Twitter | Pretraživanje | |
Ege Ozgirin
Machine Learning Scientist, Research Affiliate BCS. EECS MS from .
35
Tweetovi
1.103
Pratim
90
Osobe koje vas prate
Tweetovi
Ege Ozgirin proslijedio/la je tweet
Jaan Aru 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
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Alexia Jolicoeur-Martineau 16. lis
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!
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
KordingLab 👨‍💻🧠∇🔬📈,🏋️‍♂️⛷️🏂🛹🕺⛰️☕🦖 3. lis
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.
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
bioRxiv Neuroscience 16. ruj
The Tolman-Eichenbaum Machine: Unifying space and relational memory through generalisation in the hippocampal formation
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Ben Poole 4. lip
Big hierarchical VQ-VAEs with autoregressive priors do amazing things. Awesome work from :
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Dzmitry Bahdanau 11. svi
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 .
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Marcos López de Prado 1. svi
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:
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Kohitij Kar 29. tra
Out now! The primate ventral stream utilizes recurrent computations to identify objects in visual images. (SharedIt link: )
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Greg Dunn 12. tra
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!
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Daniel Yamins 10. tra
Neat new results on unsupervised visual learning from my student
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Jen Zhu 10. tra
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.
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
DeepMind 22. ožu
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:
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Numenta 🧠 20. ožu
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.”
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Kyle Cranmer 13. ožu
Lol Bruno Olshausen
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Kriegeskorte Lab 26. velj
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.)
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Nature Rev Neurosci 19. velj
Heterogeneity among pyramidal cells of the — a new Review by and
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Michael Levin 8. velj
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin proslijedio/la je tweet
Wieland Brendel 6. velj
Neural networks seem to use a puzzlingly simple strategy to classify images (work accepted at ICLR 2019 and liked by ;-)). Digest @ 1/8
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin 7. sij 2019.
Odgovor korisniku/ci @egeozin
A very good paper from the designers of the model that includes the overview of the method as well as the story:
Reply Retweet Označi sa "sviđa mi se"
Ege Ozgirin 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
Reply Retweet Označi sa "sviđa mi se"