Greg Yang

@TheGregYang

Researcher at AI. Morgan Prize Honorable Mention 2018.

Bellevue, WA
decentdescent.org
Vrijeme pridruživanja: veljača 2019.
121 fotografija ili videozapis Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @TheGregYang

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

  1. Prikvačeni tweet
    5. pro 2019.

    1/ Why do wide, random neural networks form Gaussian processes, *regardless of architecture*? Let me give an overview in case you are too lazy to check out the paper or the code . The proof has two parts…

    8 replies 268 proslijeđenih tweetova 1.009 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    29. sij

    With projects ranging in focus from healthcare to gaming to server workload management, 2019-20 Microsoft AI residents are pushing forward real-world applications in artificial intelligence. Learn more about the program and apply for the class of 2020-21:

    29 proslijeđenih tweetova 41 korisnik označava da mu se sviđa
    Poništi
  3. proslijedio/la je Tweet
    29. sij

    Dr. leads the Deep Learning Group at Microsoft Research. Hear how recent advances in training data and infrastructure have allowed Dr. Gao’s team to work on applications for deep learning in language, vision, and video on the :

    1 reply 71 proslijeđeni tweet 247 korisnika označava da im se sviđa
    Poništi
  4. proslijedio/la je Tweet
    28. sij

    This is a great question that I've gotten periodically. Previously it would have taken too long to put something together, but using Neural Tangents () it's really easy and fast! Here is the reproduction in a colab:

    Can someone help me in reproducing the code for the paper - by . I'm interested in reproducing the Order Chaos Transition Graph for a Non-Linear Activation function as shown on the Figure 1 in the paper.
    18 proslijeđenih tweetova 103 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet
    25. sij

    With TensorWatch, researchers seek to help other researchers/engineers get much-needed info about the state of their ML systems. The open-source debugging and visualization tool works in Jupyter Notebook to perform key analysis tasks for models and data:

    79 proslijeđenih tweetova 157 korisnika označava da im se sviđa
    Poništi
  6. proslijedio/la je Tweet
    24. sij

    Research on the Neural Tangent Kernel (NTK) almost exclusively uses a non-standard neural network parameterization, where activations are divided by sqrt(width), and weights are initialized to have variance 1 rather than variance 1/width.

    23 proslijeđena tweeta 124 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    23. sij

    Super disgusted by people working on privacy-related technologies who knowingly misrepresent "their" technology to the employer or policymakers hoping for a faster or wider adoption. Such people make matters strictly worse for everyone (except possibly the company they work for).

    6 proslijeđenih tweetova 29 korisnika označava da im se sviđa
    Poništi
  8. 21. sij

    Joint work with these awesome algebraicists and geometers Justin Chen, Christopher Eur, and, Mengyuan Zhang!

    4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  9. 21. sij

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  10. 21. sij

    Learnability (VC dim) is a *topological property*, as I proved in for parity, conjunctions, poly threshold fctns. Now this extends to downward-closed classes, conjunction of parities, and k-CNFs, as well! Just how far does this go?

    44 proslijeđena tweeta 202 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  11. proslijedio/la je Tweet
    16. sij

    Text-based games provide a platform to train RL agents that generate goal-driven language. Jericho framework by & provides benchmarks for scaling RL to combinatorially sized language action spaces:

    48 proslijeđenih tweetova 164 korisnika označavaju da im se sviđa
    Poništi
  12. 13. sij

    Did u know? The NTK & GP kernels of an MLP on the sphere have the form K(x, y) = F(x.y/dim) for scalar F. Each (eigenvalue*multiplicity) of K goes to F^{(k)}(0)/k! as dim -> infty, over unif dist on the sphere (or the bool cube, also extends to Gaussian)

    20 proslijeđenih tweetova 54 korisnika označavaju da im se sviđa
    Poništi
  13. proslijedio/la je Tweet
    9. sij

    Microsoft AirSim, now on Unity, provides an open-source system for training autonomous systems. Faster and safer than training these systems in the real world, explore how AirSim utilizes multiple learning methods to create a realistic environment:

    5 replies 62 proslijeđena tweeta 179 korisnika označava da im se sviđa
    Poništi
  14. proslijedio/la je Tweet
    8. sij

    A few weeks ago, published some interesting research on how smiles can serve as feedback for machine learning. I don't think it got enough attention.

    15 proslijeđenih tweetova 15 korisnika označava da im se sviđa
    Poništi
  15. 7. sij

    Neural networks tend to Gaussian processes (GPs) as their widths tend to infinity --- now you can play with these GP kernels in ! Try out RNN-GP, GRU-GP, Transformer-GP, or Batchnorm-GP today! Repo: Colab Entry Point:

    111 proslijeđenih tweetova 487 korisnika označava da im se sviđa
    Poništi
  16. 27. pro 2019.

    It was a pleasure to visit the Physics department of Taiwan National University! What a beautiful place!

    talking about the Tensor Program for computing correlation functions in deep learning and the connection to physics
    2 proslijeđena tweeta 30 korisnika označava da im se sviđa
    Poništi
  17. 19. pro 2019.

    15/ The tensor program framework is very powerful and has many other consequences, such as the universality of Neural Tangent Kernels. If you would like to know more, check out

    10 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. 19. pro 2019.

    14/ Finally, we verify our theory is more and more accurate as the width of the network increases. Here we measure the relative frobenius norm of the empirical kernel to the infinite-width theoretical kernel, and we see that the deviation tends to 0 like 1/sqrt(width)

    4 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  19. 19. pro 2019.

    13/ These are examples of kernels corresponding to the infinite-width limit of other architectures: GRU, transformer, batchnorm.

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  20. 19. pro 2019.

    12/ Again, see for an outline of the proof.

    1/ Why do wide, random neural networks form Gaussian processes, *regardless of architecture*? Let me give an overview in case you are too lazy to check out the paper or the code . The proof has two parts…
    Prikaži ovu nit
    1 reply 2 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  21. 19. pro 2019.

    11B/ While at face value, this may seem limited, one can in fact express almost all modern architectures in this framework: resnet, transformer, LSTM, etc. Here are some simple examples of tensor programs expressing NN computation.

    1 reply 1 proslijeđeni tweet 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi

@TheGregYang 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:

    ·