We had a NIPS workshop paper a few years ago showing how modulators can repurpose neural computing quite well. Of course, the NIPS process stripped away all the interesting neuro. But we had some great, still unpublished, results with distal dendrites biasing hippocampus dynamics
- Should neuroscientists stop using models where the neural network was trained to perform only one task? - How many different tasks does a network need to be trained on to be similar to a [fly/fish/mouse/aadvark/human] brain, and how does the task composition affect results?https://twitter.com/jaaanaru/status/1084855170700374022 …
-
-
-
Do you have more info on the workshop somewhere??
-
This workshop. https://nips.cc/Conferences/2017/Schedule?showEvent=8765 … It turned out that cognitive did not really mean neural in most cases here, but was more brain inspired than most of the rest of the conference.
-
looks interesting. We need to meet again some time soon! It's been ages since we've talked, hope we overlap at [cosyne, something], I'd love to pick your brain about this
End of conversation
New conversation -
-
-
excellent questions ::chefkiss::
-
Someone give me a lab!
End of conversation
New conversation -
-
-
what’s a task?
-
(not being flip, genuinely unsure where one “task” ends and another “task” begins)
-
yeah idk either. Sometimes I see object classification as multiple "tasks". Maybe that is appropriate in visual domain but not as appropriate more broadly?
End of conversation
New conversation -
-
-
I'd say it depends! If you're looking for general principles of network-based computation for a certain task, it's probably fine to just train a network on that task.
-
I guess it depends if general principles really...generalize to networks trained on multiple tasks?
-
My guess is you'd see a line attractor appear for evidence accumulation/decision-making tasks no matter what else a network is trained to do
-
Come see my poster on this @
#COSYNE19 !
End of conversation
New conversation -
-
-
I'd not make quite such a sweeping statement as that, but I think if we keep discovering that there are big differences between networks trained for a single task and for multiple tasks then doing so will be more and more important. Similar to using natural vs artificial stimuli.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.