Machine Learning, Recommender Systems, Causal Inference, Bayesian Inference https://xitter.cf/t.co/V5yOMCprrS
Joined April 2020
- Tweets461
- Following1.5K
- Followers1K
- Likes1.2K
Bayesian ML at Scale retweeted
A friend just informed me that our colleage, Professor Arthur Dempster, has died last month at 96. Arthur was an intellectual giant, famous for developing the EM algorithm as well as for the Shafer-Dempster theory, but remained skeptic about causation.
memoritree.com/memorial/arth…. May his memory be an inspiration.
Bayesian ML at Scale retweeted
Thanks for the engagement and the thoughtful response. It is difficult to outline my (many) points of agreement and the few points where I differ in an X post, but I will try.
> "I've never insisted on the "frequency interpretation" of probability."
In this paper statistical analysis is defined in frequentist terms.
escholarship.org/content/qt4…
I acknowledge I am being particularly purist here, but to a strict (operational subjective) Bayesian, probability does not exist, and the idea of a Bayesian estimator is a contradiction in terms.
The concept of "experimental conditions remaining the same" only makes sense with a frequentist notion of repeated draws from a low-dimensional probability model.
Bayesian ML at Scale retweeted
@RichardSSutton contrarian or classist?
youtu.be/PsjG0RSBUPU?si=YKJ8…
Bayesian ML at Scale retweeted
My latest @criteo article — The Future of Commerce Recommendation: Why Hybrid Systems Will Win — explores the pros and cons of today’s Performance Recommenders and LLM-based Agents, and how the next generation of RecSys Agents will blend both bit.ly/3J8reEg
Bayesian ML at Scale retweeted
I don't think causal models are "merely an inferential tool".
I think they operate on a realist interpretation of probability (frequentist probability is realist, even if it is estimated using Bayesian methods), to an anti-realist "probability does not exist".
The arrow directions also have a realist meaning of physical causation.
The SCM and do calculus answer questions within this realist framework. I agree that within this realist framework you need to go beyond probability theory to do causal inference.
The anti-realist position is, (as I see it), here is my data, D, a treatment that I will do t and a future outcome y which cause me to believe this about y if I set t=0 and this about y if I set t=1. The anti-realist position does not model the physical world but how past data and an intervention affect their beliefs about a future outcome. In this anti-realist scenario probability theory (with trivial additions) is sufficient for causal inference.
The reason to talk about realism/anti-realism is only to explain why different people can have different views about probability theory and its applicability to causal inference.
Bayesian ML at Scale retweeted
Anti-realists claiming that causal models are merely an inferential tool, should tell us what objective is served by this "tool", and why evolution has endowed us with this, rather than another (possibly conflicting) "tool". Once we have an answer to these questions, I would be able to tell whether SCM is "tool neutral", or tied to realism. @BayesianIn @soboleffspaces
I think that even the author @yudapearl himself seems genuinely uncertain about how to position his own SCM whether this causal model he has built should be regarded as a real entity or merely as an inferential tool (anti-realism).
Bayesian ML at Scale retweeted
Causal Inference is just (Half 🙃@yudapearl ? Bayesian @f2harrell 🙃 ? ) Inference
Really nice lecture by David Rohde!
youtube.com/watch?v=hoFRVJLn…
Bayesian ML at Scale retweeted
I like Larry Wasserman's quote, which separates the mathematics from the physics.
Thanks for the comment.
I have a short follow up post here that may clarify: hashnode.com/edit/cmfdqk3rd0…
I address your point more directly here (it is similar to a comment others have made, which I repeat) youtube.com/watch?v=hoFRVJLn…
I consider Pearl to be realist because a) he thinks probability exists; and b) he thinks arrow directions relate to real phenomena in the world.
Bayesian ML at Scale retweeted
I am honored to be characterized as "realist", but I used to believe that Causal Inference transcends the realism/anti-realism distinction. I'll need to re-read the distinction that @BayesianIn is making to decide if should hold on to that belief.
Realism vs Anti-Realism as Approaches to Causal Inference
Why in my opinion, @yudapearl is a realist and Bruno de Finetti is an anti-realist.
ml4interact.hashnode.dev/rea…
Bayesian ML at Scale retweeted
"The War on Science", edited by @LKrauss1 is a collection of essays that sheds light on what went wrong in academia in the USA in the past years. I highly recommend you grab a copy if you want to understand the current situation better.
amazon.com/War-Science-Thirt…
Bayesian ML at Scale retweeted
If you’re at #AISTATS2025, come check out my poster!
📝 Bayesian Off-Policy Evaluation and Learning for Large Action Spaces
📍 Hall A-E 58
⏰ Poster Session 1
Joint work with Victor-Emmanuel Brunel, David Rohde, @Korba_Anna
@aistats_conf
Bayesian ML at Scale retweeted
Last year, at @NeurIPSConf our team reached a new milestone: 15 papers accepted (13 papers - one spotlight, 1 dataset paper and 1 challenge paper) 🎉
Check all our contributions in this exhaustive recap from #NeurIPS2024 👇
techblog.criteo.com/all-our-…
Bayesian ML at Scale retweeted
A Point Process Model for Optimizing Repeated Personalized Action Delivery to Users ift.tt/wPnVGEv
Bayesian ML at Scale retweeted
🔴 #NeurIPS24 is happening this week and our Criteos will present their latest #ML papers (13 papers, 1 dataset paper and 1 challenge paper), bringing original #AI contributions on #AdTech topics.
@NeurIPSConf
Check this video to know more 👇
youtu.be/mgiVdEX2ssU
Bayesian ML at Scale retweeted
@OtmaneSakhi - Is Principled Offline A/B Testing Achievable? #NeurIPS2024
Bayesian ML at Scale retweeted
@imad_aouali - Scaleable Off-Policy Learning - Impossible without side information? #NeurIPS2024
Bayesian ML at Scale retweeted
@OtmaneSakhi and @imad_aouali - Is Principled Offline A/B Testing Achievable?
youtube.com/watch?v=GF2EoOB-… #NeurIPS2024
Bayesian ML at Scale retweeted
The talk will explore the practical implications of our findings and how they could help reduce the need for online A/B testing. It should be interesting for both theorists and practitioners alike, hope to see you there! #neurips2024
Thursday @OtmaneSakhi and @imad_aouali Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning #neurips2024. Register here: criteo.zoom.us/webinar/regis…
Bayesian ML at Scale retweeted
Call for StanCon 2025+
statmodeling.stat.columbia.e…