--- Log opened Tue Jul 20 00:00:11 2021 00:26 < archels_> I need to give a presentation about "links between neuroscience and AI", any recommendations for papers/material? (much appreciated!) 01:17 -!- SDr6 is now known as SDr 01:17 -!- SDr [~SDr@li1189-192.members.linode.com] has quit [Changing host] 01:17 -!- SDr [~SDr@user/sdr] has joined #hplusroadmap 02:27 < maaku> archels_: that will be a very short presentation 02:36 < archels_> what makes you say so? 02:36 < archels_> so far I have things like why do neurons spike; temporal difference learning/dopamine; attention mechanisms; backprop/gradient descent and biological plausible variants 02:52 < archels_> may or may not be bothered with Jeff Hawkin's "thousand brains theory" 05:06 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:6574:650b:abe1:d53c] has joined #hplusroadmap 07:30 < archels_> .title https://www.frontiersin.org/articles/10.3389/fncom.2016.00094/full 07:30 < saxo> Frontiers | Toward an Integration of Deep Learning and Neuroscience | Frontiers in Computational Neuroscience 09:35 -!- andytosh1 is now known as andytoshi 09:45 -!- L29Ah [~L29Ah@user/l29ah] has quit [Quit: Gateway shutdown] 09:49 -!- L29Ah [~L29Ah@user/l29ah] has joined #hplusroadmap 13:31 < maaku> archels_: AI techniques are at best loooooosly derived from armchair theories about how the brain works 13:31 < maaku> the brain does not use backpropagation 13:31 < maaku> threshold models for neural nets have little to do with actual neuron spiking 13:32 < maaku> the internal structure of neurons are not modeled at all in ANNs, but are critically important to neuron growth and synnapse strengthening 13:33 < maaku> if you squint super hard and make your eyes blurry, then I guess the general shape of artificial neural networks kinda sortof looks like actual neurons, in an abstract way 13:34 < maaku> but otherwise the term "neural net" is just a holdover from a time when we knew next to nothing about how the brain worked, and *even then* ANNs were known to diverge from actual neuroscience 13:37 < maaku> Someone who was trained in the 1950's understanding of brains was wondering how neurons can compute, and in the process stumbled upon how a weighted network with threshold signalling can approximate functions and learn. 13:37 < maaku> because they were thinking about neurons when they invented it, they called it a neural net. and we've dealt with the legacy of that terrible nomenclature ever since :( 13:38 < fenn> it seems like quite a coincidence 13:38 < fenn> are you sure you're not wrong? 13:38 * fenn pokes the bear with a stick 13:39 < maaku> what seems like a coincidence? 13:40 < fenn> that someone thinking about how neurons work just happened upon an algorithm that works really well for doing the sort of applications that neurons typically work well for 13:40 < maaku> yes but they happened upon that algorithm by making what was already known at that time to be nonbiological assumptions 13:40 < fenn> birds, airplane wings 13:42 < fenn> if you squint hard they kind look the same in an abstract way 13:42 < maaku> I just skimmed the article, but this seems like a decent summary of the differences, most of which were known at the time perceptetrons were invented: https://towardsdatascience.com/the-differences-between-artificial-and-biological-neural-networks-a8b46db828b7 13:43 < maaku> *https://towardsdatascience.com/the-differences-between-artificial-and-biological-neural-networks-a8b46db828b7 13:43 < fenn> actual living neurons have to deal with this whole other problem of building themselves and physically connecting to other neurons 13:43 < maaku> ugh *perceptrons 13:44 < maaku> fenn: great example! the aerodynamics of bird wings and powered airplanes have zilch to do with each other 13:44 < fenn> uh huh 13:44 < maaku> the mechanism by which birds fly is entirely unrelated to the mechanims by which a Cessna or 747 flys 13:44 < fenn> why did you say "powered airplane" 13:45 < maaku> because a bird in flight often does resemble a purpose-built mechanical glider 13:46 < maaku> although birds also have and rely upon an "air pumping" trick when they flap their wings that we have been unable to reproduce in a practical way 13:48 < maaku> anytime you see birds actually flapping their wings, they're making little air vorticies from which they derive their lift 13:49 < maaku> rather than pushing on the air in front of them at an angle, like all airplances since the Wright flyer have done. 13:53 < fenn> http://i.imgur.com/SBDFQWf.jpeg looks the same as airplane wing vortices to me 13:54 < fenn> insect wings are different and i don't understand them 14:01 < maaku> yes the vortices are similar, but what I'm saying is the mechanism by which they are generated is different 14:02 < maaku> a physically moving wing pump motion, vs a static cross section presented to fast moving air 14:02 < fenn> a relative motion either way 14:06 < fenn> anyway it seems silly to complain about calling them both "wings" because birds were there first 14:07 -!- pasky [~pasky@nikam.ms.mff.cuni.cz] has quit [Remote host closed the connection] 14:08 < fenn> if you actually look at a bird wing, there are many different types, and a wide variety of features that we don't see at all on airplane wings 14:09 < fenn> there are also many types of airplane wings that don't exist at all on birds, such as delta wings 14:11 < fenn> it is easier to study airplane wings, and the general laws that such study informs can teach us about how bird wings work as well 14:14 -!- Croran [~quassel@2601:601:1880:7780:9c65:2f87:e3fa:24ba] has joined #hplusroadmap 15:09 < Croran> kanzure: Hi. lol thanks for noticing me? 15:49 < jrayhawk> archels_: https://www.infoq.com/presentations/models-cognition-brain/ was an okay mostly-non-technical presentation 15:58 < jrayhawk> archels_: the various cambridge bayesian neurochemistry papers are probably also worthwhile; rebecca lawson, paul c fletcher, etc. 16:01 < jrayhawk> https://www.researchgate.net/profile/Jakob-Hohwy/publication/315598050_Bayesian_Approaches_to_Autism_Towards_Volatility_Action_and_Behavior/links/58db9c5baca2725c47162b45/Bayesian-Approaches-to-Autism-Towards-Volatility-Action-and-Behavior.pdf https://link.springer.com/article/10.1007/s00213-009-1561-0 18:44 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:6574:650b:abe1:d53c] has quit [Quit: Leaving] 20:22 -!- Codaraxis_ [~Codaraxis@ip68-5-90-227.oc.oc.cox.net] has joined #hplusroadmap 20:26 -!- Codaraxis [~Codaraxis@user/codaraxis] has quit [Ping timeout: 268 seconds] 21:42 -!- Codaraxis__ [~Codaraxis@193.138.218.171] has joined #hplusroadmap 21:45 -!- Codaraxis_ [~Codaraxis@ip68-5-90-227.oc.oc.cox.net] has quit [Ping timeout: 252 seconds] 22:32 -!- faceface [~faceface@user/faceface] has quit [Remote host closed the connection] --- Log closed Wed Jul 21 00:00:12 2021