BigData@Chalmers
Language	
  Technology	
  Tools	
  for	
  
Cognitive	
  Computing
Devdatt	
  Dubhashi
LAB
(Machine	
  Lea...
Rise	
  of	
  the	
  Robots
A	
  Tale	
  of	
  Two	
  Stanford	
  Labs
• Artificial	
  Intelligence	
  	
  	
  
(AI	
  John	
  McCarthy)
• Intelligenc...
Why	
  do	
  we	
  need	
  Cognitive	
  Assistants?
“The	
  reason	
  I	
  was	
  interested	
  in	
  interactive	
  compu...
What	
  is	
  a	
  Cognitive	
  Assistant?
A	
  software	
  agent	
  (cog)	
  that
– “augments	
  human	
  intelligence”	
...
The  Vision…
A  scientific  enterprise  where  cognitive  
mediators  regularly  serve  as an  integral  
part  of  a  res...
Dance	
  with	
  the	
  Robots
Cognitive	
  computing	
  is	
  a	
  new	
  type	
  of	
  software	
  that	
  combines	
  m...
Why	
  Language	
  is	
  difficult	
  ..
He	
  sat	
  on	
  the	
  river	
  bank	
  and	
  counted	
  his	
  dough.
She	
 ...
“Judge a man by the company he keeps.”
- Euripides
Chris	
  Anderson
IndiaOxford	
  Uni.
Pakistan
TED Future	
  Publishing
San	
  Francisco
Chris	
  Anderson
Graph	
  Communities
“Chris Anderson” “Chris Anderson”
TED WIRED
Classification	
  with	
  Graph	
  Kernels
G1
G2
v1
v2
K(G1
,G2
)
G
Graph	
  Embeddings and	
  Kernels
~u1
~u2
~u3
~u4 ~u5
~c
1p
#
• Embed	
  discrete	
  combinatorial	
  
object	
  (graph)	...
Demonstrator	
  at	
  Recorded	
  Future
• Classifies	
  names	
  as	
  ambiguous or	
  unique
• Uses	
  graph	
  classifi...
Word	
  Embeddings
Deep	
  Learning	
  (Neural	
  Networks)
• Revolutionized	
  vision	
  
and	
  speech	
  systems	
  
• Dramatic	
  improve...
Word	
  Embeddings capture	
  meaning
Word	
  sense	
  induction
him political
her government
god influence state
came us act labour given
council about authori...
Senses	
  of	
  for	
  ‘paper’
Vis	
  using	
  t-­‐sne
Medium
Essay
Scholarly article
Newspaper
Newspaper firm
Material
Dealing with
information overload
Document	
  summarization
Word	
  vectors
+	
  Multiple	
  Kernel	
  	
  	
  
learning
+	
  Submodular
optimization	
  
M....
The	
  Mycroft	
  Cognitive	
  Assistant:	
  Enhancing	
  and	
  
supporting	
  T-­‐shaped	
  skills	
  and	
  research	
 ...
The	
  Mycroft	
  Cognitive	
  Assistant®
Mission:	
  Provide	
  a	
  suite	
  of	
  tools	
  designed	
  to	
  enhance	
 ...
Watson	
  in	
  Undergrad	
  Project
• A	
  Cognitive	
  Assistant	
  Powered	
  by	
  IBM	
  Watson:	
  A	
  Tool	
  for	...
Chris	
  Biemann:
Adaptive	
  Natural	
  Language	
  Processing
• Adaptive	
  to	
  
collection	
  and	
  to	
  
human	
  ...
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Ibm cog
Upcoming SlideShare
Loading in …5
×

Ibm cog

129 views

Published on

IBM Cognitive Systems Seminar

Published in: Education
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
129
On SlideShare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
5
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Ibm cog

  1. 1. BigData@Chalmers Language  Technology  Tools  for   Cognitive  Computing Devdatt  Dubhashi LAB (Machine  Learning.  Algorithms,  Computational  Biology)     Computer  Science  and  Engineering   Chalmers  
  2. 2. Rise  of  the  Robots
  3. 3. A  Tale  of  Two  Stanford  Labs • Artificial  Intelligence       (AI  John  McCarthy) • Intelligence   Augmentation                           (IA  Douglas  Engelbart)
  4. 4. Why  do  we  need  Cognitive  Assistants? “The  reason  I  was  interested  in  interactive  computing,   even  before  we  knew  what  that  might  mean,  arose  from   this  conviction  that  we  would  be  able  to  solve  really   difficult  problems  only  through  using  computers  to   extend  the  capability  of  people  to  collect  information,   create  knowledge,  manipulate  and  share  it,  and  then  to   put  that  knowledge  to  work…Computers most  radically   and  usefully  extend  our  capabilities  when  they  extend   our  ability  to  collaborate to  solve  problems  beyond  the   compass  of  any  single  human  mind.1” 1 Improving  Our  Ability  to  Improve:  A  Call  for  Investment  in  a  New  Future.  Douglas  C.  Engelbart,  September  2003.  
  5. 5. What  is  a  Cognitive  Assistant? A  software  agent  (cog)  that – “augments  human  intelligence”  (Engelbart’s definition1 in  1962) – Performs  tasks  and  offer  services  (assists  human  in  decision  making   and  taking  actions) – Complements  human  by  offering  capabilities  that  is  beyond  the   ordinary  power  and  reach  of  human  (intelligence  amplification) 1 Augmenting  Human  Intellect:  A  Conceptual  Framework,  by  Douglas  C.  Engelbart,  October  1962 From  Cognitive  Assistance  at  Work:  Cognitive  Assistant  for  Employees  and  Citizens,  by  Hamid  R.  Motahari-­‐Nezhad,  AAAI  2015  Fall  Symposium.
  6. 6. The  Vision… A  scientific  enterprise  where  cognitive   mediators  regularly  serve  as an  integral   part  of  a  research  team throughout  the   entire research  process.
  7. 7. Dance  with  the  Robots Cognitive  computing  is  a  new  type  of  software  that  combines  machine   learning and  natural  language  processing  techniques  in  order  to  solve   complex  problems.  IBM  Watson  is  a  revolutionary  system  implementing  the   cognitive  computing  paradigm.
  8. 8. Why  Language  is  difficult  .. He  sat  on  the  river  bank  and  counted  his  dough. She  went  to  the  bank  and  took  out  some  money. Lexical  Layer Concept  Layer synonymouspolysemous
  9. 9. “Judge a man by the company he keeps.” - Euripides
  10. 10. Chris  Anderson IndiaOxford  Uni. Pakistan TED Future  Publishing San  Francisco
  11. 11. Chris  Anderson
  12. 12. Graph  Communities “Chris Anderson” “Chris Anderson” TED WIRED
  13. 13. Classification  with  Graph  Kernels G1 G2 v1 v2 K(G1 ,G2 ) G
  14. 14. Graph  Embeddings and  Kernels ~u1 ~u2 ~u3 ~u4 ~u5 ~c 1p # • Embed  discrete  combinatorial   object  (graph)  into  continuous   Euclidean  space • Define  kernel  based  on   geometry  of  Euclidean  sp. • V.  Jethava et  al  NIPS  2012,   JMLR  2013 • T.  Kerola,  L.  Hermansson,  V.   Jethava,  F.  Johansson  CIKM   2013 • F.  Johansson,  V.  Jethava et  al   ICML  2014. • F.  Johanssso,  D.  Dubhashi  et  al NIPS  2015
  15. 15. Demonstrator  at  Recorded  Future • Classifies  names  as  ambiguous or  unique • Uses  graph  classification to  classify  occurrence   graphs of  names • Achieved  state-­‐of-­‐the-­‐art  results  (CIKM,   2013). • Powerful  extension  for  complete  disambiguation   in  progress  …   • Parallel/Distributed  implementation  in  GraphLab Ambiguous  or   Unique?
  16. 16. Word  Embeddings
  17. 17. Deep  Learning  (Neural  Networks) • Revolutionized  vision   and  speech  systems   • Dramatic  improvements   in  image  classification  – near  human  level. • Skype  real  time   translation  from  English   to  Chinese.
  18. 18. Word  Embeddings capture  meaning
  19. 19. Word  sense  induction him political her government god influence state came us act labour given council about authority energy unit system battery x performance high allows engine equipment processing systems failure management provide Instance cloud for: 'power' M.  Kageback,  F.  Johansson  et  al,   “Neural  context  embeddings for   automatic  discovery  of  word   senses”,  (NAACL  2015  workshop  on   Vector  Space  Modeling   for  NLP) Used  an  innovative  clustering technique   Exploited  word  and  context   vectors. Ongoing   work  using  LSTMs
  20. 20. Senses  of  for  ‘paper’ Vis  using  t-­‐sne Medium Essay Scholarly article Newspaper Newspaper firm Material
  21. 21. Dealing with information overload
  22. 22. Document  summarization Word  vectors +  Multiple  Kernel       learning +  Submodular optimization   M.  Kågeback,  O.  Mogren et  al,   “Extractive  Summarization   using  Continuous  Vector   Space  Models”,   Workshop   on   (CVSC)  EACL 2014 Olof Mogren,   et  ql,  “Extractiv Summarization  by  Aggregatin Multiple  similarities”   RANLP  2015
  23. 23. The  Mycroft  Cognitive  Assistant:  Enhancing  and   supporting  T-­‐shaped  skills  and  research   collaborations Kara  L.  Hall,  PhD       Director,  Science  of  Team  Science  (SciTS),  Division  of  Cancer  Control  and   Population  Sciences,  National  Cancer  Institute Paul  K.  Courtney Chief  Technology  Officer,  Open  Health  Systems  Laboratory Anil  Srivastava,   President,  Open  Health  Systems  Laboratory
  24. 24. The  Mycroft  Cognitive  Assistant® Mission:  Provide  a  suite  of  tools  designed  to  enhance  the   effectiveness  and  efficiency of  biomedical  research   collaboration. Deliver  solutions  that  support  and  facilitate  collaboration   across  entire  the  research  process,  including  exploring  a   research  problems  space,  assembling  research  teams,  to   obtaining  research  funding,  launching  a  research  project,   interacting  with  team  members,  enhancing  team   functioning,  as  well  as  developing  and  disseminating   research  products.   IBM  Watson  cognitive  computing  technology  drives  the   intelligence  of  the  tools  to  provide  a  dynamic  set  of   cognitive  assistants  to  optimize  research  productivity  and   increase  innovative  capacity.  
  25. 25. Watson  in  Undergrad  Project • A  Cognitive  Assistant  Powered  by  IBM  Watson:  A  Tool  for   Finding  Researchers  Within  the  Medical  Field.   Pontus  Thome,  Erik  Sievers,  Hannes  H aggander,  Rasmus Letterkrantz,  Fredrik  Rahn.  Chalmers.  Gothenberg,  Sweden.   January  2016
  26. 26. Chris  Biemann: Adaptive  Natural  Language  Processing • Adaptive  to   collection  and  to   human  user Adaptive Machine   Learning Use   annotations   as  features Text Data Find  regularities Annotate  regularities   in  data WWW enhance  background by  focused  crawling Structure Discovery Algorithms Structure Discovery Algorithms Structure Discovery Algorithms adaptive  annotation rich  representation jaguar ISA vehicle car brand company SIM mercedes bmw cadillac jeep ISA animal species wildlife predator SIM tiger leopard lion cougar semantic  model

×