Nice! I never quite know if it is worth turning blog posts into papers, but this one is great. Random q: you say in 2 that conv bundles are retinotopic... is that true? The output is a full field activation map, so neurons have limited fov, but input at next layer is full field?
My blogpost on how & why we use convolutional neural networks as a model of the visual system is probably the most read thing I've ever written and it's now been expanded & updated into a proper review article, complete with 136 references & 5 new figures! https://arxiv.org/abs/2001.07092 pic.twitter.com/xRqYKptTsF
-
-
-
I guess I'm thinking from a clinical perspective. If you damaged the "upper left" neurons in a CNN layer, you wouldn't get visual field blindness, you would only lose responsiveness to certain visual features.
- Još 6 drugih odgovora
Novi razgovor -
-
-
Great paper. Thanks. I am curious about what do you think of microsaccades playing a role in visual attention as suggested by a recent study: https://www.cell.com/neuron/pdfExtended/S0896-6273(18)30468-9 …
-
Hm I'd consider this connected to the "pre-motor" theory of attention that links saccade planning to spatial attention (theres evidence for and against this). The idea that top-down activity (e.g. in FEF) both modulates activity in V4/IT & causes microsaccades seems reasonable
Kraj razgovora
Novi razgovor -
-
-
Curious if you have or will submit this to a journal (so I know how to cite it down the track)?
-
Yes, it's been accepted at journal of cognitive neuroscience. Not sure exactly when it is appearing yet.
- Još 1 odgovor
Novi razgovor -
-
-
fantastic &
@scidrawio Understanding both the ‘how’ & ‘why’ of biological vision. r/s b/w image encoding & memorability, incl. foveation & saccading for image classification. H&W>Neocognitiron>HMAX>Alexnet, a shift towards models that actually do something, forced new questions.pic.twitter.com/pqtkQPonqA
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
This looks great, thanks!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Great work! Beautiful figures
!Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I'm not going to mention the fact that spurious frequency based correlations which the weights in each filter represent are not what the brain represent, because I've talked about that before. But I think […]
-
that according to the thousand brains theory, each cortical column is essentially computing the same thing. This is not what CNNs do, filters getting more complex as we progress thru the layers as they build on more complex earlier filters+input convolutions.
@rhyolight
Kraj razgovora
Novi razgovor -
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.