that	
  we	
  
can	
  draw	
  upon	
  this	
  richness	
  of	
  informa�on.	
  




                                                                                                                                               3	
  
10/15/11	
  




And,	
  our	
  analy�cs	
  tools	
  are	
  becoming	
  more	
  and	
  more	
  powerful	
  and	
  sophis�cated.	
  
The	
  group	
  assembled	
  here	
  today	
  probably	
  knows	
  be�er	
  than	
  anybody	
  else	
  what	
  
tremendous	
  progress	
  has	
  been	
  made	
  in	
  the	
  field	
  of	
  analy�cs	
  and	
  data	
  mining.	
  




                                                                                                                              4	
  
10/15/11	
  




Honestly,	
  all	
  of	
  us	
  feel	
  pre�y	
  good	
  about	
  our	
  algorithms	
  and	
  advanced	
  sta�s�cal	
  
methods.	
  Our	
  services	
  as	
  data	
  scien�sts	
  are	
  certainly	
  in	
  great	
  demand,	
  not	
  only	
  here	
  
in	
  Silicon	
  Valley	
  (just	
  think	
  about	
  how	
  much	
  recrui�ng	
  is	
  going	
  on	
  here	
  today!).	
  
Between	
  big	
  data	
  and	
  supercompu�ng,	
  we	
  feel	
  indeed	
  very	
  powerful	
  with	
  our	
  
knowledge.	
  




                                                                                                                                           5	
  
10/15/11	
  




However,	
  does	
  that	
  mean	
  we	
  are	
  omniscient?	
  Do	
  we	
  really	
  understand	
  the	
  subjects	
  
we	
  are	
  studying	
  with	
  our	
  fancy	
  tools?	
  Can	
  we	
  truly	
  generate	
  a	
  deep	
  understanding	
  of	
  
our	
  problem	
  domains?	
  




                                                                                                                                             6	
  
10/15/11	
  




I	
  don’t	
  want	
  to	
  take	
  my	
  five	
  minutes	
  on	
  the	
  podium	
  here	
  to	
  go	
  into	
  a	
  metaphysical	
  
direc�on,	
  but	
  rather	
  reference	
  Judea	
  Pearl’s	
  explana�on	
  of	
  “deep	
  understanding.”	
  
He	
  says:	
  “Deep	
  understanding	
  means	
  knowing,	
  not	
  merely	
  how	
  things	
  behaved	
  
yesterday,	
  but	
  also	
  how	
  things	
  will	
  behave	
  under	
  new	
  hypothe�cal	
  circumstances.”	
  
Thus	
  he	
  makes	
  the	
  clear	
  dis�nc�on	
  between	
  observa�onal	
  and	
  causal	
  inference.	
  
Deep	
  understanding	
  requires	
  knowledge	
  of	
  the	
  causal	
  mechanism.	
  
	
  
This	
  will	
  not	
  necessarily	
  surprise	
  us,	
  as	
  we	
  o�en	
  hear	
  the	
  warning	
  “Correla�on	
  does	
  
not	
  imply	
  causa�on.”	
  We	
  will	
  all	
  nod	
  in	
  agreement	
  and	
  carry	
  on.	
  
	
  




                                                                                                                                                7	
  
10/15/11	
  




The	
  problem	
  is	
  that	
  we	
  are	
  quite	
  good	
  at	
  observa�onal	
  inference,	
  with	
  robust	
  
sta�s�cal	
  tools,	
  while	
  our	
  methods	
  for	
  causal	
  inference	
  are	
  o�en	
  rather	
  tenuous.	
  This	
  
metaphor	
  of	
  a	
  steel	
  chain	
  and	
  a	
  string	
  highlights	
  the	
  weakness	
  in	
  our	
  understanding	
  
and,	
  as	
  a	
  result,	
  our	
  reasoning.	
  




                                                                                                                                          8	
  
10/15/11	
  




As	
  a	
  consequence	
  of	
  this	
  imbalance	
  of	
  capabili�es,	
  we	
  o�en	
  do	
  not	
  address	
  causality	
  
directly,	
  but	
  rather	
  take	
  the	
  “don’t	
  ask,	
  don’t	
  tell”	
  approach.	
  I’m	
  exaggera�ng	
  to	
  
make	
  my	
  point,	
  but	
  analysts	
  o�en	
  choose	
  non-­‐commi�al	
  phrases	
  in	
  expressing	
  their	
  
findings	
  and	
  then	
  let	
  their	
  audience	
  make	
  up	
  their	
  own	
  causal	
  conclusions	
  -­‐	
  at	
  their	
  
own	
  risk.	
  




                                                                                                                                               9	
  
10/15/11	
  




There	
  is	
  indeed	
  no	
  easy	
  automated	
  method	
  for	
  discovering	
  causal	
  rela�onships	
  and	
  
genera�ng	
  causal	
  inference,	
  but	
  there	
  is	
  a	
  framework	
  that	
  facilitates	
  causal	
  
representa�on	
  in	
  very	
  formal	
  way:	
  Bayesian	
  networks.	
  They	
  allow	
  us	
  to	
  precisely	
  
encode	
  non-­‐causal	
  and	
  causal	
  dependencies	
  between	
  the	
  variables	
  of	
  interest	
  and	
  
then	
  leverage	
  this	
  knowledge	
  to	
  the	
  fullest	
  extent	
  possible.	
  




                                                                                                                                10	
  
10/15/11	
  




Beyond	
  evangelizing	
  about	
  Bayesian	
  networks,	
  we	
  are	
  here	
  to	
  promote	
  our	
  
BayesiaLab	
  so�ware	
  as	
  an	
  integrated	
  pla�orm	
  for	
  learning,	
  analyzing	
  and	
  simula�ng	
  
Bayesian	
  networks	
  and,	
  most	
  importantly,	
  carrying	
  out	
  causal	
  inference.	
  




                                                                                                                              11	
  
10/15/11	
  




Although	
  we	
  are	
  rela�vely	
  small	
  in	
  terms	
  of	
  our	
  company	
  size,	
  we	
  can	
  confidently	
  
point	
  to	
  a	
  long	
  list	
  of	
  highly-­‐respected	
  companies	
  and	
  academic	
  ins�tu�ons,	
  many	
  of	
  
which	
  are	
  Fortune	
  500	
  companies.	
  They	
  have	
  come	
  to	
  recognize	
  Bayesian	
  networks	
  
and	
  BayesiaLab	
  as	
  powerful	
  tools	
  for	
  exploring	
  and	
  researching	
  all	
  kinds	
  of	
  problem	
  
domains.	
  




                                                                                                                                        12	
  
10/15/11	
  




We	
  invite	
  you	
  to	
  visit	
  us	
  today	
  at	
  our	
  exhibi�on	
  booth	
  here	
  on	
  the	
  eBay	
  campus	
  in	
  
order	
  to	
  learn	
  more	
  about	
  the	
  power	
  of	
  Bayesian	
  networks.	
  Thank	
  you	
  for	
  your	
  
a�en�on	
  and	
  have	
  a	
  great	
  day	
  here	
  at	
  the	
  Data	
  Mining	
  Camp!	
  




                                                                                                                                                13	
  

More Related Content

Similar to Bayesian Networks & BayesiaLab

Big Data Madison: Architecting for Big Data (with notes)
Big Data Madison: Architecting for Big Data (with notes)Big Data Madison: Architecting for Big Data (with notes)
Big Data Madison: Architecting for Big Data (with notes)MIO | the data experts
 
Presentation to EASE, Tallinn, June 2012
Presentation to EASE, Tallinn, June 2012Presentation to EASE, Tallinn, June 2012
Presentation to EASE, Tallinn, June 2012Sarah Callaghan
 
Putting Great KM Ideas into Practice
Putting Great KM Ideas into PracticePutting Great KM Ideas into Practice
Putting Great KM Ideas into PracticeKate Simpson
 
Chris Lacinak
Chris LacinakChris Lacinak
Chris LacinakFIAT/IFTA
 
Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Martin Stewart-Weeks
 
Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Martin Stewart-Weeks
 
On the Emergent Semantic Web and Overlooked Issues - 2004
On the Emergent Semantic Web and Overlooked Issues - 2004On the Emergent Semantic Web and Overlooked Issues - 2004
On the Emergent Semantic Web and Overlooked Issues - 2004Yannis Kalfoglou
 
Libraries: Change and our Future
Libraries: Change and our FutureLibraries: Change and our Future
Libraries: Change and our FutureMal Booth
 
Document Databases In Online Publishing
Document  Databases In  Online Publishing Document  Databases In  Online Publishing
Document Databases In Online Publishing Irakli Nadareishvili
 
Nuvalo Nephophobia Seattle 2011
Nuvalo Nephophobia Seattle 2011Nuvalo Nephophobia Seattle 2011
Nuvalo Nephophobia Seattle 2011wlambert_2001
 
Meetup 22/2/2018 - Artificiële Intelligentie & Human Resources
Meetup 22/2/2018 - Artificiële Intelligentie & Human ResourcesMeetup 22/2/2018 - Artificiële Intelligentie & Human Resources
Meetup 22/2/2018 - Artificiële Intelligentie & Human ResourcesDigipolis Antwerpen
 
Delivering on the Promise of Big Data and the Cloud
Delivering on the Promise of Big Data and the CloudDelivering on the Promise of Big Data and the Cloud
Delivering on the Promise of Big Data and the CloudBooz Allen Hamilton
 
How To Write Literature Essays. Scholarship essay: How to write literary essay
How To Write Literature Essays. Scholarship essay: How to write literary essayHow To Write Literature Essays. Scholarship essay: How to write literary essay
How To Write Literature Essays. Scholarship essay: How to write literary essaybdg8266a
 
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-share
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-shareBigData Visualization and Usecase@TDGA-Stelligence-11july2019-share
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-sharestelligence
 

Similar to Bayesian Networks & BayesiaLab (20)

Big Data Madison: Architecting for Big Data (with notes)
Big Data Madison: Architecting for Big Data (with notes)Big Data Madison: Architecting for Big Data (with notes)
Big Data Madison: Architecting for Big Data (with notes)
 
Presentation to EASE, Tallinn, June 2012
Presentation to EASE, Tallinn, June 2012Presentation to EASE, Tallinn, June 2012
Presentation to EASE, Tallinn, June 2012
 
Interview
InterviewInterview
Interview
 
Ljc
LjcLjc
Ljc
 
Salesforce Wave
Salesforce WaveSalesforce Wave
Salesforce Wave
 
Putting Great KM Ideas into Practice
Putting Great KM Ideas into PracticePutting Great KM Ideas into Practice
Putting Great KM Ideas into Practice
 
The Gulf of Learnology
The Gulf of LearnologyThe Gulf of Learnology
The Gulf of Learnology
 
Chris Lacinak
Chris LacinakChris Lacinak
Chris Lacinak
 
Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011
 
Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011Anzsog national conference engaging citizens msw july 2011
Anzsog national conference engaging citizens msw july 2011
 
On the Emergent Semantic Web and Overlooked Issues - 2004
On the Emergent Semantic Web and Overlooked Issues - 2004On the Emergent Semantic Web and Overlooked Issues - 2004
On the Emergent Semantic Web and Overlooked Issues - 2004
 
Libraries: Change and our Future
Libraries: Change and our FutureLibraries: Change and our Future
Libraries: Change and our Future
 
2009 06 few
2009 06 few2009 06 few
2009 06 few
 
Document Databases In Online Publishing
Document  Databases In  Online Publishing Document  Databases In  Online Publishing
Document Databases In Online Publishing
 
Nuvalo Nephophobia Seattle 2011
Nuvalo Nephophobia Seattle 2011Nuvalo Nephophobia Seattle 2011
Nuvalo Nephophobia Seattle 2011
 
Vna origins
Vna originsVna origins
Vna origins
 
Meetup 22/2/2018 - Artificiële Intelligentie & Human Resources
Meetup 22/2/2018 - Artificiële Intelligentie & Human ResourcesMeetup 22/2/2018 - Artificiële Intelligentie & Human Resources
Meetup 22/2/2018 - Artificiële Intelligentie & Human Resources
 
Delivering on the Promise of Big Data and the Cloud
Delivering on the Promise of Big Data and the CloudDelivering on the Promise of Big Data and the Cloud
Delivering on the Promise of Big Data and the Cloud
 
How To Write Literature Essays. Scholarship essay: How to write literary essay
How To Write Literature Essays. Scholarship essay: How to write literary essayHow To Write Literature Essays. Scholarship essay: How to write literary essay
How To Write Literature Essays. Scholarship essay: How to write literary essay
 
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-share
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-shareBigData Visualization and Usecase@TDGA-Stelligence-11july2019-share
BigData Visualization and Usecase@TDGA-Stelligence-11july2019-share
 

More from Bayesia USA

BayesiaLab_Book_V18 (1)
BayesiaLab_Book_V18 (1)BayesiaLab_Book_V18 (1)
BayesiaLab_Book_V18 (1)Bayesia USA
 
Loyalty_Driver_Analysis_V13b
Loyalty_Driver_Analysis_V13bLoyalty_Driver_Analysis_V13b
Loyalty_Driver_Analysis_V13bBayesia USA
 
vehicle_safety_v20b
vehicle_safety_v20bvehicle_safety_v20b
vehicle_safety_v20bBayesia USA
 
Impact Analysis V12
Impact Analysis V12Impact Analysis V12
Impact Analysis V12Bayesia USA
 
Causality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact AnalysisCausality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact AnalysisBayesia USA
 
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...Bayesia USA
 
The Bayesia Portfolio of Research Software
The Bayesia Portfolio of Research SoftwareThe Bayesia Portfolio of Research Software
The Bayesia Portfolio of Research SoftwareBayesia USA
 
Causal Inference and Direct Effects
Causal Inference and Direct EffectsCausal Inference and Direct Effects
Causal Inference and Direct EffectsBayesia USA
 
Knowledge Discovery in the Stock Market
Knowledge Discovery in the Stock MarketKnowledge Discovery in the Stock Market
Knowledge Discovery in the Stock MarketBayesia USA
 
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...Bayesia USA
 
Probabilistic Latent Factor Induction and
 Statistical Factor Analysis
Probabilistic Latent Factor Induction and
 Statistical Factor AnalysisProbabilistic Latent Factor Induction and
 Statistical Factor Analysis
Probabilistic Latent Factor Induction and
 Statistical Factor AnalysisBayesia USA
 
Microarray Analysis with BayesiaLab
Microarray Analysis with BayesiaLabMicroarray Analysis with BayesiaLab
Microarray Analysis with BayesiaLabBayesia USA
 
Breast Cancer Diagnostics with Bayesian Networks
Breast Cancer Diagnostics with Bayesian NetworksBreast Cancer Diagnostics with Bayesian Networks
Breast Cancer Diagnostics with Bayesian NetworksBayesia USA
 
Modeling Vehicle Choice and Simulating Market Share with Bayesian Networks
Modeling Vehicle Choice and Simulating Market Share with Bayesian NetworksModeling Vehicle Choice and Simulating Market Share with Bayesian Networks
Modeling Vehicle Choice and Simulating Market Share with Bayesian NetworksBayesia USA
 
Driver Analysis and Product Optimization with Bayesian Networks
Driver Analysis and Product Optimization with Bayesian NetworksDriver Analysis and Product Optimization with Bayesian Networks
Driver Analysis and Product Optimization with Bayesian NetworksBayesia USA
 
BayesiaLab 5.0 Introduction
BayesiaLab 5.0 IntroductionBayesiaLab 5.0 Introduction
BayesiaLab 5.0 IntroductionBayesia USA
 
Car And Driver Hk Interview
Car And Driver Hk InterviewCar And Driver Hk Interview
Car And Driver Hk InterviewBayesia USA
 

More from Bayesia USA (17)

BayesiaLab_Book_V18 (1)
BayesiaLab_Book_V18 (1)BayesiaLab_Book_V18 (1)
BayesiaLab_Book_V18 (1)
 
Loyalty_Driver_Analysis_V13b
Loyalty_Driver_Analysis_V13bLoyalty_Driver_Analysis_V13b
Loyalty_Driver_Analysis_V13b
 
vehicle_safety_v20b
vehicle_safety_v20bvehicle_safety_v20b
vehicle_safety_v20b
 
Impact Analysis V12
Impact Analysis V12Impact Analysis V12
Impact Analysis V12
 
Causality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact AnalysisCausality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact Analysis
 
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...
Vehicle Size, Weight, and Injury Risk: High-Dimensional Modeling and
 Causal ...
 
The Bayesia Portfolio of Research Software
The Bayesia Portfolio of Research SoftwareThe Bayesia Portfolio of Research Software
The Bayesia Portfolio of Research Software
 
Causal Inference and Direct Effects
Causal Inference and Direct EffectsCausal Inference and Direct Effects
Causal Inference and Direct Effects
 
Knowledge Discovery in the Stock Market
Knowledge Discovery in the Stock MarketKnowledge Discovery in the Stock Market
Knowledge Discovery in the Stock Market
 
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...
Paradoxes and Fallacies - Resolving some well-known puzzles with Bayesian net...
 
Probabilistic Latent Factor Induction and
 Statistical Factor Analysis
Probabilistic Latent Factor Induction and
 Statistical Factor AnalysisProbabilistic Latent Factor Induction and
 Statistical Factor Analysis
Probabilistic Latent Factor Induction and
 Statistical Factor Analysis
 
Microarray Analysis with BayesiaLab
Microarray Analysis with BayesiaLabMicroarray Analysis with BayesiaLab
Microarray Analysis with BayesiaLab
 
Breast Cancer Diagnostics with Bayesian Networks
Breast Cancer Diagnostics with Bayesian NetworksBreast Cancer Diagnostics with Bayesian Networks
Breast Cancer Diagnostics with Bayesian Networks
 
Modeling Vehicle Choice and Simulating Market Share with Bayesian Networks
Modeling Vehicle Choice and Simulating Market Share with Bayesian NetworksModeling Vehicle Choice and Simulating Market Share with Bayesian Networks
Modeling Vehicle Choice and Simulating Market Share with Bayesian Networks
 
Driver Analysis and Product Optimization with Bayesian Networks
Driver Analysis and Product Optimization with Bayesian NetworksDriver Analysis and Product Optimization with Bayesian Networks
Driver Analysis and Product Optimization with Bayesian Networks
 
BayesiaLab 5.0 Introduction
BayesiaLab 5.0 IntroductionBayesiaLab 5.0 Introduction
BayesiaLab 5.0 Introduction
 
Car And Driver Hk Interview
Car And Driver Hk InterviewCar And Driver Hk Interview
Car And Driver Hk Interview
 

Bayesian Networks & BayesiaLab

  • 1. 10/15/11   It’s  a  great  pleasure  and  privilege  to  be  here  today  at  the  ACM  Data  Mining  Camp  in   San  Jose.  We  are  delighted  to  once  again  par�cipate  in  this  event  as  a  Gold  Sponsor.   Many  thanks  for  invi�ng  us  back  this  year!     1  
  • 2. 10/15/11   As  the  name  of  our  product  implies,  we  are  all  about  promo�ng  Bayesian  networks  as   a  framework  and  BayesiaLab  as  a  so�ware  tool.  Of  all  the  possible  mo�va�ons  for   using  Bayesian  network,  such  as  knowledge  discovery  in  high-­‐dimensional  domains,  I   want  to  focus  on  another,  o�en  neglected  topic,  namely  causality.  But  before  we  get   to  that  par�cular  point,  please  allow  me  to  recap  some  of  the  dominant  headlines  in   our  industry.   2  
  • 3. 10/15/11   I  don’t  need  to  tell  you  that  big  data  is  probably  the  single  most  frequently  used   buzzword  in  our  industry.  As  data  miners  we  are  (and  should  be)  delighted  that  we   can  draw  upon  this  richness  of  informa�on.   3  
  • 4. 10/15/11   And,  our  analy�cs  tools  are  becoming  more  and  more  powerful  and  sophis�cated.   The  group  assembled  here  today  probably  knows  be�er  than  anybody  else  what   tremendous  progress  has  been  made  in  the  field  of  analy�cs  and  data  mining.   4  
  • 5. 10/15/11   Honestly,  all  of  us  feel  pre�y  good  about  our  algorithms  and  advanced  sta�s�cal   methods.  Our  services  as  data  scien�sts  are  certainly  in  great  demand,  not  only  here   in  Silicon  Valley  (just  think  about  how  much  recrui�ng  is  going  on  here  today!).   Between  big  data  and  supercompu�ng,  we  feel  indeed  very  powerful  with  our   knowledge.   5  
  • 6. 10/15/11   However,  does  that  mean  we  are  omniscient?  Do  we  really  understand  the  subjects   we  are  studying  with  our  fancy  tools?  Can  we  truly  generate  a  deep  understanding  of   our  problem  domains?   6  
  • 7. 10/15/11   I  don’t  want  to  take  my  five  minutes  on  the  podium  here  to  go  into  a  metaphysical   direc�on,  but  rather  reference  Judea  Pearl’s  explana�on  of  “deep  understanding.”   He  says:  “Deep  understanding  means  knowing,  not  merely  how  things  behaved   yesterday,  but  also  how  things  will  behave  under  new  hypothe�cal  circumstances.”   Thus  he  makes  the  clear  dis�nc�on  between  observa�onal  and  causal  inference.   Deep  understanding  requires  knowledge  of  the  causal  mechanism.     This  will  not  necessarily  surprise  us,  as  we  o�en  hear  the  warning  “Correla�on  does   not  imply  causa�on.”  We  will  all  nod  in  agreement  and  carry  on.     7  
  • 8. 10/15/11   The  problem  is  that  we  are  quite  good  at  observa�onal  inference,  with  robust   sta�s�cal  tools,  while  our  methods  for  causal  inference  are  o�en  rather  tenuous.  This   metaphor  of  a  steel  chain  and  a  string  highlights  the  weakness  in  our  understanding   and,  as  a  result,  our  reasoning.   8  
  • 9. 10/15/11   As  a  consequence  of  this  imbalance  of  capabili�es,  we  o�en  do  not  address  causality   directly,  but  rather  take  the  “don’t  ask,  don’t  tell”  approach.  I’m  exaggera�ng  to   make  my  point,  but  analysts  o�en  choose  non-­‐commi�al  phrases  in  expressing  their   findings  and  then  let  their  audience  make  up  their  own  causal  conclusions  -­‐  at  their   own  risk.   9  
  • 10. 10/15/11   There  is  indeed  no  easy  automated  method  for  discovering  causal  rela�onships  and   genera�ng  causal  inference,  but  there  is  a  framework  that  facilitates  causal   representa�on  in  very  formal  way:  Bayesian  networks.  They  allow  us  to  precisely   encode  non-­‐causal  and  causal  dependencies  between  the  variables  of  interest  and   then  leverage  this  knowledge  to  the  fullest  extent  possible.   10  
  • 11. 10/15/11   Beyond  evangelizing  about  Bayesian  networks,  we  are  here  to  promote  our   BayesiaLab  so�ware  as  an  integrated  pla�orm  for  learning,  analyzing  and  simula�ng   Bayesian  networks  and,  most  importantly,  carrying  out  causal  inference.   11  
  • 12. 10/15/11   Although  we  are  rela�vely  small  in  terms  of  our  company  size,  we  can  confidently   point  to  a  long  list  of  highly-­‐respected  companies  and  academic  ins�tu�ons,  many  of   which  are  Fortune  500  companies.  They  have  come  to  recognize  Bayesian  networks   and  BayesiaLab  as  powerful  tools  for  exploring  and  researching  all  kinds  of  problem   domains.   12  
  • 13. 10/15/11   We  invite  you  to  visit  us  today  at  our  exhibi�on  booth  here  on  the  eBay  campus  in   order  to  learn  more  about  the  power  of  Bayesian  networks.  Thank  you  for  your   a�en�on  and  have  a  great  day  here  at  the  Data  Mining  Camp!   13