@xuanalogue Thank you for the kind words! You were an incredible TA, and we were so lucky to have you TA this class!

RT @renberlinghieri: So exciting to see MIT news featuring our work (accepted at ICML) on how to better predict currents and identify diver…

@shakir_za Thanks for the very kind words!

@raybai07 @PFort35 So glad you found these useful!

It's been wonderful working with an amazing team: MIT student Nicholas Bonaker, Dr. Emli-Mari Nel @NelEmli, and Prof. Keith Vertanen. We are indebted to @SpecialEffect, the @AceCentre, a variety of folks who offered us invaluable preliminary feedback, and our study participants.

@hannah_klinki Thanks for the kind note! So glad you enjoyed it!

@dan_p_simpson Thanks so much, Dan!

@bayesiangirl Thanks a ton, Marian!

At the link, you'll also find @david_sontag's great guest lecture on "Machine learning in healthcare." Big thanks to amazing lecture TAs Crystal Wang & Satvat Jagwani, super captioner Jeffrey Shen, and all the fantastic 6.036 course staff.

@forestgregg I think it's totally possible that time-variation could be the cause of many "small clusters"; a time-varying model would be super to pursue. You might enjoy our paper suggesting a generalization of exchangeability along these lines:

@economeager @alex_peys @rgiordan (IIRC you can adapt everything we're doing to the case where G depends on n not just via the data. Just put G_n instead of G into our equations.) But you can get Z-estimators in other/more ways too. See our footnote 2 for a discussion and example.

@alex_peys @economeager @rgiordan A final note: a *lot* of estimators are Z-estimators. If you're worried we don't cover an estimator, it might secretly or not-so-secretly be a Z-estimator.