Ollie Hines

@UnibusPluram @BerkeleyCTML Research statement optional. DEI statement mandatory??

Postdoc position open at Berkeley @BerkeleyCTML aprecruit.berkeley.edu/JPF04532

New preprint is out!! We construct average moment estimators using constrained neural nets with some nice links to Auto-debiasing/TMLE. arxiv.org/abs/2409.19777 Really enjoyed working on this one with my supremely talented younger brother

@shortstein Not surprised it is hard - reminds me of this fun Riemann zeta function fact

@UnibusPluram @ildiazm @mark_vdlaan Very cool! Did you know that it also represents a weighted average derivative effect too? See example 5 from arxiv.org/pdf/2308.05456

this fact is actually mentioned briefly in one of @ildiazm's papers with @mark_vdlaan and similar ideas have been discussed before but we thought it deserved a fuller treatment, thinking about in what contexts binarization can make sense + helping understand the MTP better

Lots of people binarize continuous exposures to get "ATEs". Lots of papers say this is bad. Here we say "actually it can make sense sometimes"! Turns out this "ATE" can represent a sensible causal quantity defined with a modified treatment policy: arxiv.org/abs/2405.07109

Riesznet (@VC31415 et al.) has propelled Riesz representers into the causal inference limelight, but can we say more about the representers for similar looking estimands? Answer: Yes arxiv.org/abs/2308.05456 @karlado @SVansteelandt

@wortmanncallejo @infornomics @DAcemogluMIT I just gave the repo a bit of an update and the tests seem to be passing now. Do you mind raising an issue if something is still broken? github.com/ohines/teffect…

@wortmanncallejo @infornomics @DAcemogluMIT Nice to hear this code is getting some use!

Plus @edwardhkennedy called them his new favorite estimands at #ACIC2023 🙌 Or the numerator of \Psi at least

Should you use projection estimands or derivative effects for continuous exposures? It turns out that they are the same thing* (along with ATEs) + you can use the R-learner for estimation. New preprint with @karlado @SVansteelandt arxiv.org/abs/2308.05456

@UnibusPluram @AngYu_soci @edwardhkennedy I’m honoured to be mentioned in the same sentence as Edward and Alejandro! Not much to add except that the EIF derivation you’re after is a special case (X=Y) of Example 7 in our paper: arxiv.org/abs/2107.00681

@AngYu_soci @edwardhkennedy @hines8 Yeah you should work it out to make sure but it's just Op algebra

Just updated the section on influence function derivation in my book. Closely follows the great tutorials by @edwardhkennedy and @hines8! Hope it helps folks who aren't as mathematically fluent. alejandroschuler.github.io/mci/0cb2ffe5e5…

Check out our latest preprint where we propose new CATE variable importance measures for understanding heterogeneous causal effects. Pretty excited about this one! @karlado @SVansteelandt: arxiv.org/abs/2204.06030

@d_s_rod @karlado @SVansteelandt I agree! We gave a workshop on this material at the most recent @TheEuroCIM meeting but that was some time ago now. I’ll tweet about it if/when we plan another

@hines8 @karlado @SVansteelandt Hi, looking forward to reading the paper! It would be great if tutorial papers came with a live session to go through the material:) Can I ask- are there any upcoming talks on this?

Our tutorial paper has just been published! We shed some light on the dark art of influence curve derivation, which is at the core of machine learning estimators in causal inference. Oliver Dukes @karlado @SVansteelandt doi.org/10.1080/000313…

Thankful for the Central Limit Theorem

Proud to be ranked 2nd in the world for Public, Environmental & Occupational Health & 4th for Infectious Diseases by @usnews 🎉 Based on research & reputation, this reflects the hard work of our amazing students & staff 🙌 Study with us 👉 lshtm.ac.uk #GlobalHealth

🚨🚨New preprint with @karlado @SVansteelandt on average derivative effects in causal inference. TLDR - we think ADEs are a natural generalization of the average treatment effect to continuous exposures with some nice connections arxiv.org/abs/2109.13124

@cspan IMOoiooo i

President Biden hosts Virtual COVID-19 Summit - LIVE online here: c-span.org/video/?514854-…

@SVansteelandt Developed!

For anyone attending the virtual #JSM2021 - catch my talk today on new assumption-lean mediation estimands, developped with @SVansteelandt bit.ly/3lHkUWb