All About that Bayes: Probability, Statistics, and the Quest to Quantify Uncertainty

Lawrence Livermore National Laboratory statistician Kristin Lennox delves into the history of statistics and probability in this talk, "All About that Bayes: Probability, Statistics, and the Quest to Quantify Uncertainty," given at LLNL on July 28, 2016.

Abstract:

The great Bayesian vs. Frequentist war has raged within statistics for almost 100 years, much to the confusion of outsiders. The Bayesian/Frequentist question is no longer academic, with both styles of inference appearing frequently in scientific literature and even the news. In this talk, Kristin Lennox aims to explain the great divide to non-statisticians, and also to answer the most important statistical question of all: how does probability allow us to better understand our world?

[qwosters]

I am in awe. This is my new favorite person ever.

[morganweiss]

Can't believe I watched this just because I was randomly interested, AND I thoroughly enjoyed learning and watching the entire lecture. If only I could learn more things from her! A fantastic presentation with just the right amount of humor. Being an expert does not always mean being a good teacher. She clearly is both!

[ShellyDeForte]

"Most people view statisticians as a hybrid between an accountant and a wizard, which is ridiculous, we have nothing in common with accountants."

OMG, this woman.

[miglriccardi]

I came here expecting to learn about Bayes Theorem. I did not expect I'd fall in love.

[stephenpuryear]

I have been re-watching this for years now. I still get new insights from this tremendous hour. Over time I have come to believe that a sense of humor is perhaps the single most important tool in the Lennox toolbox.

[sach]

There is lots of noise out there about Bayesian and Frequentist methods. This talk clearly lays out the different approaches. And she is hilarious to boot!

[juliocardenas]

This is the best stats talk I’ve ever seen !!
I’m a fan

[iirolenkkari]

Great talk! It is untrivial to find a knowledgeable person that knows how to communicate knowledge. There are plenty of smart people who can't explain things to less equipped people such as myself. There are also plenty of not so smart people who give 'entertaining' talks, but which don't translate any true information. Where smart meets the ability to communicate, therewithin lies gold.

[charlesaydin]

Phenomenal talk! This should be added to the material for any statistics / probability course!

[billwindsor]

@Kristin Lennox - awesome job on this talk; clear presentation and juxtapositions on Bayesian and Frequentist Statistics, and a great example application on the successful search for the submarine 'Scorpion' in the Atlantic -- *thank you!*

[jaijeffcom]

Delightful. I came back to watch again and saw I had marked it with a thumbs down. I didn’t mean to do that! What’s not to like? A clear lecture, and with jokes.

[aNytmare]

I love the descriptions of the pictures of Thomas Bayes

[roelofvuurboom]

Starting at 39:40: "De Finetti never actually worked with quantum physics...might of changed his mind." Lennox is talking about the belief that randomness as an objective reality does not exist and so probability based on the concept of randomness does not exist either but only probability as a description of our uncertainty in our beliefs or knowledge about a certain event. Lennox makes the statement that if De Finetti had met up with quantum physicists then he might have changed his mind because a lot of quantum physics has to do with fundamental randomness. The fundamental irony here is that in fact the opposite has occurred: there is now a branch of physics called quantum bayesianism - qbism for short - which has, in fact, followed De Finetti. Qbism believes that was is touted to be objective randomness of processes in quantum physics can be better explained by our (the observers) lack of certainty in what is going on at the quantum level. Although there are issues with this interpretation (as with any interpretation) it does provide intuitive - almost trivial - explanations for some highly non-intuitive - and literally incomprehensible - explanations in "classical" quantum physics including the concept of an instantly collapsing probability wavefront across all of space and the concept of quantum entanglement. This last concept has to do with the idea that if one particle is observed to have a particular property value then an entangled particle will instantly display a related property value no matter how far the particles are apart. Qbism states that all that is happening is our knowledge is being updated. So for example, if I have a pair of socks, one of which is red and one of which is blue and with my eyes closed I put one in one suitcase and the other in another suitcase and close the suitcases. I then send one suitcase to my cousin across the world. If I then open the suitcase I kept and see that the sock is blue then I instantly know that the sock in the other suitcase is red. In Qbism there is no randomness as to the color of the sock before being observed but simply the degree of certainty of our knowledge about the colors of the socks in each suitcase.

[maiconlourenco]

6:40: Probability is a measure. Distribution, Parameter and Likelihood.

[redbeangreenbean]

I wish I watched this lecture seven years ago

[pappaflammyboi]

I'm a full-blown Bayesian, but I will look to priors derived from Frequentist techniques if I can't derive nor find one of my own that I deem reasonably believable.

[Rhoude]

Bayes Theorem of subjective probability has also been employed to locate lost nuclear weapons in the Mediterranean and a lost Russian lost sub in the Pacific Ocean.

[mikaelcarneholm]

I'm going to take a Bayesian approach and say the probability that her next presentation is fantastic is greater than 0.5 and that all other predictions are ignorant.

[maiconlourenco]

17:00:The word frequentist referes to the long range frequency of experiments.

[maiconlourenco]

24:40:They calculate the conditional distribution of a parameter theta given the observed data X.