Max Little

@MaxALittle

Applied mathematician and statistician at University of Birmingham/MIT/Oxford, TED fellow: Views are entirely my own.

Oxford, UK
Vrijeme pridruživanja: ožujak 2012.

Tweetovi

Blokirali ste korisnika/cu @MaxALittle

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

  1. Prikvačeni tweet
    13. kol 2019.

    After six years of work, pleased to announce that my latest book "Machine Learning for Signal Processing" is published by Oxford University Press today:

    Prikaži ovu nit
    Poništi
  2. prije 3 sata
    Poništi
  3. proslijedio/la je Tweet
    2. velj

    Should We Trust Algorithms? by David Spiegelhalter

    Poništi
  4. 2. velj

    Agreed. Better to join forces across disciplines, rather than arguing over differences. Similar applies to ML + signal processing, which motivated me to write this textbook:

    Poništi
  5. proslijedio/la je Tweet
    31. sij

    Useful perspective by who thinks this isn't meaningful: "This particular example is not going to accelerate drug discovery much, because this is speeding up a minor part of the process ...nowhere near a rate-limiting step."

    Prikaži ovu nit
    Poništi
  6. proslijedio/la je Tweet
    31. sij

    A few years ago TensorFlow looked indomitable. Now it's on the verge of being overtaken by Pytorch. OpenAI just announced it's standardizing on Pytorch, joining the likes of Tesla, Nvidia, and FB. Google hoped TensorFlow would cement its lead in ML. It may soon be a liability.

    Poništi
  7. proslijedio/la je Tweet
    30. sij

    Excited to share some new work on convergence guarantees for GPs: . led by G. Wynne and joint with .

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    29. sij

    'Trouble replicating cancer studies a ‘wake-up call’ for science' - results from the Reproducibility Project: Cancer Biology are quite sobering

    Poništi
  9. 30. sij

    Nice application of a few current ML/AI tricks (RL, imitation learning etc.) I don't know if these computational tricks are entirely necessary, but regardless, I'd put my bets on DL eventually being profitable for such narrow tasks.

    Poništi
  10. proslijedio/la je Tweet
    23. sij

    I'm hiring postdoctoral fellows in (1) network science and (2) data science with focus on smartphone data and health, both in my group at Harvard. Please share!

    Poništi
  11. 29. sij

    "Ending Parkinson's Disease: A Prescription for Action" It's my pleasure to help disseminate my esteemed colleague's, exceptionally clear, book on conquering this formidable medical condition. All proceeds to charity.

    Poništi
  12. proslijedio/la je Tweet

    See our team’s recent research where they’ve demonstrated machine learning can reveal undetected signs of anemia in the eye from non-invasive images -

    Poništi
  13. proslijedio/la je Tweet

    *Our paper diagnosing problems in fair ML now on arXiv!* . Took a few weeks, b/c our interdisciplinary collaboration broke the paper categorization system. 😂

    Prikaži ovu nit
    Poništi
  14. 28. sij

    "The role of epigenetic-related codes in neurocomputation: dynamic hardware in the brain" Neuronal complexity is likely far greater than current AI assumes. Here is evidence that cells influence neuronal "hardware" changes in their neighbours.

    Poništi
  15. proslijedio/la je Tweet
    28. sij

    Risk models can predict death or stroke after ACS, but do not perform well for *all* patients. Stultz Lab , , & developed “unreliability score” to determine if model’s results can be trusted for given patient cc

    , , i još njih 3
    Poništi
  16. proslijedio/la je Tweet
    24. sij

    It's a pity most recent grads I interview don't know EM since teaching HMMs have fallen out of favor with the rise of DL models (HMMs are the first place many first encounter EM in NLP). General ML courses seem to skip EM too since there is already so much to cover in DL.

    Poništi
  17. proslijedio/la je Tweet
    27. sij
    Odgovor korisnicima i sljedećem broju korisnika:

    I think you’re proving his point 😉

    Poništi
  18. proslijedio/la je Tweet
    27. sij
    Odgovor korisniku/ci

    Perfect is the problem. Less than 1% of all ed ct heads contain thousands of types of fatal findings. Since at least some of these thousands will be incredibly rare in any training set, the normal detector will miss a large proportion (ie call them normal). Deadly.

    Poništi
  19. 27. sij

    "The origins of the AI world’s overconfidence here lie not in any particular vendetta against radiologists, but in the structural affinities of artificial intelligence itself."

    Poništi
  20. proslijedio/la je Tweet
    27. sij

    Image resolution affects models for chest xray diagnosis

    Poništi
  21. proslijedio/la je Tweet
    25. sij

    More praise for the proof by Ji-Natarajan-Vidick-Wright-Yuen showing MIP*=RE, reported by . Scott Aaronson: "It's a huge deal." William Slofstra: "It’s an incredible achievement." Lance Fortnow: "There’s no reason to think it’s wrong."

    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:

    ·