michaelerule
Member 1242
103 entries
87549 views

 RSS
Michael Rule (M, 25)
Pittsburgh, US
Immortal since Dec 25, 2007
Uplinks: 0, Generation 3
( related blog / mirror )
All original media ( images, animations, videos, writing ) on this account is licensed creative commons non-commercial attribution share alike 3.0. Please contact the author (mrule7404 at gmail dot com) for permission for commercial use.
  • Affiliated
    •  /  
  • Invited
    •  /  
  • Descended
    • michaelerule’s favorites
    From abhominal
    Biostructure
    From feanne
    grow, grow
    From Andy Gilmore
    =========
    From feanne
    Venus, peacock moss, earth...
    From sarahs
    FINAL PRESENTATION
    Recently commented on
    From michaelerule
    ∞ zoom
    From gamma
    Ducks fractals
    From michaelerule
    Limitless Enhancement
    From michaelerule
    Sun
    From Queston
    Physicists describe method...
    Now playing SpaceCollective
    Where forward thinking terrestrials share ideas and information about the state of the species, their planet and the universe, living the lives of science fiction. Introduction
    Featuring Powers of Ten by Charles and Ray Eames, based on an idea by Kees Boeke.
    From michaelerule's personal cargo

    Self-Organizing Maps


    These images were generated by a neural network algorithm for embedding a complex perceptual space in two dimensions. It would be very interesting to apply this algorithm to other visual data sets. If applied to auditory patches it could also be a useful generator of electronic music.

    http://en.wikipedia.org/wiki/Self-organizing_map
    http://tedlab.mit.edu/~dr/Lens/


    Using a dataset from benjiart:


    Handwritten digits dataset :


    Rudimentary animation, also from benjiart's data : drifting through the space ( sorry about the low resolution, these things kill my computer )




    Sun, Mar 8, 2009  Permanent link

      RSS for this post
    3 comments
      Promote (2)
      
      Add to favorites (2)
    Create synapse
     
    Comments:


    benjaminleemartin     Sun, Apr 12, 2009  Permanent link
    sounds very interesting although i am not sure to understand fully, I have a data set of 25,000
    human silouettes in different positions, would this qualify as a data set?

    Part of this collection of "humans" is here: www.humanography.com

    would love to hear more about your project.

    cheers
    benji


    michaelerule     Sun, Apr 12, 2009  Permanent link
    That sounds like an awesome dataset

    ( I cant see your java because I haven't got the plugin to compile for 64 bit firefox on linux yet )

    what say you we collaborate ?

    If you provide me the dataset I can dump it into the map algorithm and it might produce one of these nice pictures ( maybe, depending on how many images you have and how similar they are ). I'll post the result back here.

    There might be more you can do with it but that would be a good start. If you want to play with the algorithm yourself, you can start with the LENS example at
    http://tedlab.mit.edu/~dr/Lens/Example/kohonen.html

    This project grew out of a course assignment, so I have no part in developing these algorithms. LENS itself appears to be difficult to use ( at least for me ). The images are made with a little script in Java I wrote that processes the weight structure of the map network once it is trained.

    There are a few limitations to this algorithm at the moment, most notably it can only produce images that are linear combinations of images in the dataset. So, you'll notice with the glyph dataset you just get blurry figures and shadows rather than some sort of interpolation between characters.
    benjaminleemartin     Thu, Apr 16, 2009  Permanent link
    Super i like the results of the human pattern, the movie is quite beautiful too,
    when i have a little time i will send you some more image sets, (people sitting, lying etc)
    would be interesting to see some more results,

    peace and inspiration

    benji
     
          Cancel