Joost van Amersfoort

@joost_v_amersf

PhD student in ML at Oxford by day, walking perceptron by night. Formerly .

Oxford, England
joo.st
Vrijeme pridruživanja: rujan 2009.
24 fotografije ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @joost_v_amersf

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

  1. proslijedio/la je Tweet
    30. sij

    Our 3rd Challenge on Learned Image Compression (CLIC) is live! As always, winners will be announced at our workshop at CVPR. If you are working on image reconstruction, generative modeling or compression (learned or not), consider taking part.

    1 reply 5 proslijeđenih tweetova 23 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  2. 14. sij

    I wrote this self-contained, 150 line script that trains a ResNet-18 to ~94% accuracy on CIFAR-10. Useful for obtaining a strong baseline with minimal tricks. Uses / Torchvision: 👉

    1 reply 34 proslijeđena tweeta 226 korisnika označava da im se sviđa
    Poništi
  3. 12. sij

    If anyone else used in the past and was disappointed with the experience: they've made some big improvements. The app is now a joy to use. End-to-end encryption 🔑 + no metadata stored 🤫 (not even members of groups!)

    18 korisnika označava da im se sviđa
    Poništi
  4. 7. sij

    The superior format for an academic paper in ML is:

    2 proslijeđena tweeta 1 korisnik označava da mu se sviđa
    Poništi
  5. proslijedio/la je Tweet
    24. pro 2019.

    Hey! Where are you all sleeping tonight? Anywhere good? Bad? Indifferent? 1/5

    130 replies 227 proslijeđenih tweetova 710 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  6. 12. pro 2019.

    Interested in active learning? Do you want to acquire diverse batches of data instead of single datapoints? Learn about our approach BatchBALD at 10:45, at board #1 at ! , and me are looking forward to chat!

    15 proslijeđenih tweetova 66 korisnika označava da im se sviđa
    Poništi
  7. 6. stu 2019.

    Try it for yourself with: !

    1 proslijeđeni tweet 4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  8. 6. stu 2019.

    Testing the new GPT2-XL from with my lab mate . We prompt it with "Milad is" and "Joost is", these are the results:

    7 replies 11 proslijeđenih tweetova 42 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  9. 16. srp 2019.

    A true information-theoretic approach. Claude Shannon would be proud.

    Did you know you can classify MNIST using gzip? 🤓 You can get 45% accuracy on binarized MNIST using class-wise compression and counting bits 🤗 🔥No or needed 🔥 BASH script and classifier 👉
    Prikaži ovu nit
    5 korisnika označava da im se sviđa
    Poništi
  10. 1. srp 2019.

    Removes PIL, applies normalization just once and preloads onto GPU (h/t ). It keeps all other great features (such as automatic downloading)! Can be used as a drop in replacement if you don't need fancy image transformations.

    2 proslijeđena tweeta 7 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  11. 1. srp 2019.

    Do you often use MNIST on a GPU for prototyping? Do you use ? You can get 2-3x *training* speed up and save CPU by using this 🔥FastMNIST🔥 dataset class. Code 👉

    52 proslijeđena tweeta 215 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    26. lip 2019.

    Big thanks to for providing an open source implementation of our Glow CIFAR-10 vs SVHN experiment:

    Attending ? Interested in uncertainty, generative models and out-of-distribution robustness? Check out our poster "Do deep generative models know what they don't know?" this Thursday 430-630pm (Joint work with & Dilan )
    Prikaži ovu nit
    2 proslijeđena tweeta 27 korisnika označava da im se sviđa
    Poništi
  13. 26. lip 2019.

    I'm happy to share my implementation of Glow that reproduces results from "Do Deep Generative Models Know What They Don't Know?" ( et al.) Includes pretrained model, evaluation notebooks and training code!

    76 proslijeđenih tweetova 428 korisnika označava da im se sviđa
    Poništi
  14. 25. lip 2019.

    We present BatchBALD, a theoretically grounded batch acquisition method for active learning that promotes diversity. 🧐🥳 Check out the blog post below or the preprint () and code ().

    Very happy & proud to share some research in Deep Bayesian Active Learning from , and me at 🎉🎉🎉🤗🤗🤗
    5 proslijeđenih tweetova 39 korisnika označava da im se sviđa
    Poništi
  15. 15. sij 2019.

    Area Chair to disagreeing reviewers: “DIVISION, CLEAR THE LOBBY”.

    1 reply 8 korisnika označava da im se sviđa
    Poništi
  16. 12. pro 2018.

    "NICE", "Real NVP" and "Glow" are a lovely stack of papers that allow following and in 2 year increments. Compute and algorithmic improvements go hand in hand. NICE: Real NVP: Glow:

    11 proslijeđenih tweetova 82 korisnika označavaju da im se sviđa
    Poništi
  17. proslijedio/la je Tweet
    30. lis 2018.

    Excited to announce our new research group at : Oxford Applied and Theoretical Machine Learning Group (OATML) Have a look at the website and ping me at NIPS if you'd like to join us!

    6 replies 39 proslijeđenih tweetova 231 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. 16. lis 2018.

    Excited to announce that I’m starting a PhD in ML with and ! Looking forward to the next 3+ years here in Oxford 🤓🏰

    6 replies 94 korisnika označavaju da im se sviđa
    Poništi
  19. 10. kol 2018.

    The Idaho stop: “A law that allows cyclists to treat a stop sign as a yield sign, and a red light as a stop sign.” Sounds wonderful! Wish this was possible in London. 🚲 🇬🇧

    3 korisnika označavaju da im se sviđa
    Poništi
  20. 19. srp 2018.

    Very happy to be cycling behind more of these electric black cabs in London! My lungs are ready for the battery powered future. 🚕🔋 (Bonus points for anyone who can guess the location of this pic!)

    18 korisnika označava da im se sviđa
    Poništi

@joost_v_amersf 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:

    ·