Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

General public talk: "Logics for causal inference under uncertainty"

269 views

Published on

PhD defense general public talk.
Thesis: http://dare.ubvu.vu.nl/handle/1871/55267

Published in: Science
  • DOWNLOAD FULL BOOKS, INTO AVAILABLE FORMAT ......................................................................................................................... ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... ......................................................................................................................... ......................................................................................................................... .............. Browse by Genre Available eBooks ......................................................................................................................... Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, Cookbooks, Crime, Ebooks, Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult,
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Be the first to like this

General public talk: "Logics for causal inference under uncertainty"

  1. 1. LOGICS FOR CAUSAL INFERENCE UNDER UNCERTAINTY SARA MAGLIACANE
  2. 2. DECISION MAKING AND WHAT-IF QUESTIONS ▸ What if we gave propranolol to a patient with migraine and nausea? Image sources: https://cdn.pixabay.com/photo/2017/01/31/20/41/anatomy-2027131_960_720.png ,https://www.flickr.com/photos/possan/2352842020, http://blog.dana-farber.org/insight/2013/03/how-do-cancer-drugs-block-pathways/
  3. 3. DECISION MAKING AND WHAT-IF QUESTIONS ▸ What if we gave propranolol to a patient with migraine and nausea? ▸ What if the US withdrew from the Paris agreement? Image sources: https://cdn.pixabay.com/photo/2017/01/31/20/41/anatomy-2027131_960_720.png ,https://www.flickr.com/photos/possan/2352842020, http://blog.dana-farber.org/insight/2013/03/how-do-cancer-drugs-block-pathways/
  4. 4. DECISION MAKING AND WHAT-IF QUESTIONS ▸ What if we gave propranolol to a patient with migraine and nausea? ▸ What if the US withdrew from the Paris agreement? ▸ What if we used a certain drug on a cancerous cell? Image sources: https://cdn.pixabay.com/photo/2017/01/31/20/41/anatomy-2027131_960_720.png ,https://www.flickr.com/photos/possan/2352842020, http://blog.dana-farber.org/insight/2013/03/how-do-cancer-drugs-block-pathways/
  5. 5. DECISION MAKING AND WHAT-IF QUESTIONS ▸ What if we gave propranolol to a patient with migraine and nausea? ▸ What if the US withdrew from the Paris agreement? ▸ What if we used a certain drug on a cancerous cell? ▸ What-if questions = causal questions Image sources: https://cdn.pixabay.com/photo/2017/01/31/20/41/anatomy-2027131_960_720.png ,https://www.flickr.com/photos/possan/2352842020, http://blog.dana-farber.org/insight/2013/03/how-do-cancer-drugs-block-pathways/
  6. 6. HOW CAN WE DISCOVER CAUSAL RELATIONS? ▸ How can we discover if propranolol improves migraine?
  7. 7. HOW CAN WE DISCOVER CAUSAL RELATIONS? ▸ How can we discover if propranolol improves migraine? ▸ Classical approach: experimentation ▸ Sometimes unethical, unfeasible, too expensive ▸ For example, an ineffective drug Source: http://catbearding.com/wp-content/uploads/2013/06/cat-hate-it.gif
  8. 8. CAUSAL DISCOVERY METHODS ▸ Past 30 years: use also information from observations ▸ For example, using correlations…
  9. 9. CAUSAL DISCOVERY METHODS ▸ Past 30 years: use also information from observations ▸ For example, using correlations… https://xkcd.com/552/ Image source: https://commons.wikimedia.org/wiki/File:Crime.svg
  10. 10. CAUSAL DISCOVERY METHODS ▸ Past 30 years: use also information from observations ▸ For example, using correlations… ▸ One correlation does not imply causation, but many (especially combined with non-correlations) may ▸ Constraint-based causal discovery => Logic https://xkcd.com/552/ Image source: https://commons.wikimedia.org/wiki/File:Crime.svg
  11. 11. CONSTRAINT-BASED CAUSAL DISCOVERY EXAMPLE ▸ Observations of many* patients ▸ Assume no other relevant factors migraine nausea food poisoning Y Y N N Y Y N N N … … …
  12. 12. CONSTRAINT-BASED CAUSAL DISCOVERY EXAMPLE ▸ Observations of many* patients ▸ Assume no other relevant factors ▸ From data: ▸ Migraine is uncorrelated from food poisoning ▸ For patients with nausea, migraine and food poisoning are (negatively) correlated migraine nausea food poisoning Y Y N N Y Y N N N … … …
  13. 13. CONSTRAINT-BASED CAUSAL DISCOVERY EXAMPLE ▸ Observations of many* patients ▸ Assume no other relevant factors ▸ From data: ▸ Migraine is uncorrelated from food poisoning ▸ For patients with nausea, migraine and food poisoning are (negatively) correlated ▸ Causal relations: migraine nausea food poisoning Y Y N N Y Y N N N … … … MIGRAINE FOOD POISONING NAUSEA
  14. 14. ▸ Given enough data*, the predictions are correct CONSTRAINT-BASED CAUSAL DISCOVERY
  15. 15. ▸ Given enough data*, the predictions are correct ▸ Even with arbitrary unmeasured factors CONSTRAINT-BASED CAUSAL DISCOVERY MIGRAINE FOOD POISONING NAUSEA STRESS ICE CREAM SALES TEMPERATURE CRIME RATES
  16. 16. ▸ What happens if not enough data? ▸ Can we fully exploit multiple datasets under different conditions? ▸ Given enough data*, the predictions are correct ▸ Even with arbitrary unmeasured factors CONSTRAINT-BASED CAUSAL DISCOVERY MIGRAINE FOOD POISONING NAUSEA STRESS OPEN QUESTIONS: ICE CREAM SALES TEMPERATURE CRIME RATES
  17. 17. THESIS CONTRIBUTIONS
  18. 18. WHAT HAPPENS IF NOT ENOUGH DATA? ▸ Statistical independence tests can make errors ▸ State-of-the-art method resolves some errors, but not very fast
  19. 19. WHAT HAPPENS IF NOT ENOUGH DATA? ▸ Statistical independence tests can make errors ▸ State-of-the-art method resolves some errors, but not very fast ▸ Ancestral Causal Inference (ACI) ▸ Faster execution time on simplified problem ▸ Method to score predicted causal relations by confidence SIMULATED DATA
  20. 20. SIMULATED DATA PROTEIN SIGNALLING DATA WHAT HAPPENS IF NOT ENOUGH DATA? ▸ Statistical independence tests can make errors ▸ State-of-the-art method resolves some errors, but not very fast ▸ Ancestral Causal Inference (ACI) ▸ Faster execution time on simplified problem ▸ Method to score predicted causal relations by confidence Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK ACI (ancestral r. + indep. <= 1) Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK Weighted causes(i,j) Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK −1000 −500 0 500 1000 Weighted in Raf Mek PLCg PIP2 PIP3 Erk Akt Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK ACI (causes) Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK −1000 −500 0 500 1000 FCI Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK −1000 −500 0 500 1000 CFC Raf Mek PLCg PIP2 PIP3 Erk Akt Raf Mek PLCg PIP2 PIP3 Erk Akt PKA PKC p38 JNK Cause Effect
  21. 21. ▸ Most approaches learn causal relations on these datasets separately and then combine the results HOW CAN WE FULLY EXPLOIT MULTIPLE DATASETS?
  22. 22. ▸ Most approaches learn causal relations on these datasets separately and then combine the results ▸ Joint Causal Inference: framework that can systematically pool data* from different settings to perform independence tests TOY EXAMPLE HOW CAN WE FULLY EXPLOIT MULTIPLE DATASETS? migraine nausea food poisoning Y Y N N Y Y N N N … … … migraine nausea food poisoning Y Y N N Y Y … … … Patients prescribed with propranolol Patients MIGRAINE FOOD POISONING NAUSEA PROPRANOLOL known
  23. 23. ▸ Most approaches learn causal relations on these datasets separately and then combine the results ▸ Joint Causal Inference: framework that can systematically pool data* from different settings to perform independence tests ▸ More accurate than methods combining tests results SIMULATED DATATOY EXAMPLE HOW CAN WE FULLY EXPLOIT MULTIPLE DATASETS? migraine nausea food poisoning Y Y N N Y Y N N N … … … migraine nausea food poisoning Y Y N N Y Y … … … Patients prescribed with propranolol Patients MIGRAINE FOOD POISONING NAUSEA PROPRANOLOL known
  24. 24. CONCLUSIONS ▸ Causal inference has several important applications (e.g. systems biology)
  25. 25. CONCLUSIONS ▸ Causal inference has several important applications (e.g. systems biology) ▸ This thesis discusses methods that: 1. Improve the scalability of causal inference under uncertainty 2. Score predicted causal relations by confidence 3. Infer causal relations jointly from all available datasets
  26. 26. CONCLUSIONS ▸ Causal inference has several important applications (e.g. systems biology) ▸ This thesis discusses methods that: 1. Improve the scalability of causal inference under uncertainty 2. Score predicted causal relations by confidence 3. Infer causal relations jointly from all available datasets ▸ Not mentioned in the talk: ▸ A more scalable implementation of Probabilistic Soft Logic ▸ Future work: apply it to causal inference?

×