{"id":1815,"date":"2017-12-19T06:42:09","date_gmt":"2017-12-19T06:42:09","guid":{"rendered":"http:\/\/causality.cs.ucla.edu\/blog\/?p=1815"},"modified":"2017-12-27T23:49:35","modified_gmt":"2017-12-27T23:49:35","slug":"nips-2017-qa-follow-up","status":"publish","type":"post","link":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/2017\/12\/19\/nips-2017-qa-follow-up\/","title":{"rendered":"NIPS 2017: Q&A Follow-up"},"content":{"rendered":"
Dear friends in causal research,<\/div>\n
<\/div>\n
Last week I spoke at a workshop on machine learning and causality, which\u00a0followed the NIPS conference in Long Beach. Below please find my response to several questions I was asked<\/div>\n
after my talk. I hope you will find the questions and answers\u00a0to be of relevance\u00a0to issues discussed on this blog.<\/div>\n
-Judea<\/div>\n
<\/div>\n
———————————————–<\/div>\n
To: Participants at the\u00a0NIPS “What If” workshop<\/div>\n
<\/div>\n
\n
Dear friends,<\/div>\n
Some of you asked me for copies\u00a0of my slides. I am attaching them with this message, and you can get the\u00a0accompanying paper by clicking here:
\nhttp:\/\/ftp.cs.ucla.edu\/pub\/stat_ser\/r475.pdf<\/a><\/span><\/span><\/div>\n
\n

NIPS 17 – What If? Workshop Slides (PDF)<\/a><\/p>\n

NIPS 17 – What If? Workshop Slides (PPT [zipped])<\/a><\/p>\n

I have also received interesting questions at the end of my talk, which I could not fully answer in the short break we had.\u00a0I will try to answer them below.<\/p>\n

Q.1. What do you mean by the “Causal Revolution”?<\/strong>
\nAns.1:\u00a0<\/i>“Revolution” is a poetic word to summarize Gary King’s\u00a0observation: \u00a0“More has been learned about causal inference\u00a0in the last few decades than the sum total of everything\u00a0that had been learned about it in all prior recorded history”\u00a0(see cover of Morgan and Winship’s book, 2015).\u00a0It captures the miracle that only three decades ago we could not write a formula for: “Mud does not
\ncause Rain” and,\u00a0today<\/span><\/span><\/span>, we can formulate and estimate\u00a0every causal or counterfactual statement.<\/p>\n

Q2: Are the estimates produced by graphical\u00a0models the same as those produced by\u00a0the potential outcome approach?<\/strong>
\nAns.2:<\/em> Yes, provided the two approaches start\u00a0with the same set of assumptions. The assumptions in the graphical approach are advertised in the graph, while those in the potential outcome\u00a0approach are articulated separately by the investigator,\u00a0using counterfactual vocabulary.<\/p>\n

Q3: The method of imputing potential outcomes\u00a0to individual units in a table appears totally\u00a0different from the methods used in the graphical approach. Why the difference?<\/strong>
\nAns.3:<\/em> Imputation works only when certain assumptions of conditional ignorability hold. The table itself does not show us what\u00a0the assumption are, nor what they mean. To see what they\u00a0mean we need a graph, since no mortal can process such\u00a0assumptions in his\/her head. The apparent difference in\u00a0procedures reflects the insistence (in the graphical framework) on\u00a0seeing the assumptions, rather than wishing them away.<\/p>\n

Q4: Some say that economists do not\u00a0use graphs because their problems are different,\u00a0and they cannot afford to model the entire economy.\u00a0Do you agree with this explanation?<\/strong>
\nAns.4:<\/em> No way! Mathematically speaking, economic problems are no\u00a0different from those faced by epidemiologists (or\u00a0other social scientists) for whom graphical models\u00a0have become a second language. Moreover, epidemiologists have never complained that graphs\u00a0force them to model the entirety of the human anatomy.\u00a0Graph-avoidance among (some) economists is a cultural\u00a0phenomenon, reminiscent of telescope-avoidance among\u00a0Church astronomers in 17th century Italy.\u00a0Bottom line: epidemiologists can judge the plausibility\u00a0of their assumptions — graph-avoiding economists cannot.\u00a0(I have offered them many opportunities to demonstrate\u00a0it in public, and I don’t blame them for remaining silent; it is not a problem that can be managed by an unaided intellect)<\/p>\n

Q.5: Isn’t deep-learning more than just glorified\u00a0curve-fitting? After all, the objective of\u00a0curve-fitting is to maximize “fit”, while in deep-learning\u00a0much effort goes into minimizing “over-fit”.<\/strong>
\nAns.5:<\/em> No matter what acrobatics \u00a0you go through to\u00a0minimize overfitting or other flaws in your learning\u00a0strategy, you are still optimizing some property of the\u00a0observed data while making no reference to the world outside\u00a0the data. \u00a0This puts \u00a0you right back on rung-1\u00a0of the Ladder of Causation with all the limitations that\u00a0rung-1 entails.<\/p>\n

If you have additional questions on these or other topics, feel free to post them here on our blog causality.cs.ucla.edu\/blog,\u00a0(anonymity will be respected), and I will try my best to answer them.<\/p>\n

Enjoy,
\nJudea
\n———————————————–<\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Dear friends in causal research, Last week I spoke at a workshop on machine learning and causality, which\u00a0followed the NIPS conference in Long Beach. Below please find my response to several questions I was asked after my talk. I hope you will find the questions and answers\u00a0to be of relevance\u00a0to issues discussed on this blog. […]<\/p>\n","protected":false},"author":7,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[38,16],"tags":[],"class_list":["post-1815","post","type-post","status-publish","format-standard","hentry","category-conferences","category-general"],"_links":{"self":[{"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/posts\/1815","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/comments?post=1815"}],"version-history":[{"count":4,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/posts\/1815\/revisions"}],"predecessor-version":[{"id":1822,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/posts\/1815\/revisions\/1822"}],"wp:attachment":[{"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/media?parent=1815"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/categories?post=1815"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/causality.cs.ucla.edu\/blog\/index.php\/wp-json\/wp\/v2\/tags?post=1815"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}