Somshubra Majumdar

@HaseoX94

Chicago, IL
titu1994.github.io
Vrijeme pridruživanja: veljača 2018.
7 fotografija ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @HaseoX94

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @HaseoX94

  1. Prikvačeni tweet
    8. velj 2019.

    Making some headway in porting the PyTorch "Neural Ordinary Differential Equations" codebase to TF Eager execution. This graph below replicates the simplest example in their codebase, on TF 1.12 with Eager Execution enabled, with dopri5.

    1 reply 1 proslijeđeni tweet 2 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  2. 1. velj

    Finally finished porting Single Headed Attention RNNs to Tensorflow 2.0. I just couldn't make the time till now but satisfied that I can use it in my projects now. Fixing the last of the autograph tracing issues will have to wait.

    2 proslijeđena tweeta 7 korisnika označava da im se sviđa
    Poništi
  3. 1. velj

    So many reasons to actually use Julia instead of re-learning it for the 6th time. And that speed difference between Julia and Torchdiffeq is far too extreme. Side note, even tfdiffeq takes about the same as pytorch, last I checked.

    Universal Differential Equations for Scientific Machine Learning by and colleagues: u' = f(u, t, g(u, t)) where g is a neural network.
    2 korisnika označavaju da im se sviđa
    Poništi
  4. proslijedio/la je Tweet
    9. sij

    Training Neural SDEs: We worked out how to do scalable reverse-mode autodiff for stochastic differential equations. This lets us fit SDEs defined by neural nets with black-box adaptive higher-order solvers. With , and .

    6 replies 209 proslijeđenih tweetova 857 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet
    7. sij

    I completed my 1st data science project ~30 years ago. Since then I've been continuously developing a questionnaire I use for all new data projects, to ensure the right info is available from the start. I'm sharing it publicly today for the first time.

    55 replies 1.072 proslijeđena tweeta 3.714 korisnika označava da im se sviđa
    Poništi
  6. 1. sij

    Happy New Year to everyone !

    3 korisnika označavaju da im se sviđa
    Poništi
  7. 23. pro 2019.

    This is so much better than cloning a blank notebook

    You can now create new blank Swift Colab notebooks by going to just like other languages! (Spotted by .)
    1 korisnik označava da mu se sviđa
    Poništi
  8. proslijedio/la je Tweet
    20. pro 2019.

    Yes! I got my first big conference paper accepted at ICLR, with spotlight! We improve the previous DeepMind paper "NALU" by 3x-20x. – This took 7-8 months, working without any funding as an independent researcher. Paper: Code:

    30 replies 300 proslijeđenih tweetova 1.466 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    10. pro 2019.

    Found this gem @ ! Using orthogonal property of Legendre polynomials, Legendre Memory Units (LMU) can efficiently handle temporal dependencies spanning 100k timesteps, converge rapidly and use fewer internal state-variables compares to LSTMs.

    11 replies 161 proslijeđeni tweet 681 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  10. 3. pro 2019.

    I feel the same regarding tools. Learn both. Use PyTorch for Research. Once research phase is done, use Tensorflow to deploy or vice-versa. Or do both in Tensorflow. Or both in PyTorch. Frameworks have converged enough to allow flexible modelling and deployment.

    This whole symbolism and connectionism debacle just reminds me that the entire machine learning community is wack. Why do we always have to fight about something like pytorch vs TensorFlow or SGD vs Adam?Why can't we just come together and create dank memes?
    1 proslijeđeni tweet 11 korisnika označava da im se sviđa
    Poništi
  11. 27. stu 2019.

    A fun start to the day, and something to look forward to extending to other tasks than language modelling.

    The SHA-RNN is composed of an RNN, pointer based attention, and a “Boom” feed-forward with a sprinkling of layer normalization. The persistent state is the RNN’s hidden state h as well as the memory M concatenated from previous memories. Bake at 200◦F for 16 to 20 hours in a desktop sized oven.
    The attention mechanism within the SHA-RNN is highly computationally efficient. The only matrix multiplication acts on the query. The A block represents scaled dot product attention, a vector-vector operation. The operators {qs, ks, vs} are vectorvector multiplications and thus have minimal overhead. We use a sigmoid to produce {qs, ks}. For vs see Section 6.4.
    Bits Per Character (BPC) onenwik8. The single attention SHA-LSTM has an attention head on the second last layer and hadbatch size 16 due to lower memory use. Directly comparing the head count for LSTM models and Transformer models obviously doesn’tmake sense but neither does comparing zero-headed LSTMs against bajillion headed models and then declaring an entire species dead.
    Introducing the SHA-RNN :) - Read alternative history as a research genre - Learn of the terrifying tokenization attack that leaves language models perplexed - Get near SotA results on enwik8 in hours on a lone GPU No Sesame Street or Transformers allowed.
    2 korisnika označavaju da im se sviđa
    Poništi
  12. 24. stu 2019.

    This is invaluable. Will go through this over the week.

    I wrote an 8k word doc on machine learning systems design. This covers: 1. Project setup 2. Data pipeline 3. Training & debugging 4. Serving with case studies, resources, and 27 exercises. This is the 1st draft so feedback is much needed. Thank you!
    1 proslijeđeni tweet 4 korisnika označavaju da im se sviđa
    Poništi
  13. proslijedio/la je Tweet
    21. stu 2019.

    To help developers get started with PyTorch, we’re making the 'Deep Learning with PyTorch' book, written by Luca Antiga and Eli Stevens, available for free to the community:

    1.045 proslijeđenih tweetova 2.885 korisnika označava da im se sviđa
    Poništi
  14. 19. stu 2019.

    ... Well. I'm receiving my Stadia today, hope it's not as bad as it looks here. If it is, hope it gets better eventually.

    here is my Google Stadia review in one GIF. this is on The Washington Post's Gigabit Ethernet last week.
    Prikaži ovu nit
    1 reply
    Poništi
  15. proslijedio/la je Tweet
    18. stu 2019.

    *New paper* RandAugment: a new data augmentation. Better & simpler than AutoAugment. Main idea is to select transformations at random, and tune their magnitude. It achieves 85.0% top-1 on ImageNet. Paper: Code:

    5 replies 169 proslijeđenih tweetova 622 korisnika označavaju da im se sviđa
    Poništi
  16. 12. stu 2019.

    Oh but the amount of compute and data. Oh boy.

    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  17. 12. stu 2019.

    A semi-simple method that I will probably try soon.

    Want to improve accuracy and robustness of your model? Use unlabeled data! Our new work uses self-training on unlabeled data to achieve 87.4% top-1 on ImageNet, 1% better than SOTA. Huge gains are seen on harder benchmarks (ImageNet-A, C and P). Link:
    Prikaži ovu nit
    1 reply 1 proslijeđeni tweet 2 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. 11. stu 2019.

    Wow. That's gonna increase my use of colab even more at this point.

    ✨📒 now has P100s! 🙌 (Reddit thread: )
    4 proslijeđena tweeta 25 korisnika označava da im se sviđa
    Poništi
  19. proslijedio/la je Tweet
    9. stu 2019.

    1/ A friend of mine pointed me to this article with a bunch of really cool images from Japan in 1908. So naturally I ran them through my new DeOldify model. Enjoy!

    20 replies 180 proslijeđenih tweetova 563 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  20. proslijedio/la je Tweet
    4. stu 2019.

    Computing Receptive Fields of Convolutional Neural Networks -- A new Distill article by André Araujo, Wade Norris, and Jack Sim.

    105 proslijeđenih tweetova 359 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  21. proslijedio/la je Tweet
    30. lis 2019.

    Some people asked about resources/tips on debugging machine learning models. Here you go :-) 1. 2. 3. 4. 5. 6.

    I saw a guy debugging his model today. No idpb. No unittest. No visualization. No disabling regularization. He just sat there staring at every line of code and cursing TensorFlow. Like every machine learning researcher I know.
    14 replies 697 proslijeđenih tweetova 2.228 korisnika označava da im se sviđa
    Poništi

@HaseoX94 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:

    ·