| --- Log opened Tue Aug 30 00:00:40 2016 | ||
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has joined ##hplusroadmap | 01:08 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:6d46:22a3:1e49:3f9a] has joined ##hplusroadmap | 01:16 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:6d46:22a3:1e49:3f9a] has quit [Changing host] | 01:16 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 01:16 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has joined ##hplusroadmap | 01:55 | |
| -!- iaglium [~i@45.63.97.181] has quit [Ping timeout: 240 seconds] | 02:11 | |
| -!- iaglium [~i@45.63.97.181] has joined ##hplusroadmap | 02:11 | |
| -!- wrldpc1 [~ben@p597030-ipngn200504kyoto.kyoto.ocn.ne.jp] has joined ##hplusroadmap | 02:37 | |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has quit [Remote host closed the connection] | 02:38 | |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has joined ##hplusroadmap | 02:41 | |
| -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-thfutzafbbnmbzfq] has quit [Quit: Connection closed for inactivity] | 02:42 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 02:45 | |
| -!- Burnin8 is now known as Burninate | 02:48 | |
| -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-ahlabixzkuedikuf] has joined ##hplusroadmap | 02:49 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has joined ##hplusroadmap | 02:52 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has quit [Changing host] | 02:52 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 02:52 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has quit [Read error: Connection reset by peer] | 02:56 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has joined ##hplusroadmap | 02:57 | |
| -!- augur [~augur@2602:304:cdac:e260:4c99:e8f7:9f21:c0c8] has quit [Remote host closed the connection] | 02:57 | |
| -!- augur [~augur@76-218-206-38.lightspeed.sntcca.sbcglobal.net] has joined ##hplusroadmap | 02:59 | |
| -!- augur [~augur@76-218-206-38.lightspeed.sntcca.sbcglobal.net] has quit [Read error: Connection reset by peer] | 03:01 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has quit [Max SendQ exceeded] | 03:01 | |
| -!- augur [~augur@2602:304:cdac:e260:1867:85b6:26b4:6196] has joined ##hplusroadmap | 03:01 | |
| -!- augur [~augur@2602:304:cdac:e260:1867:85b6:26b4:6196] has quit [Remote host closed the connection] | 03:02 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has joined ##hplusroadmap | 03:02 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 03:10 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 03:10 | |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap | 03:16 | |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Quit: Leaving] | 03:26 | |
| -!- justan0theruser [~justanoth@unaffiliated/justanotheruser] has quit [Ping timeout: 244 seconds] | 03:27 | |
| -!- jtimon [~quassel@38.110.132.37.dynamic.jazztel.es] has joined ##hplusroadmap | 03:27 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 03:34 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has joined ##hplusroadmap | 03:35 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has quit [Changing host] | 03:35 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 03:35 | |
| -!- maaku [~quassel@173-228-107-141.dsl.static.fusionbroadband.com] has quit [Quit: No Ping reply in 180 seconds.] | 03:42 | |
| -!- wrldpc1 [~ben@p597030-ipngn200504kyoto.kyoto.ocn.ne.jp] has quit [Quit: wrldpc1] | 03:43 | |
| -!- maaku [~quassel@173-228-107-141.dsl.static.fusionbroadband.com] has joined ##hplusroadmap | 03:44 | |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has quit [Read error: No route to host] | 03:49 | |
| -!- ArturShaik [~ArturShai@185.66.252.104] has joined ##hplusroadmap | 04:03 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 04:04 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:4c13:28b3:2acd:43e9] has joined ##hplusroadmap | 04:04 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:4c13:28b3:2acd:43e9] has quit [Changing host] | 04:04 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 04:04 | |
| -!- ArturSha1 [~ArturShai@185.66.252.104] has quit [Ping timeout: 244 seconds] | 04:04 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 04:08 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has joined ##hplusroadmap | 04:08 | |
| -!- ebowden [~ebowden@CPE-101-180-249-107.lnse3.cha.bigpond.net.au] has quit [Changing host] | 04:08 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 04:08 | |
| -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-sdqhryvtwbucpztg] has joined ##hplusroadmap | 04:17 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 04:23 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 04:24 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 05:09 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:fd99:3e62:dd6d:9e51] has joined ##hplusroadmap | 05:09 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:fd99:3e62:dd6d:9e51] has quit [Changing host] | 05:09 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 05:09 | |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap | 05:10 | |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has joined ##hplusroadmap | 05:19 | |
| -!- wrldpc1 [~ben@122.31.0.30] has joined ##hplusroadmap | 05:35 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 06:17 | |
| kanzure | hrmph | 06:19 |
|---|---|---|
| JayDugger | Feh. | 06:20 |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Quit: Leaving] | 06:48 | |
| -!- oleo [~oleo@unaffiliated/oleo] has joined ##hplusroadmap | 07:31 | |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap | 07:46 | |
| -!- arthurl [~arthurl@38.117.157.141] has joined ##hplusroadmap | 07:47 | |
| arthurl | anyone here take melatonin on a regular schedule? | 07:47 |
| __mz_o | i had experimented it but it made me feel shitty in the morning | 07:49 |
| __mz_o | Maybe ask #reddit-nootropics | 07:49 |
| arthurl | cool thanks __mz_o - i'll ask in there | 07:50 |
| arthurl | i sit in front of a computer all day and have done so for many years (i'm a computer engineer) and it's always hard for me to fall asleep pretty sure my body doesn't regulate melatonin production the way it should | 07:51 |
| maaku | Turn off blue light in your monitor. | 07:59 |
| kanzure | and stop forcing yourself to sleep | 07:59 |
| __mz_o | yea maaku said. f.lux is the prog for windoze | 08:00 |
| __mz_o | dont know the linux alternative | 08:01 |
| CaptHindsight | how will you know if windoze has crashed? | 08:01 |
| arthurl | kanzure what do you mean? i don't really force myself to sleep i try to exert all my energy during the day so i'm actually tired by 11/12 at night but that's rarely the case | 08:02 |
| __mz_o | you have to assume its perpetually crashing | 08:02 |
| kanzure | "so i'm actually tired by 11/12 at night but that's rarely the case" so which is it | 08:02 |
| abetusk | arthurl, I use it occasionally when I can't get to sleep but not on a regular basis | 08:03 |
| CaptHindsight | have kids or join the army, learn how little sleep you really need | 08:03 |
| arthurl | i said i try to exert my energy so i'm tired by that time- but that's rarely the case- meaning i'm usually still not tired by then | 08:03 |
| __mz_o | arthurl: have you tried excercise in the morning/afternoon? | 08:03 |
| kanzure | arthurl: then stop trying to sleep at 11/12. | 08:03 |
| arthurl | __mz_o i'd love to try that- i'm just terrible at waking up early in the morning- and i have to be in the office by 9:30/10 - but it's something i've honestly been really wanting to try | 08:04 |
| arthurl | kanzure my point is i need a solid 8 hours or i'm tired that's why i try to sleep at that time | 08:04 |
| kanzure | get a better job; engineers are in demand. tell HR you need to adjust a sleep schedule, make it equitable for them. | 08:04 |
| arthurl | lol | 08:04 |
| CaptHindsight | 8-6 jobs are tough for people that have a 26 hour day | 08:06 |
| arthurl | CaptHindsight :) good point wrt kids/army | 08:06 |
| __mz_o | arthurl: id recommend that. I suck at waking up to but an afternoon workout will use some energy | 08:06 |
| arthurl | i think a big issue for me is the fact that i try to go to the gym after work (usually 8-9pm) and will then have dinner after that | 08:07 |
| kanzure | if you are serious about forcing yourself to sleep, i suggest an iv line drip | 08:07 |
| arthurl | and the gym/late dinner gives me energy | 08:07 |
| __mz_o | try going to the gym right after work when youre still in "work" mode | 08:08 |
| __mz_o | kanzure: iv line drip what? | 08:08 |
| arthurl | kanzure i'm very serious- my screwed up sleep cycle is quite annoying- hate feeling tired mid day when i eat healthy and got 8 hours of sleep night prior- doesn't make sense | 08:08 |
| kanzure | if you want to force the biology to be different, then anesthesia seems prudent | 08:08 |
| arthurl | __mz_o also great advice- i'm thinking going earlier will help for sure | 08:08 |
| kanzure | also, switch to steroids and skip the gym | 08:09 |
| __mz_o | lol then he really wont be able to sleep | 08:09 |
| CaptHindsight | have you tried sex near bedtime? | 08:09 |
| arthurl | CaptHindsight sex definitely helps unfortunately it's not always on tap for me :) | 08:11 |
| kanzure | it's pretty cheap | 08:11 |
| arthurl | kanzure it is but i fear being set up and or STDs | 08:11 |
| __mz_o | lol "Officer I just needed help going to sleep" | 08:12 |
| arthurl | :) | 08:12 |
| kanzure | both a set up and STDs? terrible luck. | 08:12 |
| CaptHindsight | sleep therapist | 08:12 |
| __mz_o | push comes to shove you can always do ketamine | 08:12 |
| -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-sdqhryvtwbucpztg] has quit [Quit: Connection closed for inactivity] | 08:13 | |
| CaptHindsight | sounds like you need better friends if that is who you get setup with | 08:13 |
| arthurl | lmao | 08:13 |
| arthurl | __mz_o i think i'll pass on the ketamine :P | 08:14 |
| arthurl | have considered weed but don't want to become dependent | 08:14 |
| CaptHindsight | look for a job with a flex schedule or starts later, like bar tender | 08:15 |
| arthurl | my job is actually great i can be in by 10- usually leave my apt by 9 | 08:15 |
| arthurl | which is pretty late | 08:15 |
| arthurl | in comparison to most people i know | 08:15 |
| arthurl | but i also leave the office later, 7ish usually | 08:15 |
| arthurl | so not sure it's actually helping my problem | 08:16 |
| CaptHindsight | you might grow out of it, I used to rarely ever be up before noon | 08:16 |
| arthurl | i'm 30 years old- this sleep issue has been a thing for years now | 08:17 |
| CaptHindsight | most productive 6-12am | 08:17 |
| CaptHindsight | most creative after midnight | 08:17 |
| arthurl | what about some anti-narcolepsy drugs | 08:18 |
| arthurl | have researched modafinil etc never tried | 08:18 |
| CaptHindsight | by 40 you should be out of it | 08:18 |
| arthurl | so only 10 more years of suffering? | 08:19 |
| arthurl | i want my 30's to be productive | 08:19 |
| CaptHindsight | yeah, then it will become joint and back problems, thinning hair etc | 08:20 |
| CaptHindsight | things could be worse | 08:22 |
| arthurl | oh trust me i know- i'm a greatful person | 08:22 |
| arthurl | just trying to fix this sleep problem :) | 08:22 |
| CaptHindsight | but thats what you have to deal with if you choose working 9-5 (8-6) | 08:23 |
| arthurl | already have joint and back problems, along with the thinning hair :) | 08:23 |
| arthurl | CaptHindsight what's the alternative? working for myself yeah? | 08:23 |
| CaptHindsight | then you have nothing to worry about | 08:23 |
| CaptHindsight | I stopped working for other people at age 20 | 08:26 |
| arthurl | you're a smart man | 08:27 |
| arthurl | i tried doing my own thing as well (IT consulting) but i needed to build a team and find people i could trust- easier said than done | 08:27 |
| __mz_o | thats always the hardest part about starting something yourself | 08:28 |
| __mz_o | its all about who you know ;) | 08:28 |
| -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-hsbjjwavqvmywmde] has joined ##hplusroadmap | 08:30 | |
| CaptHindsight | know, location, skills | 08:31 |
| CaptHindsight | the docs in the US would blabber about hormones and rhythms | 08:32 |
| __mz_o | regarding sleep? | 08:33 |
| CaptHindsight | yeah | 08:33 |
| __mz_o | yea sleep is definitely a combination of factors including that | 08:35 |
| arthurl | i think another big factor for me is the fact that once i come home- i wind down by being on the computer reading stuff i like to read etc | 08:35 |
| arthurl | which stimulates my brain | 08:35 |
| CaptHindsight | suggest melatonin and possibly Diphenhydramine | 08:35 |
| arthurl | i already have melatonin- don't take it often but it definitely works when i decide to take it, usually ~2.5g | 08:37 |
| CaptHindsight | and if that doesn't work something more fun like Eszopiclone, Temazepam etc | 08:37 |
| arthurl | took some last night- was asleep by 11 slept like a baby | 08:37 |
| CaptHindsight | sounds like you have it worked out | 08:40 |
| arthurl | well i don't want to have to take melatonin every night- i'd think my body would then get used to it and adjust it's own melatonin production accordingly which would make things even worse | 08:43 |
| arthurl | i use it more to catch up on sleep every now and then when i'm feeling really tired (which i was yesterday all day) | 08:43 |
| arthurl | i also think the years of drinking coffee moderately (sometimes excessively maybe) didn't help either- my adrenal glands might be shot lol | 08:43 |
| arthurl | and now that i drink much less caffeine and have been trying to get off it actually i think my body might not know what to do | 08:44 |
| CaptHindsight | pretty much everyone I come into contact with is miserable in the morning, some all day | 08:44 |
| arthurl | lol | 08:46 |
| CaptHindsight | does you heart start to race when chased by a bear? | 08:46 |
| arthurl | i think it also has a lot to do with how much REM sleep i'm getting every night | 08:46 |
| arthurl | lol not sure i understand your question | 08:47 |
| CaptHindsight | referencing your adrenal gland theory | 08:47 |
| arthurl | the answer is probably yes so you're saying my adrenal glands are fine- i get it :) | 08:47 |
| arthurl | i've heard excessive caffeine can definitely have an impact on that | 08:48 |
| arthurl | maybe i should go get some legit bloodwork done | 08:48 |
| arthurl | and go from there | 08:48 |
| CaptHindsight | change countries, jobs, philosophies, etc | 08:49 |
| arthurl | need to find a wife first- once that happens i'm open to all of that | 08:50 |
| CaptHindsight | trade the tech journal before bed for accounting publications | 08:50 |
| arthurl | accounting publications such as ? | 08:51 |
| CaptHindsight | there must be some | 08:51 |
| CaptHindsight | but kanzure is right, stop forcing yourself to sleep | 08:52 |
| arthurl | so stay up- then get inadequate sleep and be miserable the next day? | 08:53 |
| CaptHindsight | it only lasts a few days | 08:53 |
| arthurl | what do you mean? | 08:53 |
| arthurl | i've been tired like this for past few years | 08:54 |
| CaptHindsight | most people I know with similar experiences and work 8-6 catch up on weekends and are generally miserable | 08:55 |
| CaptHindsight | it's not until they change their lifestyle that they become less miserable | 08:56 |
| arthurl | lol | 08:56 |
| arthurl | well yeah i agree it's on the weekends when i can catch up i feel better- but i think most people combat their inadequate sleep schedule etc with caffeine and whatnot | 08:56 |
| arthurl | i'd think there is a logical/natural solution to this | 08:57 |
| kanzure | how is "stop forcing yourself to sleep" not a logical solution? | 08:59 |
| CaptHindsight | the actual question is" how do I solve this scheduling problem without changing my schedule?" | 09:03 |
| CaptHindsight | let us know what you find out | 09:04 |
| kanzure | ask the company to relocate you to their office in ((pick a location that has absurdly short days)) | 09:05 |
| CaptHindsight | plus you have waited a long time to start working on this problem | 09:09 |
| kanzure | hm? | 09:09 |
| CaptHindsight | he's 30 | 09:09 |
| CaptHindsight | wasn't this a problem for the past several years | 09:10 |
| kanzure | ignoring a problem for a few years makes it unsolvable? | 09:11 |
| kanzure | CaptHindsight: the space cloud stuff was about arbitrary manufacture in orbit from dna; it's the only near-term way we know of to get self-replication to work. otherwise you have to bring all your own (bulky) non-biological manufacturing equipment. | 09:13 |
| kanzure | oh i misunderstood your message actually ("ponders the market for anti-bacterial space wipes") nevermind | 09:13 |
| kanzure | nmz787_: and that's also a good reason to get protein radio stuff to work..... space comms. | 09:15 |
| kanzure | oh actually, light and rhodpsins would work well enough | 09:24 |
| kanzure | or, alternatively, if you have a bunch of dna synthesis and dna sequencing equipment in that cloud, then you could also communicate with dna in both directions, if necessary. however, you would be limited by the amount of radio equipment that you could launch. | 09:25 |
| CaptHindsight | space wipes was an attempt at humor | 09:34 |
| kanzure | not sure how to get information out of a bacteria nebula. pigment? probably something about astro-spectroscopy signal stuff. | 09:35 |
| maaku | .title https://deepmind.com/blog#decoupled-neural-interfaces-using-synthetic-gradients | 09:57 |
| yoleaux | Google DeepMind | 09:57 |
| maaku | that's a useless title | 09:57 |
| maaku | "Decoupled Neural Interfaces using Synthetic Gradients" | 09:57 |
| -!- c0rw1n_ [~c0rw1n@91.181.11.254] has joined ##hplusroadmap | 10:00 | |
| -!- c0rw1n [~c0rw1n@65.158-246-81.adsl-dyn.isp.belgacom.be] has quit [Ping timeout: 252 seconds] | 10:00 | |
| maaku | .title http://news.berkeley.edu/2016/08/29/center-for-human-compatible-artificial-intelligence/ | 10:03 |
| yoleaux | UC Berkeley launches Center for Human-Compatible Artificial Intelligence | Berkeley News | 10:03 |
| maaku | .title http://www.openphilanthropy.org/focus/global-catastrophic-risks/potential-risks-advanced-artificial-intelligence/uc-berkeley-center-human-compatible-ai | 10:04 |
| yoleaux | UC Berkeley — Center for Human-Compatible AI | Open Philanthropy Project | 10:04 |
| maaku | (different writeup) | 10:04 |
| maaku | .title https://github.com/baidu/paddle | 10:08 |
| yoleaux | GitHub - baidu/Paddle: PArallel Distributed Deep LEarning | 10:08 |
| kanzure | shouldn't there be some cells that produce viruses as a defensive mechanism? | 10:08 |
| kanzure | maaku: http://www.paddlepaddle.org/doc/cluster/opensource/cluster_train.html | 10:09 |
| -!- c0rw1n_ is now known as c0rw1n | 10:09 | |
| maaku | kanzure: hrm.. we have a lot of viral DNA | 10:09 |
| maaku | herpes lays dormant for months, or years before becoming active again | 10:10 |
| __mz_o | kanzure: i could see how the virus dna could evolve but how would the cell create the virus body? | 10:10 |
| maaku | i suppose it wouldn't be too hard to weaponize that | 10:10 |
| kanzure | it is not clear to me how they are combining the training results | 10:10 |
| kanzure | __mz_o: virus capsids are created in cells | 10:10 |
| maaku | have you seen anyone do any work on combining separately trained networks? | 10:11 |
| kanzure | yes the literature calls this "data parallelism" because they suck at naming things | 10:11 |
| maaku | I'm not finding any code for that. maybe I should expand my search to the literature | 10:11 |
| maaku | ok cool thanks | 10:11 |
| maaku | yeah i would not have guessed that's what 'data parallelism' meant :\ | 10:11 |
| kanzure | http://diyhpl.us/~bryan/papers2/ai/machine-learning/GeePS:%20Scalable%20deep%20learning%20on%20distributed%20GPUs%20with%20a%20GPU-specialized%20parameter%20server%20-%202016.pdf | 10:12 |
| kanzure | bottom of page 3 http://diyhpl.us/~bryan/papers2/ai/machine-learning/FireCaffe:%20near-linear%20acceleration%20of%20deep%20neural%20network%20training%20on%20compute%20clusters%20-%202016.pdf | 10:13 |
| kanzure | "One weird trick for parallelizing convolutional neural networks" http://arxiv.org/pdf/1404.5997.pdf | 10:13 |
| kanzure | "stochastic gradient descent" | 10:14 |
| kanzure | averaging https://github.com/fchollet/keras/issues/106#issuecomment-157130117 | 10:14 |
| kanzure | and this stuff https://github.com/mila-udem/platoon | 10:14 |
| -!- c0rw1n [~c0rw1n@91.181.11.254] has quit [Ping timeout: 265 seconds] | 10:14 | |
| -!- c0rw1n [~c0rw1n@42.156-201-80.adsl-dyn.isp.belgacom.be] has joined ##hplusroadmap | 10:18 | |
| maaku | hrm. i'll have to formalize better what I'm trying to do | 10:19 |
| maaku | what I'd like to be able to do is to train a network on pictures of cats, and train another network on pictures of flowers | 10:20 |
| maaku | and then combine the two, e.g. by merging the feature sets layer by layer | 10:20 |
| kanzure | you mean copy the same network to two GPU, do training on different data, then merge the results? | 10:20 |
| maaku | pretty much | 10:22 |
| kanzure | ya that is the above stuff, including the "one weird trick" article | 10:23 |
| -!- xmgx [~xmgx@189.5.227.175] has joined ##hplusroadmap | 10:23 | |
| maaku | ok cool I'll read more closely thanks | 10:24 |
| maaku | well actually different network models, but the models are composed of smaller repeated components that are the same | 10:25 |
| maaku | so if there's a way to do it for identical network parameters, then I think I can figure out a divide and conquer approach | 10:26 |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has quit [Ping timeout: 276 seconds] | 10:26 | |
| kanzure | i don't really know what the "model parallelism" stuff does; they split subcomponents to multiple GPUs but i'm not sure how that's helpful (i think you still need to communiucate the results from the previous module, which means gpu-gpu communication) | 10:27 |
| maaku | This is separate from the transcription project (I'm not charging time), but it seems that this would let you do cross-domain expertise transfer | 10:28 |
| maaku | which is the holy grail of agi.. | 10:28 |
| maaku | Skimming through, it looks like that is exactly what they are doing, but batching the gpu-gpu communication so as to not stall. But I'll read more closely. | 10:30 |
| kanzure | i don't see how batching prevents stalls there; at any given clock tick you need data at all layers at each gpu to do any relevant training.... | 10:30 |
| kanzure | perhaps it's just enough that /some/ or /most/ stalls are eliminated, not every stall | 10:30 |
| -!- c0rw1n [~c0rw1n@42.156-201-80.adsl-dyn.isp.belgacom.be] has quit [Ping timeout: 260 seconds] | 10:32 | |
| -!- c0rw1n [~c0rw1n@161.47-244-81.adsl-dyn.isp.belgacom.be] has joined ##hplusroadmap | 10:32 | |
| -!- c0rw1n_ [~c0rw1n@91.181.1.218] has joined ##hplusroadmap | 10:36 | |
| -!- jtimon [~quassel@38.110.132.37.dynamic.jazztel.es] has quit [Ping timeout: 258 seconds] | 10:38 | |
| -!- c0rw1n [~c0rw1n@161.47-244-81.adsl-dyn.isp.belgacom.be] has quit [Ping timeout: 264 seconds] | 10:40 | |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has joined ##hplusroadmap | 11:00 | |
| -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-ahlabixzkuedikuf] has quit [Quit: Connection closed for inactivity] | 11:12 | |
| -!- HEx1 [~HEx@hexwab.plus.com] has quit [Ping timeout: 240 seconds] | 11:33 | |
| -!- ArturShaik [~ArturShai@185.66.252.104] has quit [Ping timeout: 240 seconds] | 11:33 | |
| -!- HEx1 [~HEx@hexwab.plus.com] has joined ##hplusroadmap | 11:34 | |
| -!- jtimon [~quassel@38.110.132.37.dynamic.jazztel.es] has joined ##hplusroadmap | 11:45 | |
| -!- c0rw1n_ [~c0rw1n@91.181.1.218] has quit [Ping timeout: 264 seconds] | 11:52 | |
| -!- gourneau [~gourneau@173.250.197.104.bc.googleusercontent.com] has quit [Ping timeout: 264 seconds] | 11:52 | |
| -!- amiller [~socrates1@unaffiliated/socrates1024] has quit [Ping timeout: 265 seconds] | 11:59 | |
| -!- Jawmare [~Jawmare@unaffiliated/jawmare] has quit [Read error: Connection reset by peer] | 11:59 | |
| -!- Jawmare [~Jawmare@unaffiliated/jawmare] has joined ##hplusroadmap | 12:00 | |
| fenn | arthurl: 2.5 mg of melatonin is way too much, you should be taking less than a fifth of that | 12:00 |
| fenn | too much melatonin will just screw you up even more | 12:00 |
| __mz_o | that makes sense | 12:00 |
| arthurl | fenn i get them in 5 g pills and i usually bite off half or a quarter- hard to break it down smaller than that | 12:01 |
| __mz_o | i have one thats 3mg in one pill | 12:01 |
| -!- Guest55662 [~socrates1@li175-104.members.linode.com] has joined ##hplusroadmap | 12:01 | |
| __mz_o | too groggy in the morning | 12:01 |
| arthurl | but i will say sometimes don't feel the affect if i break off a quarter | 12:01 |
| pasky | cross-domain transfer is done very commonly in NNs | 12:02 |
| arthurl | fenn i take it rather infrequently have had the bottle for years now- i'd say on average maybe two or three times a month so far | 12:03 |
| fenn | arthurl i also recommend keeping a sleep journal and plotting a graph | 12:03 |
| arthurl | fenn and plotting what exactly? hours of sleep? | 12:03 |
| -!- c0rw1n_ [~c0rw1n@91.181.1.218] has joined ##hplusroadmap | 12:04 | |
| fenn | whatever you're interested in, i guess the best would be something like a wearable sleep monitor that logs exactly when you started sleeping and woke up | 12:04 |
| fenn | also make note of whether you're tired during the day and when | 12:04 |
| fenn | someone has a patent on not-absurdly-huge-doses of melatonin | 12:05 |
| fenn | so you have to break the pill into tiny pieces unfortunately | 12:05 |
| arthurl | oh is that the reason | 12:06 |
| arthurl | that's insane | 12:06 |
| fenn | or get the 0.5mg capsules | 12:06 |
| arthurl | i always wondered why they sell in such large doses | 12:06 |
| fenn | because more is better right? | 12:06 |
| arthurl | people unknowingly definitely take 5mg | 12:06 |
| fenn | 5-HTP is another option which seems to be more precise, it boosts your natural melatonin production so it's released at the right time (presuming your brain is working mostly normally of course) | 12:07 |
| fenn | so you don't have to take it at exactly the right time | 12:08 |
| arthurl | i've heard of it- never did much research on it however- let me take a look | 12:08 |
| __mz_o | you have to take Green Tea Extract with 5-htp | 12:09 |
| arthurl | another thing i noticed is if i get quality sleep lets say two nights in a row | 12:10 |
| arthurl | the third night | 12:10 |
| arthurl | no matter what i do | 12:10 |
| arthurl | i have too much eenrgy | 12:10 |
| arthurl | energy* | 12:10 |
| arthurl | even if i go to the gym have a long day at work etc | 12:10 |
| arthurl | and so then i stay up and the vicious cycle repeats | 12:10 |
| arthurl | https://www.bulletproofexec.com/sleep-hacking-part-3-falling-asleep-fast-with-biochemistry/ | 12:12 |
| fenn | well i'm skeptical until you show some data | 12:13 |
| maaku | kanzure: https://openai.com/blog/infrastructure-for-deep-learning/ | 12:14 |
| maaku | pasky: references? | 12:14 |
| -!- c0rw1n_ is now known as c0rw1n | 12:17 | |
| kanzure | well damn. i guess i should throw out my docker-swarm/nvidia stuff and switch to this kubernetes stuff. | 12:18 |
| kanzure | hehe they are using terraform | 12:19 |
| kanzure | bleh they are using chef. no. | 12:19 |
| kanzure | this is pretty close. | 12:20 |
| maaku | chef? ugh | 12:20 |
| kanzure | oh this is the openai people. that's a little weird. i would have expected them to have better infrastructure planz than me. | 12:20 |
| kanzure | hehe they released only a tiny bit of their infrastructure. hehehe. | 12:22 |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has quit [Remote host closed the connection] | 12:39 | |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has joined ##hplusroadmap | 12:39 | |
| pasky | maaku: http://arxiv.org/abs/1603.06127 ;) for one thing, in general pretrained word embeddings used for all sorts of stuff; what we have found in https://arxiv.org/abs/1605.04655 is that learning a RNN sentence model on IRC chat logs helps logical inference on wikipedia articles and children stories; etc. | 12:40 |
| pasky | in images using imagenet-pretrained sets of all sorts of classification tasks is a common trick | 12:40 |
| pasky | which is one of the things we do in http://vize.it ;) | 12:40 |
| kanzure | "Sentence Pair Scoring: Towards Unified Framework for Text Comprehension" | 12:42 |
| kanzure | "Joint Learning of Sentence Embeddings for Relevance and Entailment" | 12:42 |
| kanzure | "a single GTX 1080 GPU deep learning box would come around 1500$, if you pay 0.7$/hr for your cloud server, you should buy if you use more than 1500/0.7 = 2142h. So, if you need more than 90 days of GPU time, you should probably buy your box. Of course, if the cloud server is slower than GTX 1080, then the benefit is multiplied. But ... your own box would not be scalable. You'd still need AWS to speed up training." | 13:04 |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has quit [Ping timeout: 240 seconds] | 13:30 | |
| abetusk | any ideas on cost effective TCE detection? | 13:32 |
| arthurl | 4:37 pm- my first yawn of the day :) | 13:34 |
| arthurl | feel fine though | 13:35 |
| -!- Malvolio [~Malvolio@unaffiliated/malvolio] has joined ##hplusroadmap | 13:42 | |
| CaptHindsight | arthurl: some have concierge docs that administer propofol, with mixed results :) | 13:45 |
| arthurl | lol | 13:46 |
| arthurl | what do you guys think about that bulletproof coffee stuff? and the claim that microtoxins exist in a lot of the coffee that's sold here in the states? | 13:54 |
| Aurelius_Work | bulletproof coffee is, to my knowledge, marketing bullshit | 13:55 |
| kanzure | it's definitely bullshit marketing, at least | 13:56 |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Quit: Leaving] | 14:06 | |
| -!- augur [~augur@2602:304:cdac:e260:89d4:4fd8:ce9d:7f18] has joined ##hplusroadmap | 14:07 | |
| -!- anoncicada [anoncicada@gateway/shell/elitebnc/x-vpqmxikywvgaqxfk] has joined ##hplusroadmap | 14:07 | |
| kanzure | "USBee: Air-gap covert-channel via electromagnetic emission from USB" http://cyber.bgu.ac.il/t/USBee.pdf | 14:17 |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has quit [Read error: Connection reset by peer] | 14:51 | |
| -!- anoncicada [anoncicada@gateway/shell/elitebnc/x-vpqmxikywvgaqxfk] has quit [Quit: faggory daggory doo!] | 14:52 | |
| -!- anoncicada [anoncicada@gateway/shell/elitebnc/x-knvobzxdpcbhmyud] has joined ##hplusroadmap | 14:54 | |
| maaku | pasky thanks I'll look at it later (I'm on my phone) | 14:56 |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has joined ##hplusroadmap | 14:58 | |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has quit [Client Quit] | 14:59 | |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has joined ##hplusroadmap | 15:00 | |
| -!- kenn123 [~kenn123@CPEf0f2490c77c3-CMf0f2490c77c0.cpe.net.cable.rogers.com] has joined ##hplusroadmap | 15:06 | |
| docl | https://www.youtube.com/watch?v=0SUSJZaPIcs | 15:09 |
| docl | .title | 15:09 |
| yoleaux | Ray Kurzweil and Robert Freitas Chat About the Future of Nanotechnology - YouTube | 15:09 |
| -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-hsbjjwavqvmywmde] has quit [Quit: Connection closed for inactivity] | 15:13 | |
| kanzure | docl: by any chance did you read the backlog? e.g. http://gnusha.org/logs/2016-08-28.log | 15:14 |
| docl | I've been trying to stay on top of the backlogs, yeah | 15:15 |
| docl | Interesting thought about getting biology to work in space conditions with tiny droplets | 15:16 |
| kanzure | yeah i was thinking about that for space manufacturing reasons and self-replication reasons | 15:17 |
| kanzure | not sure about how to do large-scale object fabrication with only biology | 15:18 |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap | 15:20 | |
| -!- Guest55662 [~socrates1@li175-104.members.linode.com] has quit [Ping timeout: 244 seconds] | 15:21 | |
| docl | you might use something nonbiological (a laser, say) as an external growth trigger | 15:21 |
| kanzure | i think you need the laser at the very least to keep everything somewhat together- like pushing around droplets and such | 15:22 |
| kanzure | nutrition delivery by dust seems to only work for droplets. if you want to make big things, you need large clumps of cells working together (or dying in some patterned design). and then you need to have some way to get nutrition to those big groups of cells. | 15:23 |
| kanzure | maybe you could handwave with something like: you fatten up the cells with nutrients, which they store, then they join together and harden into large-scale objects (during which time they receive no food because we don't know how to make circulatory systems yet) | 15:23 |
| -!- Guest43826 [~socrates1@li175-104.members.linode.com] has joined ##hplusroadmap | 15:24 | |
| kanzure | hardened biofilms and such. | 15:24 |
| kanzure | wasn't there something about cellular crosslinking as a method of manufacturing | 15:38 |
| kanzure | "fused deposition modelling (adding layers of cells to create a structure)" | 15:40 |
| kanzure | rna scaffold as an enzyme conveyor belt http://2012.igem.org/Team:UIUC-Illinois/Project/Future/Scaffold | 15:42 |
| kanzure | wat "To start, the iGEM team designed the drone’s shape in 3D modeling software. This design file was sent to biomaterials company Ecovative Design, which fabricated the drone body from an 8-inch square of fungal mycelium via vacuforming." | 15:45 |
| kanzure | "bio-photolithography" http://2011.igem.org/Team:Glasgow "light-responsive promoters linked to proteins which can either disperse the biofilm or cement it" "light-controlled 3D sculpting of biofilms" | 15:49 |
| kanzure | and http://2012.igem.org/Team:Peking/Project/3D/3D | 15:50 |
| kanzure | bah "At this time, no quantitative model of target morphology during pattern formation exists" from "On a model of pattern regeneration based on cell memory" http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0118091 | 15:55 |
| xentrac | so I think I have a rough quantitative cost model of how much my local laser cutting shop will charge me | 15:57 |
| xentrac | on 3 mm MDF | 15:58 |
| xentrac | it's about US$0.0012 per millimeter plus US$0.011 per vertex | 15:58 |
| xentrac | is that good? | 15:58 |
| kanzure | hmm these guys seem to be modeling 3d morphology from genetics, http://allencenter.tufts.edu/wp-content/uploads/Whitepaper.pdf | 15:58 |
| kanzure | "Perhaps most crucially, briefly altering the bioelectric connectivity of a cellular network enables permanent rewriting of an organism’s target morphology [93]: genomically-normal worms can be changed to a 2-headed form that regenerates with 2 heads in perpetuity, illustrating the ability to stably re-wire bioelectric circuits with permanent changes to the overall anatomy (Fig. 3)." | 16:02 |
| kanzure | "The “bioelectric code” 3 is defined as the mapping of real-time electric circuit dynamics among tissues to the pattern-regulatory functions that cells carry out. What we have learned, after 16 years of focused effort in this field, is that bioelectrical signaling 1) exerts profound control over large-scale morphogenetic properties in a range of model systems [64, 66], 2) facilitates exploiting native modularity (such as triggers ... | 16:06 |
| kanzure | ... complex downstream patterning outcomes as a kind of master regulator) [75, 94], 3) is transduced by a set of known mechanisms into downstream chemical signals (neurotransmitters and other morphogens) [65, 95] and gene transcription changes [51], and 4) forms feedback loops with genetic pathways, often over-riding competing signals from other modalities [64, 89, 96]." | 16:06 |
| kanzure | and: | 16:06 |
| kanzure | "The striking data showing that rewriting the bioelectric circuit dynamics leads directly to the reprogramming of shape in vivo suggest a new metaphor for understanding morphogenesis. The current textbooks say that the DNA is the software while the cell is the hardware that interprets it. We believe this metaphor needs to be revised to encompass the true circularity of the process: the DNA determines the hardware (by encoding the ... | 16:06 |
| kanzure | ... specific gap junction and ion channel proteins that can support electrical dynamics in cell networks), while the resulting bioelectric circuits have their own dynamics that regulate gene expression, which in turn may affect the number and type of channels, in a continuous interplay between genetics and physiology. The hardware is important, in that it limits what can happen. But it does not fully determine the outcome - the ... | 16:06 |
| kanzure | ... morphogenetic outcome is in large part the result of bioelectric software - circuit dynamics that run on the cellular ion channel hardware (very much like what happens in the brain, where cognitive content and specific memories derive from the bioelectric software, not the genome directly). Learning to manipulate the bioelectric software running on somatic cell networks is the key to top-down programming of morphogenesis." | 16:06 |
| kanzure | biology is awful | 16:08 |
| docl | horrible | 16:09 |
| -!- anoncicada [anoncicada@gateway/shell/elitebnc/x-knvobzxdpcbhmyud] has quit [Ping timeout: 255 seconds] | 16:09 | |
| kanzure | http://ase.tufts.edu/biology/labs/levin/ | 16:11 |
| kanzure | alright well i think an artificial pattering system is probably easier to implement than hijacking existing stuff. probably it's a matter of "pick amorphous structure morphology technique from computer science literature, then encode into regulatory gene network, also pick some adherins and membrane-permeable communication system and coordinate system" which would probably be much simpler than reverse engineering whatever it is ... | 16:12 |
| kanzure | ... mammals/protists are doing | 16:13 |
| kanzure | "Synthetic morphology: prospects for engineered, self-constructing anatomies." http://www.ncbi.nlm.nih.gov/pubmed/18510501 (2008) | 16:13 |
| docl | yeah that's what I was thinking | 16:15 |
| kanzure | "2- and 3-dimensional synthetic large-scale de novo patterning by mammalian cells through phase separation" http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4746622/ | 16:16 |
| kanzure | er, this uses cadhesins instead of adhesins. | 16:17 |
| -!- anoncicada [anoncicada@gateway/shell/elitebnc/x-vmjhefuelkdwynrl] has joined ##hplusroadmap | 16:18 | |
| -!- arthurl [~arthurl@38.117.157.141] has quit [Ping timeout: 276 seconds] | 16:32 | |
| -!- augur [~augur@2602:304:cdac:e260:89d4:4fd8:ce9d:7f18] has quit [Remote host closed the connection] | 16:40 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has quit [Ping timeout: 240 seconds] | 16:43 | |
| -!- strangewarp [~strangewa@c-76-25-206-3.hsd1.co.comcast.net] has quit [Read error: Connection reset by peer] | 16:51 | |
| -!- oleo [~oleo@unaffiliated/oleo] has quit [Read error: Connection reset by peer] | 17:16 | |
| -!- oleo [~oleo@unaffiliated/oleo] has joined ##hplusroadmap | 17:16 | |
| docl | what's the main advantage of vacuum-tolerant biology over just growing it in vats? lack of need to create/maintain vats? | 17:31 |
| -!- ebowden [~ebowden@2001:8003:100e:c500:c13c:d5c1:1198:fe76] has joined ##hplusroadmap | 17:33 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:c13c:d5c1:1198:fe76] has quit [Changing host] | 17:33 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 17:33 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 17:34 | |
| docl | ability to steer the droplets around with lasers without worrying about air currents is good. but the droplets don't necessarily need to be alive for that approach to fabrication. | 17:40 |
| docl | there could be some interesting applications of bacteria to mineral enrichment. it could be energy-cheaper than plasmafying the raw material and sorting it through a mass spectrometer type mechanism. | 17:43 |
| -!- ebowden [~ebowden@2001:8003:100e:c500:1ded:615f:644:2d89] has joined ##hplusroadmap | 18:02 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:1ded:615f:644:2d89] has quit [Changing host] | 18:02 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 18:02 | |
| -!- jtimon [~quassel@38.110.132.37.dynamic.jazztel.es] has quit [Ping timeout: 265 seconds] | 18:02 | |
| -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Remote host closed the connection] | 18:06 | |
| -!- kenn123 [~kenn123@CPEf0f2490c77c3-CMf0f2490c77c0.cpe.net.cable.rogers.com] has quit [Quit: I Use Textual IRC on OS X] | 18:06 | |
| -!- yashgaroth [~yashgarot@2602:306:35fa:d500:f5e0:f867:a11d:8d52] has joined ##hplusroadmap | 18:07 | |
| kanzure | docl: right, presumably you don't have too many vats in space :/ | 18:09 |
| kanzure | and also you might not have that much material processing infrastructure built up | 18:11 |
| kanzure | "principle of least action" | 18:14 |
| kanzure | http://ase.tufts.edu/biology/labs/levin/research/spatial.htm | 18:15 |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 18:32 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:308f:ee57:393:6c79] has joined ##hplusroadmap | 18:35 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:308f:ee57:393:6c79] has quit [Changing host] | 18:35 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 18:35 | |
| kanzure | "Proceedings of the artificial life conference 2016" https://mitpress.mit.edu/sites/default/files/titles/free_download/9780262339360_ALIFE_2016.pdf includes pargellis paper (page 60) and some lipson stuff, although nothing looks particularly new here. | 18:36 |
| kanzure | page 562 has a visual simulation of eukaryote | 18:38 |
| kanzure | conference site http://xva.life/ | 18:40 |
| -!- Gurkenglas [Gurkenglas@dslb-094-223-128-173.094.223.pools.vodafone-ip.de] has quit [Ping timeout: 244 seconds] | 18:41 | |
| kanzure | prokaryote cell visualization stuff http://www.lindsayvirtualhuman.com/?page_id=469 | 18:42 |
| kanzure | macosx http://www.lindsayvirtualhuman.com/wp-content/uploads/Prokaryo_LC2.0.zip | 18:42 |
| kanzure | http://www.lindsayvirtualhuman.com/wp-content/uploads/insideEcoliWithFrame-e1418483583981.png | 18:43 |
| kanzure | page 616 has some weird stuff about combinators and ribosomes for self-replication | 18:59 |
| kanzure | er, 634 | 18:59 |
| -!- augur [~augur@2601:640:8001:4222:7d38:aaa1:1cdc:a8ec] has joined ##hplusroadmap | 19:00 | |
| kanzure | re: the nasty ion channel stuff above about "bioelectricity", it seems like the NEURON-style simulators would probably be effective, as a system of ordinary differential equations describing ion channels plus gene network accumulative effects and switching behaviors | 19:02 |
| -!- augur [~augur@2601:640:8001:4222:7d38:aaa1:1cdc:a8ec] has quit [Ping timeout: 258 seconds] | 19:04 | |
| genehacker | well you also don't have too much water in space | 19:08 |
| -!- augur [~augur@c-67-160-198-240.hsd1.ca.comcast.net] has joined ##hplusroadmap | 19:09 | |
| kanzure | got a bunch of ice | 19:26 |
| -!- augur [~augur@c-67-160-198-240.hsd1.ca.comcast.net] has quit [Remote host closed the connection] | 19:31 | |
| -!- gene_hacker [~chatzilla@c-50-188-129-64.hsd1.or.comcast.net] has joined ##hplusroadmap | 19:36 | |
| kanzure | i wonder if tesseract still sucks for OCR these days | 19:51 |
| kanzure | "data parallelism and parameter averaging" http://deeplearning4j.org/spark explanation https://news.ycombinator.com/item?id=12394444 | 19:52 |
| kanzure | "The description for multi-GPUs is through this link: http://deeplearning4j.org/gpu under the subhead "Multi-GPU data parallelism " | 19:54 |
| kanzure | https://github.com/deeplearning4j/deeplearning4j/blob/77b836cd7daba5fbd4c7a77d341d489cf6e9a220/deeplearning4j-core/src/main/java/org/deeplearning4j/parallelism/ParallelWrapper.java | 19:54 |
| kanzure | "Large models with shared weights get tricky but less frequent asynchronous updates with schemes like hogwild seem to work with SGD. I believe TF has support for this too. It won't scale linearly but might be good enough. There's some excitement about synthetic gradients to allow less communication and further parallelism." | 19:54 |
| -!- Guest43826 is now known as amiller | 19:57 | |
| -!- amiller [~socrates1@li175-104.members.linode.com] has quit [Changing host] | 19:57 | |
| -!- amiller [~socrates1@unaffiliated/socrates1024] has joined ##hplusroadmap | 19:57 | |
| -!- xmgx [~xmgx@189.5.227.175] has quit [Remote host closed the connection] | 20:13 | |
| -!- ArturShaik [~ArturShai@185.66.252.104] has joined ##hplusroadmap | 20:15 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 20:16 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:b180:4c8a:9c7e:c816] has joined ##hplusroadmap | 20:17 | |
| -!- ebowden [~ebowden@2001:8003:100e:c500:b180:4c8a:9c7e:c816] has quit [Changing host] | 20:17 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 20:17 | |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has joined ##hplusroadmap | 20:34 | |
| -!- ArturSha1 [~ArturShai@185.66.252.104] has joined ##hplusroadmap | 20:44 | |
| -!- ArturShaik [~ArturShai@185.66.252.104] has quit [Ping timeout: 244 seconds] | 20:48 | |
| -!- augur [~augur@2601:640:8001:4222:bc26:3dbe:6ff1:6d5a] has joined ##hplusroadmap | 20:48 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 20:49 | |
| justanotheruser | Since freecad is dead, does anyone here know how to combine parts so they are treated a single part that cannot be rotated or moved relative to all the other parts? | 20:56 |
| kanzure | maybe nmz787_ will remember freecad things | 20:58 |
| -!- cynsia [cyn@2600:3c03::f03c:91ff:fe79:d5d] has joined ##hplusroadmap | 21:06 | |
| -!- ebowden [~ebowden@147.69.145.152] has joined ##hplusroadmap | 21:10 | |
| -!- ebowden [~ebowden@147.69.145.152] has quit [Changing host] | 21:10 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 21:10 | |
| xentrac | freecad is dead? | 21:26 |
| xentrac | what did people switch to? | 21:27 |
| justanotheruser | #freecad I mean | 21:27 |
| kanzure | they switched to brlcad and solidworks | 21:27 |
| kanzure | and solvespace | 21:28 |
| CaptHindsight | justanotheruser: combining parts is one operation | 21:28 |
| CaptHindsight | what do you mean by " cannot be rotated or moved relative to all the other parts" | 21:29 |
| justanotheruser | CaptHindsight: I mean if they are glued together | 21:29 |
| justanotheruser | which workbench do I combine parts in | 21:29 |
| CaptHindsight | Part | 21:30 |
| CaptHindsight | you select the two parts and then use Join | 21:30 |
| -!- sandeepkr [~sandeepkr@182.75.187.110] has quit [Ping timeout: 265 seconds] | 21:31 | |
| CaptHindsight | took me a while to find it since it works like no other Cad application | 21:31 |
| justanotheruser | Neat | 21:31 |
| justanotheruser | you mean union right? because that seems to give me the desired result | 21:32 |
| CaptHindsight | Join is for walled objects | 21:32 |
| justanotheruser | sorry, walled? | 21:33 |
| CaptHindsight | might be Union | 21:33 |
| CaptHindsight | what it says | 21:33 |
| justanotheruser | In this context I'm not sure what the difference would be | 21:34 |
| CaptHindsight | wish they would have looked at NX, SW or Catia for guidance on making it more intuitive | 21:34 |
| CaptHindsight | Part is usually for one part | 21:34 |
| CaptHindsight | Assembly is for many parts | 21:34 |
| justanotheruser | right | 21:34 |
| CaptHindsight | Part and Parts would have been better | 21:35 |
| CaptHindsight | I just occasionally use it as a viewer before i open the file on a real CAD system :) | 21:36 |
| xentrac | heh | 21:38 |
| xentrac | I've been doing parametric panel-cutting CAD in raw PostScript this week | 21:39 |
| -!- augur [~augur@2601:640:8001:4222:bc26:3dbe:6ff1:6d5a] has quit [Remote host closed the connection] | 21:39 | |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has quit [Ping timeout: 244 seconds] | 21:39 | |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has joined ##hplusroadmap | 21:39 | |
| xentrac | /chamfer { 45 rt chamferwidth sq2 mul fd 45 rt } def | 21:40 |
| justanotheruser | crap sorry, my battery died | 21:40 |
| justanotheruser | should move my znc from my laptop to my server... | 21:40 |
| xentrac | 3 { 0 octsize rmoveto currentpoint row moveto } repeat | 21:40 |
| xentrac | it has problems. you can't generate geometry and later modify it (e.g. by adding chamfers; I have a bunch of bullshit like /chamferside { freesidelength chamferwidth sub fd } bdef), the paradigm of instantiating a parametric shape is procedural rather than constraint-solving, and there's no 3-D visualization of the assembled panels | 21:44 |
| -!- augur [~augur@c-67-160-198-240.hsd1.ca.comcast.net] has joined ##hplusroadmap | 21:45 | |
| xentrac | but | 21:46 |
| xentrac | well who am I trying to kid, it sucks :) | 21:46 |
| xentrac | it does eventually work | 21:47 |
| xentrac | but it's a lot like programming in assembly language | 21:47 |
| -!- oleo [~oleo@unaffiliated/oleo] has quit [Quit: Leaving] | 21:49 | |
| juri_ | I'm still hacking away on implicitcad. | 21:57 |
| -!- justanotheruser [~justanoth@unaffiliated/justanotheruser] has quit [Ping timeout: 244 seconds] | 22:04 | |
| CaptHindsight | ImplicitCAD is around four thousand lines of code | 22:17 |
| CaptHindsight | I think NX has more than 4K files in the suite | 22:17 |
| juri_ | our code is better. ;) | 22:18 |
| -!- drewbot [~cinch@ec2-54-80-14-177.compute-1.amazonaws.com] has quit [Remote host closed the connection] | 22:22 | |
| -!- drewbot [~cinch@ec2-54-235-26-134.compute-1.amazonaws.com] has joined ##hplusroadmap | 22:22 | |
| -!- CaptHindsight [~2020@unaffiliated/capthindsight] has quit [Quit: Leaving] | 22:38 | |
| -!- Orpheon [~Orpheon@48.188.62.81.dynamic.wline.res.cust.swisscom.ch] has joined ##hplusroadmap | 22:41 | |
| -!- cynsia [cyn@2600:3c03::f03c:91ff:fe79:d5d] has quit [Quit: This computer has gone to sleep] | 22:43 | |
| -!- CaptHindsight [~2020@unaffiliated/capthindsight] has joined ##hplusroadmap | 22:44 | |
| -!- yashgaroth [~yashgarot@2602:306:35fa:d500:f5e0:f867:a11d:8d52] has quit [Quit: Leaving] | 22:53 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 22:57 | |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has joined ##hplusroadmap | 23:22 | |
| xentrac | I should try implicitcad; I guess I have this idea that I will need to understand Haskell and that will take me a long time | 23:41 |
| xentrac | but I need to understand Haskell sooner or later anyway | 23:41 |
| -!- ebowden [~ebowden@unaffiliated/ebowden] has quit [Remote host closed the connection] | 23:56 | |
| --- Log closed Wed Aug 31 00:00:41 2016 | ||
Generated by irclog2html.py 2.15.0.dev0 by Marius Gedminas - find it at mg.pov.lt!
