Technically, I compute integrals for a living.
Joined June 2020
- Tweets1.5K
- Following481
- Followers480
- Likes901
Pinned Tweet
So, for people who want to reproduce, I spent a bit of time writing up what was happening exactly + cleaning up the code:
colab.research.google.com/dr…
Have fun everybody
Adrien Corenflos retweeted
posteriors 𝞡 published at ICLR!
I’ll be in Singapore next week, let’s chat all things scalable Bayesian learning! 🇸🇬👋
This is what "prompt engineering" should have been from day 1.
Is that @willkurt and I teaching a @DeepLearningAI course with @AndrewYNg? I think it is!
You'll learn how to:
- Use OpenAI's structured output API to build a nifty social media bot
- Apply retry-based methods with Instructor
- Go under the hood with Outlines
@dottxt
Adrien Corenflos retweeted
One #postdoc position is still available at the National University of Singapore @NUSingapore to work on sampling, high-dimensional data-assimilation, and diffusion/flow models. Applications are open until the end of January. Details:
alexxthiery.github.io/jobs/2…
Adrien Corenflos retweeted
🔥You enjoyed @ArnaudDoucet1 talk but want even more Schrodinger Bridge? Come talk to me at our poster!
🔷Schrodinger Bridge Flow for Unpaired Data Translation
🔊 East Exhibit Hall A-C #2504
Work done with my amazing collaborators Ira Korshunova @AndriyMnih and @ArnaudDoucet1
I saw Chris present this a few months back. Pretty interesting for people (like me) into particle MCMC. He questions calibrating it based on the variance of the log likelihood and shows that (for once!) one should work in the likelihood space directly.
Chris Sherlock: Variance bounds and robust tuning for pseudo-marginal Metropolis--Hastings algorithms arxiv.org/abs/2411.10785 arxiv.org/pdf/2411.10785
Adrien Corenflos retweeted
New preprint on importance sampling, (particle) independent MCMC and their unbiased versions! One bit of math is we use an anti-concentration inequality (L^p version of Paley-Zygmund) to lower bound the acceptance probability. Curious if it can be used for other MCMC algorithms!
George Deligiannidis, Pierre E. Jacob, El Mahdi Khribch, Guanyang Wang: On importance sampling and independent Metropolis-Hastings with an unbounded weight function arxiv.org/abs/2411.09514 arxiv.org/pdf/2411.09514
Adrien Corenflos retweeted
Apple released a hearing aids feature for the AirPods Pro a while ago. I bought a pair for grandma, but then realized that the feature was geoblocked in India
So we at @_lagrangepoint decided to unblock it. It ended up involving a leaky microwave and building a Faraday cage:
ALT An Airpods case, an iPad, and a Faraday cage made of copper and aluminum
Adrien Corenflos retweeted
Hidemaro Suwa, Synge Todo. [cond-mat.stat-mech]. Control of probability flow in Markov chain Monte Carlo -- Nonreversibility and lifting. (Replacement). arxiv.org/abs/1207.0258
The authors may consider simply using the glowing rocks to light the room in the first place.
Adrien Corenflos retweeted
This coming Tuesday, our seminar welcomes Diana Cai, presenting on "Batch and match: score-based approaches for black-box variational inference". Subscribe to our mailinglist groups.google.com/g/internat… and calendar calendar.google.com/calendar… for up-to-date schedule!
This is a paragraph I like to read.
Looking for good news this morning?
We just posted this on arXiv: we present the first local problem that shows a super-constant separation between the classical randomized LOCAL model of distributed computing and its quantum counterpart!
arxiv.org/abs/2411.03240
Adrien Corenflos retweeted
NEW PAPER: Provable robustness of Bayesian filters with simple modifications (nudging) against model misspecification in the transition kernel arxiv.org/abs/2411.00218
Reviewer 2 likes this.
$100K or 100 Days: Trade-offs when Pre-Training with Academic Resources
paper: arxiv.org/abs/2410.23261
repo: github.com/apoorvkh/academic…
Adrien Corenflos retweeted
Aaditya Ramdas, Ruodu Wang. [math.ST]. Hypothesis testing with e-values. arxiv.org/abs/2410.23614v1
Maybe I'll finally turn frequentist.
Aaditya Ramdas, Ruodu Wang: Hypothesis testing with e-values arxiv.org/abs/2410.23614 arxiv.org/pdf/2410.23614
