Aaron Fisher

@ajfishr

Statistician at ; alum; recent postdoc; food hobbyist; casual cyclist and swimmer.

Vrijeme pridruživanja: srpanj 2013.

Tweetovi

Blokirali ste korisnika/cu @ajfishr

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

  1. Prikvačeni tweet
    10. stu 2018.

    Confused about why we divide by n-1 for the sample variance? The variance tells us how much people tend to be different from each other. And, for each person in our sample, there are only n-1 people we can compare them to. More in my blog post:

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    26. sij

    PLEASE RETWEET: Jacob Bien and I are hiring a post-doc!!!! It's the perfect set-up: ✅live in sunny Los Angeles ✅work on interesting statistical methodology problems ✅collaborate with Jacob and me!! Submit your application via the link by MARCH 15!!!

    Prikaži ovu nit
    Poništi
  3. 22. sij

    I struggled to understand influence functions for years. When the intuition finally clicked, writing this paper with was a delight. Our paper builds intuition with a series of visualizations, equations, and silly analogies to magicians. (2/2)

    Prikaži ovu nit
    Poništi
  4. 22. sij

    Our educational paper on visual intuition for influence functions is out in The American Statistician! arXiv PDF: IFs are central to many stat methods and can bridge machine learning with inference, but they're hard to learn (1/2)

    Prikaži ovu nit
    Poništi
  5. 12. sij
    Prikaži ovu nit
    Poništi
  6. 12. sij

    One last, silly bookkeeping note. The paper title changed twice during the revision process, but the content is largely unchanged. So, earlier arxiv versions have different titles (). (6/6)

    Prikaži ovu nit
    Poništi
  7. 12. sij

    We've also added a high level summary of the paper in Section 2, with visual illustrations. (5/6)

    Prikaži ovu nit
    Poništi
  8. 12. sij

    Our proposed measure of conditional variable importance can be estimated using matching, weighting, or imputation. (see Sections 3 and 8.2 of ) (4/6)

    Prikaži ovu nit
    Poništi
  9. 12. sij

    We define conditional importance by permuting X1 within subgroups with identical values of X2. This can be thought of as adding noise to X1 that cannot be detected by looking at the covariates alone, as the joint distribution of the covariates (X1 and X2) is unchanged. (3/6)

    Prikaži ovu nit
    Poništi
  10. 12. sij

    In the revision process, we added a section on *conditional* variable importance. That is, how much does a prediction model rely on the information in a covariate (X1) that cannot alternatively be gleaned from other covariates (X2)? (2/6)

    Prikaži ovu nit
    Poništi
  11. 12. sij

    Our variable importance paper is out in JMLR!! Often, many prediction models achieve high accuracy. A variable could be important to some models, but not others. We study how important a variable could be to *any* model (in a pre-specified class). (1/6)

    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    10. sij

    Aesthetics are important, but the main point of a figure is NOT to make the paper look pretty. When adding a figure to a paper, think hard about how the visual cues help the reader understand a result. Published network hairballs, for example, rarely covey anything useful to me.

    Poništi
  13. 8. sij
    Poništi
  14. proslijedio/la je Tweet
    10. stu 2019.

    Hey academics on Twitter: Have you ever been scooped? Do you worry about getting scooped? My job market paper estimates the consequences of getting scooped in science using cool data from structural biology. Let’s do a thread!

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    10. stu 2019.
    Poništi
  16. 28. lis 2019.

    We revised our educational paper on visual intuition for influence functions, adding 2 figures that show a (low-dimensional, discrete) special case where pathways through the space of distributions can be seen more explicitly. with

    Poništi
  17. proslijedio/la je Tweet
    17. lis 2019.

    Following the deconfounder wars? Here's my comment on "The Blessings of Multiple Causes." Summarizes my back-and-forth with and Dave Blei about the deconfounder over the last couple of years.

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    17. lis 2019.
    Odgovor korisnicima

    I have a comment that should appear on arXiv in a day or two. The Ogburn et al comment and my comment will both appear alongside "Blessings" in JASA.

    Poništi
  19. proslijedio/la je Tweet
    26. ruj 2019.

    Just read a really interesting paper that finally helped me understand influence functions -- putting this out here in case there's one other person also struggling with it :

    Poništi
  20. proslijedio/la je Tweet
    3. tra 2019.

    I'm hiring! Looking for a full-time (most likely) master's level research associate to help with stat methods, data work, lit reviews, etc. Great fit for someone looking to work for a few years in between a master's and a PhD. Share widely! @JohnsHopkinsDM

    Prikaži ovu nit
    Poništi
  21. proslijedio/la je Tweet

    Thrilled to be awarded an R01 grant from on Bayesian methods for cortical surface data with Co-I's and . Thank you also to collaborators , , & Ying Guo for all the support and help. Couldn't have done it without you!

    Poništi

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:

    ·