4 ms·

I am not surprised at all how many applications you can get from insect-brain sized networks, since they only solve one specific purpose. If the same network co

by ma2rten 10y ago

I am not surprised at all how many applications you can get from insect-brain sized networks, since they only solve one specific purpose. If the same network could solve all those problems that would be surprising. I also think neural networks could save alot of parameters if they use sparse connections. That just doesn't happen to work well with our current hardware.

gwern 10y ago
> I am not surprised at all how many applications you can get from insect-brain sized networks, since they only solve one specific purpose. If the same network could solve all those problems that would be surprising. Transfer learning demonstrates that the 'specialization' objection is moot. The CNNs are learning large parts of what goes into general-purpose vision, not super-narrow specific things, otherwise it would be impossible to take a classification CNN and reuse only a tweaked version for image captioning, localization, tagging, segmentation etc and find the intermediate layers learning semantically meaningful things.