Björn Holzhauer

@BjornHolzhauer

Biostatistician in drug development - quite Bayesian. I do dose finding, machine learning, deep learning and chess. Our geese let me share their garden.

Vrijeme pridruživanja: siječanj 2019.

Tweetovi

Blokirali ste korisnika/cu @BjornHolzhauer

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

  1. Prikvačeni tweet
    9. srp 2019.

    Now out: my wife's fantastic book on solar power finance!

    Poništi
  2. 15. ruj 2019.

    I can confirm: She rates "Where is Mr. Penguin" much more highly.

    Poništi
  3. 18. srp 2019.

    The video abstract of "Solar Power Finance without the Jargon":

    Poništi
  4. 28. lip 2019.

    Fascinating and plausible - you wonder though whether can distinguish higher numbers, but just don't see a point for it while on the hunt... What freaks me out more than all those spiders is that the article is bizarrely null hypothesis testing obsessed.

    Poništi
  5. 19. lip 2019.

    Exciting! As a test-reader during the writing, I can say that it's a great read. Declaration of conflicts of interest: married to author.

    Poništi
  6. proslijedio/la je Tweet
    29. svi 2019.

    EfficientNets: a family of more efficient & accurate image classification models. Found by architecture search and scaled up by one weird trick. Link: Github: Blog:

    Poništi
  7. proslijedio/la je Tweet
    30. svi 2019.

    I’m honoured and excited to be keynoting PyCode Conference this year!

    Poništi
  8. 24. svi 2019.

    Predictable result of our toddler wanting to hug the goslings.

    Poništi
  9. 2. svi 2019.

    What also worked well: a single random effects model with an extra parameter that lets historical data differ from new data. That parameter gets a prior that approximates model averaging between borrowing & no borrowing of information (spiky near 0, but with long tails).

    Prikaži ovu nit
    Poništi
  10. 2. svi 2019.

    I looked at mixtures of a vague + an informative (based on historical data) joint prior for the mean & SD of random trial main effects. Weights <=0.5 (and probably lower) for the informative prior component looked good in my simulations.

    Prikaži ovu nit
    Poništi
  11. 2. svi 2019.

    The accepted version of my paper on this was just posted at the Statistics in Biopharmaceutical Research website. I focused on first events/patient-year data, but I think findings should generalize to other data types. Let me know, if you want to know more/the full article.

    Prikaži ovu nit
    Poništi
  12. 2. svi 2019.

    How to use prior information from historical trials in Bayesian meta-analyses? How to make them robust to prior-data conflicts (old vs. new controls)? How much weight to give the historical prior information?

    Prikaži ovu nit
    Poništi
  13. 21. tra 2019.

    Eventually the goslings deigned to hatch, too.

    Prikaži ovu nit
    Poništi
  14. 21. tra 2019.

    We had hoped the goose eggs would hatch today. Instead: 11 ducklings in our garden that a duck hatched in the reeds of our pond. Since then she's made them jump off a 2.5m wall, ended up on the wrong side of our 2m electrified fence (we rescued them) and barged in on our dinner.

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    6. tra 2019.

    What a time to be alive: collecting whale mucus via UAV! No doubt the first stage in a complex pipeline for data analysis or machine learning. Anyone else have equally exotic data collection techniques in work they're doing?

    Poništi
  16. proslijedio/la je Tweet
    5. tra 2019.

    Today I took the day off work to go through the edited proofs of my book, Solar Power Finance without the Jargon. Really wish I hadn't got so many 2017 numbers to update to 2018. Also I had to log into work to tell my clients about SolarReserve Aurora being cancelled.

    Prikaži ovu nit
    Poništi
  17. proslijedio/la je Tweet

    anomalies 1880-2017 by country 🌡. No matter how you visualize it, it looks scary! Download / watch hi-res 🎞:

    Poništi
  18. proslijedio/la je Tweet
    25. ožu 2019.

    ULMFit from + Data Augmentation with backtranslation can get 80+% validation accuracy using only 50 training examples on IMDB sentiment classification! Full paper for at . Thread below.

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    5. velj 2019.
    Odgovor korisniku/ci

    Work by pharma statisticians shows that a huge number of historical patients can easily be beaten by a few concurrent controls. The problem is commentators who don’t understand the stats.

    Poništi
  20. proslijedio/la je Tweet
    2. velj 2019.

    state-of-the-art in , - 530 leaderboards • 974 tasks • 712 datasets • 9205 papers with code - Computer Vision - Natural Language Processing - Medical - Speech and more

    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:

    ·