Hello Everyone,
My name is Jean Michel Sellier and, a few days ago, I released the first version (0.0.1) of Gneural Network. First of all, let me thank you all for your interest in this new project and for offering your help! I actually received an overwhelming number of emails from volunteers, this is really encouraging! I am sure that together we will achieve something interesting and release a great tool to the community!
I suppose that you subscribed to this mailing list also because you are interested in knowing more about this new GNU package and that's what I would like to discuss about in this email. In particular, I would like to tell you more about the future directions I can already foresee and I really would like to know your opinions.
Very soon, the following features will be released (I am in the process of rewriting some code that I already developed and tested):
- a scripting language, so that the user can define and run a neural networks without even knowing how to program,
- a repository will be created to make interactions easier,
- recurrent networks, in particular LSTM neurons,
- an automatic algorithm which select the best network topology given a specific problem,
- parallelization of the training process for multi-core CPUs and GPUs,
- improvement of the optimizers for the training process,
- applications.
If you have any interesting comments on these directions and/or you have ideas you would like to share, this is your chance. Just reply to this email so that everyone can read your suggestions and can comment on them.
I do not see any particular difficulty in developing these points. Actually, they should be released quite quickly. Where I see the great chance to do something different than others is in the applications. In fact, as stated on the website of Gneural Network, trained networks are practically never shared (and neither the data). What I have in mind in particular is to train this network on meaningful problems that everybody can use.
For example, I have an idea in my mind since several years now. Some of you might be aware of the online game EteRNA. People play a game and actually solve important problems related to health and nanotechnologies. In fact, every time they solve a puzzle, they actually provide a solution which can be tested in a lab, consequently speeding up research and development. Imagine now if we could train Gneural Network on such a game and then release its trained version. A LOT of people could benefit from something like that, may be even saving human lives! Not to mention that such a trained network could also provide insights on how to use RNA to achieve (gate) quantum computing...
So, these are my two cents ;) I am looking forward to hear from you guys!
Thanks again for your interest and help!
Happy Neural Hacking!
JM
--
Jean Michel Sellier, Ph.D., M.Sc.