From: RaL Date: 2008-03-20T19:20:38+09:00 Subject: Re: Google Summer of Code 2008? I want to create a project devoted to Neural Networks. Would Ruby Central be interested to support me and this project for Google Summer of Code? Right now, there are two gems with a Neural Network support - http://ai4r.rubyforge.org/ É http://ruby-fann.rubyforge.org/. But both of them implement only one learning algorithm - back-propagation (and accordingly to that fact only one type of Neural Network - multilayer perceptron) - also they don't have ability to choose different activation functions for different layers and so you can't even build Radial Basis Function Network with those gems. They don't have a built- in interface to use even widespread learning strategies (e.g., begin learning with a high learning rate and then switch to a low learning rate after approaching some threshold). So as a GSoC project I decide to implement several Neural Network types, including multilayer perceptron, self-organizing feature map (Kohonen net), auto-associative memory (Hopfield net) and recurrent neural network. I want to build an easy but powerful interface to use with. So it will support a range of learning strategies for fast start but it will be very customizable in case of need. It will have a built- in visualization module so you could see a net topology with all values of weights, errors, input and output values in any moment of time. You could also see a graphic of values (good to see a final approximation) and errors (to understand a dependence on epoch count). And as the last step i want to build a Neural Network web-editor via silverlight and IronRuby. So what does RubyCentral think about my proposal? On Mar 19, 5:05špm, James Gray wrote: > On Mar 19, 2008, at 12:16 AM, Chiyuan Zhang wrote: > > > Is there any project ideas collected? Maybe something about Ruby 1.9, > > Rubinius etc. > > I'm sure we can brainstorm some for you. > > Regarding your Ruby 1.9 idea: šmany libraries don't work correctly yet š > in Ruby 1.9. šYou could come up with a list of favorites that don't š > work and offer to patch them for Ruby 1.9. šI think that could be š > quite a resource to the community. šI got this idea from Greg Brown š > who may be doing a similar project soon, but I think there are more š > than enough libraries to go around. > > Beyond that, here are a few other random ideas: > > * There's been a lot of talk about better documentation tools lately. š > I believe Eric Hodel is working on revamping RDoc. šPerhaps you could š > see what you could do to help with that. šJames Britt has also looked š > at better ways to assemble the documentation for ruby-doc.org, so that š > may be another place you could help out. > * I've recently had need of a multiprocess event safe logger. šWhat I š > mean by that is that I have two processes writing to the same log š > file. šThey may log five things for each event. šIn the log file, I š > would prefer the resulting ten entries not be interleaved. šYou could š > provide a start event method and an end event method and only add š > events to the log file after the event ended. šThis is more š > complicated than it sounds since you will want to deal with large š > logging data for a single event (probably using Tempfile), badly š > behaved programs that don't manage to call the end event method, long š > running events, etc. > * Daniel Berger has requested a pure Ruby implementation of zlib a few š > times now. šThis might give new options for Windows installers. > > I do encourage you to think up your own ideas though. šI've been a š > mentor for the last two years and I can easily say that the š > applications for unique ideas were generally more appealing. šThere š > are two reasons for that. šThe first is that lots of applications come š > in for the ideas we post publicly, so you're competing with a lot of š > people who want to do the exact same thing. šThe other is that you š > will just be more creative when running with your own idea. šDon't š > hesitate to latch onto one of your own needs and run with it. > > James Edward Gray II