[erlang-questions] DXNN: Topology and Parameter Evolving Universal Learning Network system/platform, released to GitHub.

G.S. <>
Sat Jun 16 07:01:00 CEST 2012


Hello Shahrdad,

Thanks.
The reason for the concentration on Neuroevolution is because the goal of
my research is to develop computational intelligence systems that are
flexible, scalable, general, and capable of learning and self modification.
Seed computational intelligence... At this time we know with absolute
certainty of one approach that has worked and is capable of producing such
systems; the method is Evolution, the evolution of neural networks, and the
proof that it works, is us, biological thinking machines. Our brains are
the result of billions of years of evolution, which has carved out in flesh
our neural circuitry through trial and error.

Because DXNN MK2 is fully decoupled, it allows for other search methods,
selection, mutation... to be implemented and simply plugged in. As others
begin using the system, they will most likely modify and try out and create
new modules and functions with regards to search optimisation (artificial
immune system, ant colony, swarm, CMA-CS...), selection, mutation..., and
hopefully contribute those new functions and modules, thus making them
selectable as options within the DXNN, and make the system even more useful
within the field, and in general.

Best regards,
-Gene

On Fri, Jun 15, 2012 at 9:52 AM, Shahrdad Shadab <>wrote:

> Grate work! I cannot wait to read your book.
>  I am also working on using Erlang in statistical machine learning, this
> requires of mathematical/statistical
> library functions (like linear algebra / statistical libraries and so on)
> to be implemented in Erlang which is taking a lot of time from me.
> I wonder if you ever looked at statistical approach to AI and why you
> didn't follow that path as opposed to neuron-genetic approach.
>
> Thanks a lot
> Best regards
> Shahrdad
>
>
>
>  On Tue, Jun 12, 2012 at 1:31 PM, G.S. <> wrote:
>
>>  Hello all,
>>
>> DXNN [1,4]  is Topology and Parameter Evolving Universal Learning Network
>> (TPEULN) system, similar to topology and weight evolving artificial neural
>> network, but more general, and not constrained to the use of only sigmoid
>> based activation function neurons. Erlang was chosen because of its perfect
>> and complete mapping to the neural network architecture.
>>
>> DXNN is a TPEULN platform that uses direct and indirect encoding (neural
>> and substrate respectively [5]), has a cross-validation system for
>> experimentation, decoupled sensor/actuator systems, decoupled
>> learning/selection/... algorithms (in MK2), a built in 2d world simulator
>> called flatland for ALife experiments (all in gs()).
>>
>> The second generation (mk2) DXNN is available as a branch of the original
>> project, and is a clean implementation of this computational intelligence
>> evolving system. It is also the system explained and created in my Springer
>> book: Handbook of Neuroevolution Through Erlang [2,3], with a foreword
>> written by Joe Armstrong. The book will go into print this September.
>>
>> There are not a lot of comments within the source code on github, but I
>> will continue to add more comments as time permits.
>>
>> Upcoming features:
>> 1. Visualisation system.
>> 2. New selection algorithm modules.
>> 3. New speciation and diversification functions.
>> 4. An improved cross-validation system for the experiment database.
>> 5. Full population backup, so that all agents are saved, and only
>> manually deleted at the researcher's request (they don't take much space,
>> and it would make for an interesting visualisation, and ability to traverse
>> from the seed agent to the current agent).
>>
>> -Gene
>> [1] https://github.com/CorticalComputer/DXNN First generation DXNN has a
>> convoluted implementation. DXNN mk2 is a very clean implementation and is
>> currently on the non master branch, it will eventually overwrite the master
>> branch but both have the same features (almost) at this time.
>> [2] http://www.springer.com/computer/swe/book/978-1-4614-4462-6
>> [3]
>> http://www.amazon.com/Handbook-Neuroevolution-through-Erlang-Gene/dp/1461444624/ref=sr_1_1?ie=UTF8&qid=1338163875&sr=8-1<https://github.com/CorticalComputer/DXNN>
>> [4] http://www.erlang-factory.com/conference/SFBay2012/speakers/GeneSher
>> [5] http://arxiv.org/abs/1111.5892
>>
>> _______________________________________________
>> erlang-questions mailing list
>> 
>> http://erlang.org/mailman/listinfo/erlang-questions
>>
>>
>
>
> --
> Software Architect & Computer Scientist
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20120616/8832add5/attachment.html>


More information about the erlang-questions mailing list