Nathan Kallus

@nathankallus

Researcher interested in: personalization; optimization, especially under uncertainty; causal inference; credible+robust inference; bandits+RL; algo fairness.

Vrijeme pridruživanja: prosinac 2010.

Tweetovi

Blokirali ste korisnika/cu @nathankallus

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

  1. Prikvačeni tweet
    28. pro 2019.

    My research group (located at Cornell Tech campus in NYC) is looking to recruit a postdoc to work on topics related to causal inference, fairness in ML, and sequential decision making (bandits+RL). Positions are renewable (1-2 years). Please retweet to spread the word. 🙏

    Prikaži ovu nit
    Poništi
  2. 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"

    Poništi
  3. proslijedio/la je Tweet
    23. sij

    Generalization Bounds and Representation Learning for Estimation of Potential Outcomes and Causal Effects by Fredrik D. Johansson et al. including ,

    Poništi
  4. proslijedio/la je Tweet
    16. sij

    They're teaching partial identification pretty early these days.

    Poništi
  5. proslijedio/la je Tweet

    This George Saunders quote captures so much of what I feel (and miss) about my few happy years as an unqualified PhD student. [Author’s Note to CivilWarLand in Bad Decline]

    Poništi
  6. 13. sij

    We cannot fix what we cannot measure! Thank you for funding my FAI proposal on *credible* fairness assessments and robustly fair algorithms: Proud+excited to be working with the amazing people at on this project.

    Poništi
  7. proslijedio/la je Tweet
    13. sij

    "But too seldom is the question asked: how can AI help correct these disparities?" Hot off the press! Check out our new commentary "Treating Health Disparities with AI" w/ coauthors Shalmali Joshi +

    Poništi
  8. proslijedio/la je Tweet
    12. sij

    Feeling so safe. Pinch me, I must be dreaming to live in a “world class” cycling city with infrastructure this good 👼

    Poništi
  9. 10. sij

    So I've tried and given up on setting up an application system. To apply: - Send me an email with subject “[PostDoc App] <Applicant Name>” with cover letter and CV. - Ask two recommenders to send me an email with subject “[PostDoc Rec] <Applicant Name>” with their rec letter.

    Poništi
  10. proslijedio/la je Tweet
    7. sij

    NYU Ops day on Mar 6, 2020, at Cornell Tech, Roosevelt Island, NYC. Exciting set of speakers! Details and registration: . Co-organizers: myself, and Huseyin Topaloglu.

    Poništi
  11. proslijedio/la je Tweet
    6. sij

    These slides and this project are so exciting! It's especially very cool to see some design decisions that reflect broader challenges (and also maybe opportunities!) for algorithmic recs in decision-making and healthcare.

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

    I finally gave in and made a twitter account. I'm planning to keep the Machine Learning (Theory) blog for longer form discussions, but I've often found myself wanting to discuss relatively short things for which Twitter seems more natural than a blog.

    Poništi
  13. proslijedio/la je Tweet
    3. sij

    My new paper – Are Dissenters Epistemically Arrogant? – is out! You can read it here:

    Poništi
  14. 1. sij

    Sneak peak: under a Fréchet-deriv (stronger than Gateaux deriv used in DML) orthogonality (holds for quantile est + other cases), the oracle estimation equation is asymp equivalent to one where nuisances are evaluated at true parameter value. LDML targets this new formulation. 5/

    Prikaži ovu nit
    Poništi
  15. 1. sij

    Instead LDML uses an initial bad guess (eg IPW) to localize the estimation. Via a new 3-way cross-fold method and a finer analysis, we can ensure oracle-like behavior for our estimator without ever learning such complicated nuisances: just plain ML classification/regression. 4/

    Prikaži ovu nit
    Poništi
  16. 1. sij

    including DML and TMLE, would require we learn a whole conditional distribution function nonparametrically, which is practically challenging for ML -- especially compared to standard classification/regression, which is all we'd need for efficient average effect estimation. 3/

    Prikaži ovu nit
    Poništi
  17. 1. sij

    Causal inference on quantile treatment effects is important in assessing the risk to the population to be treated. When dealing with rich confounders/relationships we need to use ML to adjust them. But existing ML-based approaches for efficient estimation in this case, 2/

    Prikaži ovu nit
    Poništi
  18. 1. sij

    To bring in the new year +Masa+I just posted a paper on Localized Debiased ML for estimating causal quantities using ML methods when hi-dim nuisances depend on estimand In this thread I'll explain why this prob is so important and what we did 1/

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    27. pro 2019.

    My 2 cents on the (re-ignited) NIPS vs NeurIPS debate. 1. I personally was not aware of connotations of this word (except one time when I google searched it) and was never teased by anyone about it. But I have grown to understand that we live in many worlds within this world and

    Prikaži ovu nit
    Poništi
  20. 28. pro 2019.

    Link to application system will be shared soon.

    Prikaži ovu nit
    Poništi
  21. 28. pro 2019.

    Applicants should submit a cover letter, CV, and 2 letters of recommendation. Applications will be considered on a rolling basis and sending your materials early is encouraged. Applicants are also encouraged to email/DM me to notify me of intent to submit a complete application.

    Prikaži ovu nit
    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:

    ·