Prof. Anima Anandkumar

I am not surprised at cluelessness of young #DeepLearning researchers given their leader is a dogmatic bully. Bayesian methods will be increasingly important as safety-critical applications like #SelfDriving and #healthcare deploy #AI. @andrewgwils has a great threadhttps://twitter.com/andrewgwils/status/1210354001041969152 …

I work closely with @carlesgelada and he is one of the most open-minded, least dogma-driven researchers I know. We have been thinking hard about safety, and it's for exactly that reason that we are critical of Bayesian NNs. 1/3

Calling an entire area useless is the opposite of open minded outlook and very definition of dogma. Provocation can start interesting discussions but it's wrong way to have a discourse.

Carles's tweets were quite measured & non-provocative. He did not say BNNs were "useless" -- he said "makes no sense", a clear invitation for people to explain & inform. Later on in the thread, he even explicitly gives examples where BNNs may be useful (though he is bearish).

Twitter makes it hard to infer tone and intentions. In any case "makes no sense" is still negative. Instead he could have explicitly said he would like to learn more about it. Instead he himself called it controversial. A clear attempt at provocation

I certainly don't think that forbidding all negative comments will lead to productive scientific discourse. Also, leading with "Controversial Opinion" was for the explicit purpose of inviting discussion -- the tweet would be far more provocative without it! 1/2

Also, Carles had about 800 followers at the time. I'm confident most of those understood his intentions. In contrast, how many of your 25k followers do you think judged Carles negatively on the basis of your Tweet, without even reading his original, nuanced criticism? 2/2