Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Jeffrey Phillips Freeman
    @freemo
    @Lord-of-the-Galaxy there ya go
    Lord of Galaxy
    @Lord-of-the-Galaxy
    Better
    Jeffrey Phillips Freeman
    @freemo
    :)
    Lord of Galaxy
    @Lord-of-the-Galaxy
    I think, i think im not gonna get through all this in any sensible period of time
    So sorry, but I want to jump to Neural Networks
    I hope thats okay with you, i wont disturb you in making the graph based systems
    But ill start work on neural networks (ive already been through both the neural network architectures that existed in Ferma (i.e. SOM and Backprop), so uderstanding wouldnt be a problem
    I would just need advixe on how to optimise that on Aparapi when relying on a Neuron class instead of matrices.
    Lord of Galaxy
    @Lord-of-the-Galaxy
    (i do realise that a neuron class will have far more uses (different types of NN), so I'll stick to that)
    Jeffrey Phillips Freeman
    @freemo
    @Lord-of-the-Galaxy so still want to work together on the current design just focusing on the ANN part? or you going to go off on your own project?
    Lord of Galaxy
    @Lord-of-the-Galaxy
    Ill work with you
    As faf as i remember,
    *far
    The original code had the Neural Networks independent (mostly) of the rest of the code
    So it shouldnt be a problem
    (@freemo )
    ( @freemo )
    Jeffrey Phillips Freeman
    @freemo
    @Lord-of-the-Galaxy ok that works, I can do the infrastructure and you can focus on the neural network side if youd like
    Lord of Galaxy
    @Lord-of-the-Galaxy
    @freemo Cool, let me know as soon as youre done, ill learn up on new and better neural network achitectures till then.
    Jeffrey Phillips Freeman
    @freemo
    @Lord-of-the-Galaxy will do
    automenta
    @automenta
    i took a look at GRAIL again because i was thinkking about something like it but not exactly
    for laying various graph algorithms on top of NARS but basically NARS without NARS, just extracting out the general purpose symbolic system
    and making NARS a particular layer that could operate on it, along with other graph algorithms like dANN/GRAIL etc
    for example the 'concept' class is just a map it can store arbitrayr objects
    i have a few link networks running in paralllel but these are currently hardcoded. by making these adapters and adding more support for several contextually identified graph layers to intersect at each concept
    and certain compound concepts would act as proxies for the lower level links, so that they can be accessed by higher level logic but also efficiently by the lower level algorithms
    then i thought about signaling and asynch timing and thats when i remembered grail with its signal propagators