Alan Yuille

@karpathy There is evidence that brains have approximate generative models (used “as needed”). And some people have eidetic visual memories (but not clear this implies generative)

Do brains build generative models all the way down to pixel level? I happened to get woken up this morning just as I was scrutinizing a visual detail in the dream, which gave me a strong sense that it does. Previously I've been less sure. Anyone else try to debug?

@glynmoody Not really. Relying only on impact factors is crazy because of their biases and abuses. But ignoring them completely can throw out useful information.

@glynmoody Hello Moody. One problem is that AI is not tested properly and isn’t explainable. Tougher performance measures are needed like adversarial examiners (Yuille & Liu 2020). Of course it may still be a threat even then. Will radiologists believe our algorithms or their eyes?

@cihangxie @yuyinzhou_cs I confirm that Cihang and Yuyin both did excellent jobs! I’m happy and sad that they graduated last year.

My fiancée @yuyinzhou_cs and I do PhD together with @YuilleAlan from 2016-2020. I think both of us do good jobs :)

Very happy that Roger Penrose got the Nobel Prize. He introduced the maths which could make predictions from General Relativity like the formation of Black Holes. Sorry that Stephen Hawking isn’t alive to share.

@deniswirtz @JohnsHopkins @APSPhysicsDC @APSphysics @JHUArtsSciences @Stanford @UCBerkeley @Harvard @MIT @PhysRevLett @JPhysCM Extremely sad and surprising news. Mark was really nice and helpful. He lived in our neighborhood and we’d meet him sometimes. (My wife and I walked past his house only yesterday)

I have just learned that my wonderful colleague, world-renowned @JohnsHopkins Physics Prof. Mark Robbins has just died. He was not only brilliant and original, but also warm and welcoming of new colleagues and ideas. Short bio: physics-astronomy.jhu.edu/directory/mark…

@deniswirtz @JohnsHopkinsSPH @HopkinsMedicine @hopkinskimmel @HopkinsEngineer Very sad. I was honored to meet him last year when he invited me to lecture in one of his classes. A pity that we don’t have wise men like him running the country.

I just learned Prof. PC Huang has died. He was a brilliant scientist, world famous for his work on metallothionein-encoding genes, which are inducible by metals and other stress conditions. Also a great mentor and advisor, and wonderful colleague. Really sad news.

@chenxi116 @openreviewnet Those prize awarding committees usually do a better job because they have time to reflect, can consult experts, and for some prizes are able to observe the impact of the work over time. Conference reviewers are often more junior, under time pressure, and can’t consult experts.

@YuilleAlan @openreviewnet I agree given the reviewing noise. Though as the submission history accumulates, this noise may diminish (e.g. by taking the median).

Random realization: now that CMT3 and @openreviewnet have user profiles (as opposed to the pre-2015 days), they can easily rate (and rank) researchers based on the history of review scores they've received. Complementary to (better than?) h-index. Scary or no? Thumbs up or down?

@tallinzen There are theorems showing that error rates for two class classification for large numbers of iid samples falloff exponentially with Bhattacharya or Chernoff distance (depending on details on how you formulate it). I can point you to the literature if you like.

Why would you choose to use the Bhattacharyya distance over other measures of similarity of probability distributions (e.g. KL divergence)? Same for "Chernoff coefficient".

@SeyounP Thanks! A few less than last year 😿, but I’m happy because they accepted our most novel papers (which often get rejected and need to be resubmitted multiple times).

@YuilleAlan Congratulations on your 9 #CVPR2020 acceptances!!!