Timeline for How is it possible for brain neurons to learn if they don't do backpropagation?
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 12 at 17:17 | comment | added | Jiminy Cricket. | That's a lot of big questions. I can't answer, certainly not in the space of an answer here, much less in comments. The background reading listed in the link you posted as a comment would be worth a dive down the rabbit hole. | |
| Dec 12 at 16:43 | comment | added | yters | Does neuroscience have a theory of how the learning mechanism works? I don't mean a listing of all the biological components of a learning system, but how the components allow the brain to go from no knowledge to some knowledge. How are all the signals from various senses turned into concepts? How are the concepts stored? How are correct concepts distinguished from false concepts? How does the brain use these concepts to make predictions about the world? All of this is described in computer science, so I'm trying to understand how neuroscience has a different definition. | |
| Dec 12 at 15:59 | comment | added | Jiminy Cricket. | "How does neuroscience define learning [....]?" Not in a single way, it's multi-layered - all the way up from molecular levels through gene-expression changes, protein manufacture, long term potentiation, growth (and retraction) of dendrites etc.. Overall the definition would be the same - to do with laying down or modification of memory leading to behavioural change. | |
| Dec 12 at 15:16 | comment | added | yters | How does neuroscience define learning in a scientific way that is different than in the fields of AI and machine learning? At any rate, it seems the takeaway here is that insofar as computer science defines learning, the brain does not do anything like that at all. | |
| Dec 12 at 4:04 | comment | added | Jiminy Cricket. | Or to put it another way, wetware is very different from hardware. I'm aware of a great gulf between a programmer's understanding of AI, and a neuroscientist's understanding - AI is designed, the brain has evolved and it's activity is emergent. This signifies that the investigation of such things is not finished, but known to be much more complex than a set of algorithms and logical operations. There's a whole mess of neurotransmitters and hormones at play too. | |
| Dec 12 at 3:16 | comment | added | Bryan Krause♦ | They. Are. Not. The. Same. At all. | |
| Dec 12 at 3:00 | comment | added | yters | Here's a good related question: psychology.stackexchange.com/questions/26883/… | |
| Dec 12 at 2:58 | comment | added | yters | @BryanKrause Feedforward ANNs are massively parallel (see NVIDIA's stock boom), and variants are recurrent. However, recurrency dramatically impacts the learning rate in a negative way, which is why the transformer architecture was adopted instead of recurrent neural network. So, still it is very mysterious how the brain learns much more effectively than modern neural networks, yet modern neural networks rejected all the potential learning mechanisms we've identified in the brain as being much to inefficient to learn even in a massively parallel format. | |
| Dec 11 at 20:25 | comment | added | Bryan Krause♦ | The brain is massively parallel and recurrent, it doesn't make much sense to discuss algorithms for feedforward ANNs in the context of real brain circuitry. | |
| Dec 11 at 13:28 | comment | added | yters | @ArnonWeinberg thanks, the first question seems related. However, Hebbian learning is not a very good learning algorithm. It seems implausible the brain uses Hebbian learning, since the brain seems to learn things that even our best neural networks cannot learn yet, and SOTA neural networks use much, much more effective algorithms than Hebbian learning. | |
| Dec 10 at 18:57 | comment | added | Arnon Weinberg♦ | Possible duplicate of: How do neurons decide how to alter their output signals? Also see: Biological plausibility of RBMs. | |
| Dec 10 at 18:38 | history | asked | yters | CC BY-SA 4.0 |