Robert Peharz

@ropeharz

Assistant Professor

Netherlands
Vrijeme pridruživanja: lipanj 2019.
31 fotografija ili videozapis Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @ropeharz

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @ropeharz

  1. Prikvačeni tweet
    8. ruj 2019.

    This paper has quite a history. When Martin started his PhD in 2015, and said he'd be interested in Bayesian structure learning in SPNs, my reaction was: interesting -- and challenging 🙂. But here we are, just a few years later 😄, and accepted at !

    Finally, "Bayesian Learning of Sum-Product Networks" is out as preprint on Arxiv. Amazing work, done together with amazing people!
    Prikaži ovu nit
    5 proslijeđenih tweetova 23 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    1. velj

    a solution: (PCs) which are the version of mixture models! PCs let you 1) still marginalize in poly time 2) compactly encode exponentially large mixtures stay tuned for a tutorial on with YooJung Choi at 2020!

    5 replies 14 proslijeđenih tweetova 74 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  3. 31. sij

    Hey EU, something's different about you today... lost weight?

    8 korisnika označava da im se sviđa
    Poništi
  4. 30. sij

    Tired of doing grad student descent? Try some principled post-doc sample correction! 👇

    Happy to announce latest joint work with and Fred Roosta is now on arXiv! A complete analysis of Stein importance sampling for Markov chains - a black-box, post-hoc, rejection-free correction method.
    1 proslijeđeni tweet 9 korisnika označava da im se sviđa
    Poništi
  5. proslijedio/la je Tweet
    27. sij

    Two new University Lectureships in the Department of Engineering of the University of Cambridge, in the broad area of Machine Learning and/or Computer Vision. Application deadline 1 March 2020.

    22 proslijeđena tweeta 39 korisnika označava da im se sviđa
    Poništi
  6. 24. sij

    I can recommend this book (~200 p.), which I red during my PhD time, and which fixes the biggest knowledge gaps:

    8 proslijeđenih tweetova 54 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  7. 24. sij

    Most computer scientist don't get a proper education in measure theory/rigorous probability theory. However, if you are working on probabilistic machine learning/applied statistics, you're missing out!

    8 proslijeđenih tweetova 73 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    24. sij

    Researchers had had it with and was willing to take a 3x performance hit going with a nice productive language like . To their surprise, they did not get a performance hit, but a 3x performance BOOST! Julia rocks!

    1 reply 29 proslijeđenih tweetova 82 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    17. sij

    This review on normalizing flows is excellent. It's full of clear writing, precise claims, and useful connections.

    Check out our extensive review paper on normalizing flows! This paper is the product of years of thinking about flows: it contains everything we know about them, and many new insights. With , , , . Thread 👇
    Prikaži ovu nit
    27 proslijeđenih tweetova 141 korisnik označava da mu se sviđa
    Poništi
  10. proslijedio/la je Tweet
    18. sij

    You know what I love? Machine learning and applied science researchers offering explanations of how Bayesian inference works whose fallaciousness is exceeded only by the confidence in which they are presented.

    9 replies 9 proslijeđenih tweetova 92 korisnika označavaju da im se sviđa
    Poništi
  11. proslijedio/la je Tweet
    16. sij

    Turing.jl is the killer package for probabilistic programming!

    1 reply 15 proslijeđenih tweetova 55 korisnika označava da im se sviđa
    Poništi
  12. proslijedio/la je Tweet
    15. sij

    For anyone doing computational* work involving measure theory, what capabilities would you look for in a "measures" software library? * Turing machines are cool, but here I mean "computed using an actual physical computer"

    5 proslijeđenih tweetova 5 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  13. 7. sij

    I guess the best feature is that it is a well-defined process model, inheriting tractable inference from its "parents", Tractable Probabilistic Circuits and Gaussian Processes. Work with , Franz Pernkopf, and Carl Rasmussen.

    1 proslijeđeni tweet 4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  14. 7. sij

    Super happy that our paper on Deep Structured Mixtures of Gaussian Processes was accepted at AISTATS! I truly think that this line of work adds some nice new dimensions to GP-style of models.

    Our paper on deep structured mixtures of GPs got accepted at !
    1 reply 1 proslijeđeni tweet 14 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    6. sij

    Now I have to drop everything and read this: [1912.13170] Schrödinger Bridge Samplers by Espen Bernton, Jeremy Heng, Arnaud Doucet ⁦⁩, Pierre E. Jacob ⁦⁩

    5 proslijeđenih tweetova 23 korisnika označavaju da im se sviđa
    Poništi
  16. proslijedio/la je Tweet
    6. sij

    Despite Deep 's popularity, there are precious few good intro tutorials! This is a really nice one. It combines: - toy implementation - math concepts - intuitive explanations

    8 replies 446 proslijeđenih tweetova 1.837 korisnika označava da im se sviđa
    Poništi
  17. proslijedio/la je Tweet
    30. pro 2019.

    For the morning crowd: there's plenty of new stuff by me to keep you busy in January! Introduction to Probabilistic Computation: Markov chain Monte Carlo:

    1 reply 13 proslijeđenih tweetova 92 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    27. pro 2019.
    Odgovor korisnicima

    I think the ability to express a lot structure with network architectures has been one of the driving forces behind the success of deep learning based approaches. If we can add even more structure with clever priors/interactions of priors/architectures, that would be a huge win.

    2 proslijeđena tweeta 8 korisnika označava da im se sviđa
    Poništi
  19. proslijedio/la je Tweet
    25. pro 2019.

    The phrase "the right scientific model" should never be taken seriously. How could we ever.know that we've found it?

    Odgovor korisnicima @BethCarey12 @GaryMarcus i sljedećem broju korisnika: 2
    I believe that DL (defined as gradient-based learning applied to networks of differentiable modules), is here to stay. So yes: I think DL is the right scientific model. More accurately: DL is part of the solution. Though obviously, its current incarnations are insufficient.
    1 reply 2 proslijeđena tweeta 12 korisnika označava da im se sviđa
    Poništi
  20. proslijedio/la je Tweet
    24. pro 2019.

    Now that everyone is again into logic/symbols/reasoning vs deep/learning, I'd like to repost my C&T talk: 📽️ I discuss: - some history of this false dilemma in AI - logic and pure learning are *both* brittle - probabilistic world models as middle ground

    36 proslijeđenih tweetova 206 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  21. proslijedio/la je Tweet
    18. pro 2019.

    Here are some notes I wrote up on her 2013 Shannon Lecture (back in the days when I had the time and the mojo to maintain a blog): 3/3

    1 reply 1 proslijeđeni tweet 19 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi

@ropeharz još nije objavio nijedan Tweet.

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·