Ege Ozgirin

@egeozin

Machine Learning Scientist, Research Affiliate BCS. EECS MS from .

Vrijeme pridruživanja: travanj 2011.

Tweetovi

Blokirali ste korisnika/cu @egeozin

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

  1. proslijedio/la je Tweet
    2. sij

    Brains are amazing. Our lab demonstrates that single human layer 2/3 neurons can compute the XOR operation. Never seen before in any neuron in any other species. Out now in . Congrats Albert, Tim  & CO

    Poništi
  2. proslijedio/la je Tweet

    My new paper is out! We show a framework in which we can both derive and gradient penalized ! We also show how to make better gradient penalties!

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet

    This work, by the brilliant to me was possibly the most unexpected result I have seen in a very long time. Many ReLU nets can be almost entirely reconstructed (~full weight matrix, architecture) from measuring the output as a function of the inputs.

    Poništi
  4. proslijedio/la je Tweet

    The Tolman-Eichenbaum Machine: Unifying space and relational memory through generalisation in the hippocampal formation

    Poništi
  5. proslijedio/la je Tweet
    3. lip 2019.

    Big hierarchical VQ-VAEs with autoregressive priors do amazing things. Awesome work from :

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

    If you want to do research on instruction following and/or language grounding, consider using our BabyAI platform: 10^19 synthetic instructions, 19 levels of varying difficulty. Work done by with the help of .

    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet

    A common misconception is that the risk of overfitting increases with the number of parameters in the model. In reality, a single parameter suffices to fit most datasets: Implementation available at:

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    29. tra 2019.

    Out now! The primate ventral stream utilizes recurrent computations to identify objects in visual images. (SharedIt link: )

    Poništi
  9. proslijedio/la je Tweet
    11. tra 2019.

    The , chock full of modulatory nuclei, reward processing centers, and traversing axons. This image contains the VTA, LC, Raphe, superior and inferior colliculi, substantia nigra, red nucleus...its got it all baby!

    Poništi
  10. proslijedio/la je Tweet
    10. tra 2019.

    Neat new results on unsupervised visual learning from my student

    Poništi
  11. proslijedio/la je Tweet
    10. tra 2019.

    Her name is Katie Bouman, an MIT graduate. 3 years ago she led the creation of a new algorithm to produce the first-ever image of a black hole we are seeing today.

    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    22. ožu 2019.

    Interested in unsupervised object decomposition & representation learning? We're excited to share two new approaches: MONet, which uses sequential decomposition & more recently IODINE, which uses iterative refinement MONet: IODINE:

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet
    20. ožu 2019.

    Our researchers and collaborated with associate professor Ila Fiete on a new paper, titled “Flexible Representation and Memory of Higher-Dimensional Cognitive Variables with Grid Cells.”

    Poništi
  14. proslijedio/la je Tweet
    13. ožu 2019.

    Lol Bruno Olshausen

    Poništi
  15. proslijedio/la je Tweet

    Behold on the right: the missing panel in textbook illustrations of overfitting. Overly simple model can’t fit the data. Intermediate-complexity model fits ok. Complex model overfits. Super-complex model fits best of all. (Low-norm fits minimizing squared error.)

    Prikaži ovu nit
    Poništi
  16. proslijedio/la je Tweet

    Heterogeneity among pyramidal cells of the — a new Review by and

    Poništi
  17. proslijedio/la je Tweet
    8. velj 2019.
    Poništi
  18. proslijedio/la je Tweet
    6. velj 2019.

    Neural networks seem to use a puzzlingly simple strategy to classify images (work accepted at ICLR 2019 and liked by ;-)). Digest @ 1/8

    Prikaži ovu nit
    Poništi
  19. 7. sij 2019.

    A very good paper from the designers of the model that includes the overview of the method as well as the story:

    Prikaži ovu nit
    Poništi
  20. 7. sij 2019.

    Flint crisis --> ML model does reasonable (acc ~70%) in detecting lead pipes --> contractor changes --> priorities change, communication between different teams fail--> model is ignored (acc goes to 15 %) --> New contractor decides to go back to the model

    Prikaži ovu nit
    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:

    ·