| Tweetovi |
| Ege Ozgirin proslijedio/la je tweet | ||
|
Jaan Aru
@jaaanaru
|
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 @sciencemagazine. Congrats Albert, Tim @mattlark @YiotaPoirazi & CO science.sciencemag.org/content/367/64…
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Alexia Jolicoeur-Martineau
@jm_alexia
|
16. lis |
|
My new paper is out! We show a framework in which we can both derive #SVMs and gradient penalized #GANs! We also show how to make better gradient penalties!
ajolicoeur.wordpress.com/MaximumMarginG…
arxiv.org/abs/1910.06922 twitter.com/hardmaru/statu…
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
KordingLab 👨💻🧠∇🔬📈,🏋️♂️⛷️🏂🛹🕺⛰️☕🦖
@KordingLab
|
3. lis |
|
This work, by the brilliant @david_rolnick 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. twitter.com/david_rolnick/…
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
bioRxiv Neuroscience
@biorxiv_neursci
|
16. ruj |
|
The Tolman-Eichenbaum Machine: Unifying space and relational memory through generalisation in the hippocampal formation biorxiv.org/cgi/content/sh… #biorxiv_neursci
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Ben Poole
@poolio
|
4. lip |
|
Big hierarchical VQ-VAEs with autoregressive priors do amazing things. Awesome work from @catamorphist @avdnoord @OriolVinyalsML: arxiv.org/abs/1906.00446 pic.twitter.com/JpEbEJnXk4
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Dzmitry Bahdanau
@DBahdanau
|
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 @MILAMontreal with the help of @Element_AI. github.com/mila-iqia/baby…
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Marcos López de Prado
@lopezdeprado
|
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: arxiv.org/abs/1904.12320
Implementation available at: github.com/Ranlot/single-… pic.twitter.com/2gGlrSuRGj
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Kohitij Kar
@KohitijKar
|
29. tra |
|
Out now! The primate ventral stream utilizes recurrent computations to identify objects in visual images. (SharedIt link: rdcu.be/bzuF9) pic.twitter.com/ZX1qyApAJG
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Greg Dunn
@GDunnArt
|
12. tra |
|
The #midbrain, 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! buff.ly/2s2tYIs #art #neuro #brain #sciart pic.twitter.com/hebxPNc53o
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Daniel Yamins
@dyamins
|
10. tra |
|
Neat new results on unsupervised visual learning from my student @ChengxuZhuang arxiv.org/abs/1903.12355
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Jen Zhu
@jenzhuscott
|
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.
#BlackHole #EventHorizonTelescope pic.twitter.com/peZcLSjQmJ
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
DeepMind
@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: arxiv.org/abs/1901.11390
IODINE: arxiv.org/abs/1903.00450 pic.twitter.com/sB67lWlxBD
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Numenta 🧠
@Numenta
|
20. ožu |
|
Our researchers @MrcsLws and @MirkoKlukas collaborated with @MITBrainAndCog associate professor Ila Fiete on a new paper, titled “Flexible Representation and Memory of Higher-Dimensional Cognitive Variables with Grid Cells.” doi.org/10.1101/578641 pic.twitter.com/mzygTyB4hp
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Kyle Cranmer
@KyleCranmer
|
13. ožu |
|
Lol Bruno Olshausen pic.twitter.com/KpmA4ObsFI
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Kriegeskorte Lab
@KriegeskorteLab
|
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.) pic.twitter.com/HaVjpwh2kJ
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Nature Rev Neurosci
@NatRevNeurosci
|
19. velj |
|
Heterogeneity among pyramidal cells of the #hippocampus — a new Review by @MarkCembrowski and @nspruston go.nature.com/2SJ26si pic.twitter.com/GfNO1w6nZ3
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Michael Levin
@drmichaellevin
|
8. velj |
|
To help study brain-body plasticity (like jeb.biologists.org/content/216/6/…), put brains into simpler bodies that we already understand:
ncbi.nlm.nih.gov/pubmed/18002625
link.springer.com/chapter/10.100…
ieeexplore.ieee.org/stamp/stamp.js…
ncbi.nlm.nih.gov/pmc/articles/P…
|
||
|
|
||
| Ege Ozgirin proslijedio/la je tweet | ||
|
Wieland Brendel
@wielandbr
|
6. velj |
|
Neural networks seem to use a puzzlingly simple strategy to classify images (work accepted at ICLR 2019 and liked by @karpathy ;-)). Digest @ medium.com/bethgelab/neur… @MatthiasBethge @bethgelab @GaryMarcus 1/8
|
||
|
|
||
|
Ege Ozgirin
@egeozin
|
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:
arxiv.org/pdf/1806.10692…
|
||
|
|
||
|
Ege Ozgirin
@egeozin
|
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
theatlantic.com/technology/arc…
|
||
|
|
||