This is insane. I was just introducing myself to the EM algorithm through a computational stats book and was trying to figure out the case for exponential families. Thanks for this, I'll be sure to listen carefully and take notes.

I really enjoyed this video!  EM feels like frequentists dipping a toe into the Bayesian pool — not quite diving in for the full posterior, but definitely hanging out near the deep end. Love seeing the connections between the two worlds!

Maybe I missed if you mentioned this, but it should be noted that EM is only guaranteed to converge to a local maximum of the likelihood. This is not a problem if you know that the only local maxima are also global, but this is not guaranteed in general.

Thx for the video! =) very illustrative =) At 14:17 I belive you missed a 1/theta( . ) arround the light blue section. I was doing the derivation by hand and that threw me off xD

Great work, as always! The video is very beginner-friendly, especially considering how complex the algorithm is. There is only one little concern about how the Estep is shown in the video. It is somewhat unclear under which parameter the expectation of the complete loglikelihood is taken. This can cause some confusion, especially for people who aren't familiar with algorithm.

This is incredible stuff. But every time I watch a stats video I feel like I know less about stats than when I started.

Why can't we use the actual lifetimes for the customers that were less than 6 months, rather than the conditional expectations of their lifetimes, given that they left before 6 months?

(I'm sorry I couldn't help my self professor. 😂😂😂)

Roger was given no prior to create the belief that the hypothesis that new customers and old customers weren't sampled from the same population.

If he had run some Chi tests on their demographic's he would have seen so... Roger was later reprimended for wasting company resources fell into a deep depression and got fired soon after.

Nicely explained, thank you! Just a note, at 8:45 the number of people who stayed and left are swapped

What a nice video, really good for the intuition. You are a wonderful teacher.

Great video! Peace out

Awesome channel..I am a stats major and this is helping me a lot. Please keep doing more.

Great video, thank you. A good follow up would be a video explaining how would one do interval estimation in a dataset like this.

Wow beautifully explained... Cannot Thank you enough...🙏

That is amazing!!!

Wonderful!!!

Great work. Using your video to learn/create an example for my class. I replicated your example and checked the 'Exercises in EM' document and I believe the term 593(theta_(t_1)+6) needs to be devided by theta. Thanks!

Word!

Could you make a video on the em algorithms connection to reverse kl divergence?

Hello, awesome video as always! I am curious if survival analysis methods like Kaplan-Meier would also work in your example, and if so what are the differences between these two cases?