--- Log opened Tue Nov 09 00:00:27 2021
00:10 -!- juri_ [~juri@178.63.35.222] has quit [Ping timeout: 268 seconds]
00:11 -!- juri_ [~juri@178.63.35.222] has joined #hplusroadmap
00:44 -!- Malvolio [~Malvolio@user/malvolio] has quit [Ping timeout: 264 seconds]
01:16 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 250 seconds]
01:44 -!- Malvolio [~Malvolio@user/malvolio] has joined #hplusroadmap
05:14 < docl> .wik Clay
05:14 < saxo> "Clay is a type of fine-grained natural soil material containing clay minerals. Clays develop plasticity when wet, due to a molecular film of water surrounding the clay particles, but become hard, brittle and non–plastic upon drying or firing." - https://en.wikipedia.org/wiki/Clay
05:18 < docl> .wik Clay mineral
05:18 < saxo> "Clay minerals are hydrous aluminium phyllosilicates, sometimes with variable amounts of iron, magnesium, alkali metals, alkaline earths, and other cations found on or near some planetary surfaces. /  Clay minerals form in the presence of water and have been important to [...]" - https://en.wikipedia.org/wiki/Clay_mineral
05:24 < fenn> ok i was wrong about the formation mechanism, i think?
05:29 < docl> yeah I think it's just generally water erosion of silicate rocks. but the phyllosilicates that form the sheets are all hydrated, so you wouldn't find them naturally on the moon
05:29 < docl> .wik Silicate_mineral#Phyllosilicates
05:29 < saxo> "Silicate minerals are rock-forming minerals made up of silicate groups. They are the largest and most important class of minerals and make up approximately 90 percent of Earth's crust. /  In mineralogy, silica (silicon dioxide) SiO2 is usually considered a silicate mineral.  [...]" - https://en.wikipedia.org/wiki/Silicate_mineral#Phyllosilicates
05:30 < docl> "
05:30 < docl> Phyllosilicates (from Greek φύλλον phýllon 'leaf'), or sheet silicates, form parallel sheets of silicate tetrahedra with Si
05:30 < docl> 2O
05:30 < docl> 5 or a 2:5 ratio. The Nickel–Strunz classification is 09.E. All phyllosilicate minerals are hydrated, with either water or hydroxyl groups attached. "
05:30 < fenn> says it comes from chemical weathering of feldspar, so i'm trying to figure out what a feldspar is, and failing
05:31 < docl> .wik Feldspar
05:31 < saxo> "Feldspars are a group of rock-forming aluminium tectosilicate minerals, containing sodium, calcium, potassium or barium. The most common members of the feldspar group are the plagioclase (sodium-calcium) feldspars and the alkali (potassium-sodium) feldspars." - https://en.wikipedia.org/wiki/Feldspar
05:33 < docl> basically aluminum silicates... I think gravitational sorting during earth's formation is the root cause of having lots of those in the crust
05:33 < docl> stony asteroids should be pretty similar
05:38 < fenn> the wikipedia redirects seem to be getting worse in general
05:38 < fenn> what causes the sheets to form in phyllosilicates?
05:40 < fenn> i guess the empty gap with the K ion is why... i guess. nothing explicitly says so https://en.wikipedia.org/wiki/File:Illstruc.jpg
06:00 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::f2f0] has joined #hplusroadmap
06:10 < docl> .title https://www.tandfonline.com/doi/full/10.1080/24749508.2017.1361128#_i10
06:11 < docl> "Clay minerals can be described very simply by the stacking of two kinds of layers: 1:1 layers and 2:1 layers. They are layered by silicate in which each layer in the structure in reality consists of two sublayers. The sublayer consists of octahedral coordinates and structural water in the form of hydroxyl groups."
06:27 < docl> there's a picture of kaolinite where Al2O6 octahedra are stacked against SiO4 tetrahedra. https://www.tandfonline.com/na101/home/literatum/publisher/tandf/journals/content/tgel20/2017/tgel20.v001.i03/24749508.2017.1361128/20170914/images/medium/tgel_a_1361128_f0001_oc.gif
06:28 < fenn> but those layers are only a few atoms wide
06:28 < fenn> surely that's not related to large scale repeating structures?
06:30 < docl> well it's going to give the rock flaky mechanical properties
06:33 < docl> which I think is why the dust sized particles take on the shape of sheets -- strong planes with weak dividing points
08:05 < muurkha> fenn: it depends on what you mean by "large scale".  clay crystals can be hundreds of microns wide
08:05 < fenn> many orders of magnitude bigger than an atom
08:07 < muurkha> as I understand it, clays are typically formed by oxidation of other aluminum silicates
08:09 < muurkha> so the large-scale order in them might derive from large-scale order that's present in the unweathered mineral, or it might just be the energetically favorable crystal structure at the low temperatures where clays typically form
08:11 < muurkha> the key point about clays is not that they're weak but that they form particles with a very high aspect ratio, can be plasticized by water because those particles are hydrophilic, have very little elasticity, and can be sintered into much harder and stronger materials
08:14 < muurkha> I think kaolinite is a 1:1 Al:Si molar ratio while feldspar is 1:3, a property which makes it much easier for the silicon in feldspar to form a continuous network in all three dimensions
08:16 < muurkha> but calcic plagioclase feldspars are closer to kaolin in their molar ratio
08:19 < muurkha> Montmorillonite is 1:2, with (divalent!) magnesium substituting for some of the sodium
08:22 < muurkha> anyway, yes, the large-scale repeating structure of clay crystals, like the large-scale repeating structures of other crystals, are very strongly related to the atomic structure
08:25 < muurkha> I said "oxidation of other aluminum silicates" but that's wrong; things like mullite and feldspar are already totally oxidized.  clays do often incorporate other elements like potassium, sodium, and calcium, but so do feldspars.  the clays, though, are full of hydroxyls
08:26 < muurkha> the hydroxyls are volatilized during firing well before the sintering stage
10:11 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
11:51 -!- mrdata_ [~mrdata@135-23-182-185.cpe.pppoe.ca] has joined #hplusroadmap
11:51 -!- mrdata_ [~mrdata@135-23-182-185.cpe.pppoe.ca] has quit [Changing host]
11:51 -!- mrdata_ [~mrdata@user/mrdata] has joined #hplusroadmap
11:55 -!- otoburb_ [~otoburb@user/otoburb] has joined #hplusroadmap
11:58 -!- mrdata [~mrdata@user/mrdata] has quit [Killed (NickServ (GHOST command used by mrdata_))]
11:58 -!- mrdata_ is now known as mrdata
12:00 -!- Netsplit *.net <-> *.split quits: otoburb
12:09 -!- otoburb_ is now known as otoburb
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=16b4cc28 Michael Folkson: Add October 4th c-lightning developer call >> http://diyhpl.us/diyhpluswiki/transcripts/c-lightning/2021-10-04-developer-call.md
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=ada40035 Michael Folkson: Merge pull request #234 from michaelfolkson/c-lightning-2021-10-04 >> http://diyhpl.us/diyhpluswiki/
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=af330fa2 Michael Folkson: Add October 18th c-lightning call >> http://diyhpl.us/diyhpluswiki/transcripts/c-lightning/2021-10-18-developer-call.md
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=891ab155 Michael Folkson: Merge pull request #235 from michaelfolkson/c-lightning-2021-10-18 >> http://diyhpl.us/diyhpluswiki/
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=2e65a761 Michael Folkson: Add November 1st c-lightning call >> http://diyhpl.us/diyhpluswiki/transcripts/c-lightning/2021-11-01-developer-call.md
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=0404661f Michael Folkson: Merge pull request #236 from michaelfolkson/c-lightning-2021-11-01 >> http://diyhpl.us/diyhpluswiki/
12:16 < gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=126803e6 Bryan Bishop: Merge remote-tracking branch 'github/master' into master >> http://diyhpl.us/diyhpluswiki/
12:30 -!- spaceangel [~spaceange@ip-89-176-181-220.net.upcbroadband.cz] has joined #hplusroadmap
13:58 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::f2f0] has quit [Quit: Leaving]
15:13 -!- spaceangel [~spaceange@ip-89-176-181-220.net.upcbroadband.cz] has quit [Remote host closed the connection]
16:35 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 246 seconds]
18:22 < kanzure> https://www.3dsystems.com/press-releases/3d-systems-announces-acquisition-volumetric-biotechnologies
18:22 < kanzure> jmil^
18:26 < kanzure> "galvo driving a DLP projection, so they can raster it across the build surface which is nonplanar"
19:23 < lsneff> I wonder if a stack of 10 or so wafer scale chips designed for very sparse spiking neural networks (akin to https://cerebras.net/chip/, but for SNNs) with a couple hundred TB of NAND flash soldered to each wafer would be enough to do human brain scale simulations
19:31 < lsneff> The cerebras chip draws 1.5 kW, but that’s because power is being distributed to the whole thing. The human brain is limited to roughly 1% firing sparsity at any given time, probably a lot less, so each wafer would only sip a few watts
19:35 < lsneff> I think storage is the major issue actually. It seems to me like we might be in a compute overhead.
19:37 < muurkha> lsneff: SNNs use less power?  how do you train them?
19:38 < muurkha> is there a regular continuous approximation you can run gradient descent on?
19:38 < lsneff> Some sort of STDP
19:38 < muurkha> (these are probably super basic questions, I know)
19:38 < muurkha> hmm, like Hebbian learning?  does that work?
19:39 < lsneff> To train a network this big, it really can’t be global optimization like backpropagation
19:39 < lsneff> Of a sort, it’s unclear exactly how the brain does it
19:39 < lsneff> But it seems that neurons usually only have local information anyway
19:39 < lsneff> Unsupervised
19:40 < lsneff> SNNs use significantly less power and less compute
19:40 < lsneff> But they tend to be a little less accurate
19:40 < muurkha> how do we know, if we don't know how to train them?
19:41 < muurkha> or is there a training setup that works in practice even if it isn't how the brain does it?
19:44 < lsneff> Yeah, the various unsupervised training methods work
19:44 < lsneff> You can also convert feed forward ANNs to SNNs
19:55 < muurkha> oh really?  how does that work?
19:55 < muurkha> that sounds a lot more interesting than unsupervised training
19:55 < muurkha> do you know about avalanche transistors?
19:55 < lsneff> Ah, I think it’s much less interesting personality
19:56 < lsneff> You basically convert the rate-coding approximation that is ANNs to the equivalent parameters in an SNN
19:56 < lsneff> No, what’s an avalanche transistor?
19:57 < lsneff> The brain is larger unsupervised training with reward neurotransmitters, so that’s more interesting to me
19:58 < lsneff> The conversation of ann to snn results in an snn that’s less efficient than a bespoke trained snn unfortunately
19:59 < muurkha> well, if I want to control a CNC to move to a certain position, guide a search through an exponentially large search space, or distinguish my smell from someone else's, I can use supervised training, but not unsupervised training
19:59 < muurkha> how do you build SNN circuits?  maybe you have something better than avalanche transistors
20:00 < muurkha> or are you just simulating them?
20:02 < muurkha> if I want to upload my brain into silicon, supervised training can help me, unsupervised training can't
20:21 < lsneff> The brain is able to supervise itself at a very large scale, but individually neurons seem to learn using local data, aka mostly unsupervised
20:24 < lsneff> For supervised training, you have to know the answer already, which doesn’t really make sense at the neuronal level in brains.
20:34 < lsneff> It is surprising perhaps, but SNNs where neurons have STDP and some other rules will, if they’re architected right, learn to differentiate inputs
--- Log closed Wed Nov 10 00:00:28 2021