by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Apr 05 17:00

    kephale on scenery

    (compare)

  • Apr 05 17:00

    kephale on renderer

    (compare)

  • Apr 05 17:00

    kephale on gui-updates

    (compare)

  • Apr 05 17:00

    kephale on travis

    (compare)

  • Apr 05 17:00

    kephale on natives_cleanup

    (compare)

  • Apr 05 17:00

    kephale on funimage_update

    (compare)

  • Apr 05 17:00

    kephale on slurm

    (compare)

  • Apr 05 17:00

    kephale on Physics

    (compare)

  • Apr 05 17:00

    kephale on GraphicsRewrite

    (compare)

  • Apr 05 17:00

    kephale on prerelease

    (compare)

  • Sep 16 2019 16:50

    kephale on functional-sim

    Fix vec3? and vec4? predicates (compare)

  • Jun 09 2019 10:59

    kephale on master

    Change repositories to scijava … (compare)

  • Jun 08 2019 12:17

    kephale on master

    Use newer fun.imagej (compare)

  • May 13 2019 00:34

    kephale on kdtree

    (compare)

  • May 13 2019 00:34

    kephale on master

    Kdtree (#24) * Replace brevis … (compare)

  • May 13 2019 00:34
    kephale closed #24
  • May 11 2019 23:40
    kephale synchronize #24
  • May 11 2019 23:40

    kephale on kdtree

    Increment fun.imagej version (compare)

  • May 07 2019 21:54
    kephale synchronize #24
  • May 07 2019 21:54

    kephale on kdtree

    Update fun.imagej (compare)

Kyle I S Harrington
@kephale
they seem to have a lot of good distributed code
what i mean is situations where the current AI/neural net is being used and trained at the same time
Caue Rego
@cauerego
yeah, that’s what I thought then :)
Kyle I S Harrington
@kephale
it is a slight difference, but the code can end up structured quite differently if they only care about taking a dataset, training on it, then using that for prediction later
Caue Rego
@cauerego
if that’s not legacy code, makes whole sense that it should be able to do it, though.
yup
Kyle I S Harrington
@kephale
i tend to get more excited by online/real-time learning since its more like real intelligence
Caue Rego
@cauerego
maybe their commercial secret is on the training, then.
Kyle I S Harrington
@kephale
it looks like they might be more focused on text learning
Caue Rego
@cauerego
yeah
Kyle I S Harrington
@kephale
but that could be interesting, since one of the things we've been working on is getting agents to learn how to communicate
which has been via sequences of symbols so far
Caue Rego
@cauerego
learning to talk human should be the next logical step towards Ai next gen :)
Kyle I S Harrington
@kephale
and autoencoders/RNNs are one of the ways that has worked for that before
hehe
Caue Rego
@cauerego
also dangerously close to the last step towards singularity, imho.
Kyle I S Harrington
@kephale
i would be impressed enough to see AIs learning how to talk to each other about simple things that they learned about by exploring their environment
mmm, AIs actually understanding human language is probably a little ways down the road still though
Caue Rego
@cauerego
yeah, but, why? in my mind, it’s only due to software research.
Kyle I S Harrington
@kephale
mimicking human language might not be that far away
Caue Rego
@cauerego
it’s certainly not. here:
Kyle I S Harrington
@kephale
but being able to associate language to sensory/motor information
Kyle I S Harrington
@kephale
or vice versa is actually probably the more appropriate direction
hahaha
i'm amused
Caue Rego
@cauerego
:)
Caue Rego
@cauerego
the way I see it, if computers can actually understand human language, there is far from enough data on the web for them to learn everything else unsupervised.
including motor information. maybe not so much sensory, but just maybe.
it depends on how offline that data is. it’s surely digitalized from hospitals to cinema.
Caue Rego
@cauerego
Kyle, just wanted to share with you, hear a bit of your thoughts if you will: Mariox is alive! www.basiux.org
let me try this… oh okay, I see how it worked. :point_up: @basiux square November 29, 2015 3:59 PM
Kyle I S Harrington
@kephale
@cauerego hehehe, how long until it gets past the first baddie
and you're using what, a GA?
Caue Rego
@cauerego
very, very long! in my case, it was less than 8hours and 280 generations
Kyle I S Harrington
@kephale
hehehe
Caue Rego
@cauerego
I hear in Seth’s it could taken him over 24h, but just 30 generations
Kyle I S Harrington
@kephale
gotta get some reinforcement learning in there
Caue Rego
@cauerego
and yeah, it’s a genetic algo
but I’m still trying to understand how it works, since I just ported it
also think it’s broken. there is some reinforcements that should be in place.
I’m currently streaming on https://www.livecoding.tv/cregox/ ;P
Kyle I S Harrington
@kephale
heh
Caue Rego
@cauerego
I’m pretty sure I’m quickly filling up my indexedDB...
Kyle I S Harrington
@kephale
Kyle I S Harrington
@kephale
there still is life in here
Caue Rego
@cauerego
ok dokey
that's so last year!
Kyle I S Harrington
@kephale
tyb ,
Kyle I S Harrington
@kephale
dd