Johan Ugander

@jugander

Assistant Professor, MS&E. Social networks and social data. Occasionally disappear into the mountains.

San Francisco, CA
stanford.edu/~jugander/
Vrijeme pridruživanja: travanj 2009.
170 fotografija ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @jugander

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

  1. proslijedio/la je Tweet
    prije 10 sati

    Why do random embedding methods for high-dimensional Bayesian optimization produce inconsistent results? Find out in our new paper w/ et al . Implementation of new method now in Ax. See for usage + replication materials.

    3 proslijeđena tweeta 30 korisnika označava da im se sviđa
    Poništi
  2. proslijedio/la je Tweet
    30. sij

    Are you an Assistant Professor doing research at the interface of Data Science and Management Science? If yes, then apply to attend the - workshop that I, with Shunyuan Zhang, Allie Feldberg, and Alex Volfovsky, are organizing.

    7 proslijeđenih tweetova 25 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  3. 29. sij

    Covering and packing problems:

    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  4. 29. sij

    Posting some things I've dug up. Fast shortest-path algo heuristic: using the "beacons" idea:

    1 reply
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  5. 29. sij

    Know about nearest neighbor lit, e.g. Karger-Ruhl, and p2p network distance ([30] above). Most interested in applied _graph_ problems that are studied under metric restrictions on growth (double dim or otherwise; bounded degree doesn't count). 2/2

    1 reply
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  6. 29. sij

    In their 2003 FOCS paper, Gupta, Krauthgamer, and Lee made the following comment about the relevance of doubling metrics to algorithm design: Seeking pointers to further work in applied/practical algorithm design where bounded growth has played a role. 1/2

    1 reply 4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    29. sij

    New postdoc starting in Fall 2020, working on ML or data science approaches for combating abusive behavior in online communities:

    40 proslijeđenih tweetova 44 korisnika označavaju da im se sviđa
    Poništi
  8. proslijedio/la je Tweet
    29. sij

    ML conferences should also start having COI slides, ... some talks feel like advertisements. So many faculty have "dual" industry-academia roles now in ML.

    When I was in an engineering department for grad school, COI was never mentioned ... not even once. I think traditional science and engineering disciplines should include COI on presentations like we do in med schools and hospitals, this would change the culture ...
    3 proslijeđena tweeta 12 korisnika označava da im se sviđa
    Poništi
  9. proslijedio/la je Tweet
    27. sij

    Have you ever calculated the sample size for an and come up with a sample size that is bigger than you can ever practically get? Does this mean you shouldn't run the test? No! A paper thread for 1/17

    28 proslijeđenih tweetova 120 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  10. proslijedio/la je Tweet
    27. sij

    The 2020 Atlantic Causal Inference Conference (ACIC) will be in Austin, Texas, May 27-29. Submissions due Feb 7. This coincides with the conference being renamed "American" rather than "Atlantic"

    13 proslijeđenih tweetova 46 korisnika označava da im se sviđa
    Poništi
  11. proslijedio/la je Tweet
    27. sij

    Now with due credit to

    1 reply 44 proslijeđena tweeta 137 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    24. sij

    The call for postdoc fellows at Michigan in data science for 2020 is up now: . Feel free to reach out if you have questions.

    22 proslijeđena tweeta 28 korisnika označava da im se sviđa
    Poništi
  13. 23. sij

    Following up, this new work was inspired by work Arjun & I had at ICML'19 on modeling IIA violations: A very practical model, connects old and new ideas, and nests MNL so can be used for testing (but not all violations!)

    Very happy to share a paper that student Arjun Seshadri and I have been working on for quite a while: "Fundamental Limits of Testing the Independence of Irrelevant Alternatives in Discrete Choice" Feedback sought! Thread! 1/n
    Prikaži ovu nit
    9 korisnika označava da im se sviđa
    Poništi
  14. 23. sij

    We suggest (in the conclusion) a number of open testing question in discrete choice: strong stochastic transitivity (SST), regularity, etc. It'd be nice to see efficient constructive test for at least some of these, if not IIA! 10/10

    3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  15. 23. sij

    IIA testing has a much messier geometry than any of the problems we've seen results for. Our lower bounds come from counting Eulerian orientations of cycle decompositions of a bipartite graph constructed from data(!!), which each give ways to step outside the IIA model class. 9/n

    1 reply 4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  16. 23. sij

    In this way, our works builds on recent work by , , , Sivaraman Balakrishnan, , Greg and Paul Valiant, and others. Read their papers! 8/n

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  17. 23. sij

    This work entailed a new foray for me, into the exciting recent literature on finite-sample analysis of various testing questions, where many things turn out to be testable before they're estimable. 7/n

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. 23. sij

    While we do not propose any new tests in this work, we hope that our lower bounds (and proof techniques) can open the door to new constructive tests that test IIA rigorously, if not also efficiently. 6/n

    1 reply 2 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  19. 23. sij

    Our main result is a series of structure-dependent lower bounds on the minimax risk of the best possible test. "Structure" here is the quilted structure of how the sets that are relevant overlap. 5/n

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  20. 23. sij

    If you're concerned about every possible violation of IIA on every possible subset, it's pretty intuitive (albeit hard to prove!) that the test will require a huge number of samples. What's less clear is: what if you're only concerned with specific types of sets (e.g. pairs)? 4/n

    1 reply 1 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi

@jugander 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:

    ·