So one day I was thinking, it would be kinda cool to have handwriting recognition on advanced monitors. And as I am coming closer and closer to the actual implementation(I'm currently like reading and reading and reading how it works in theory) I don't exactly know, whether a neural net is "efficient".
And if it actually comes to it, I would definitely want to also train the nets in ComputerCraft and bake the net to a single function at the end of the training for speed purposes. But I'm not too experienced with neural nets and maybe some of you have already made something AI like in CC and could tell me if the training would exceed my lifetime?
Is pattern/handwriting recognition a bad idea to train in plain lua code?
I could theoretically train it in something like Torch or similar and try to export it, but I don't know if thats practical
Are there already neural nets in use of some programs? (haven't seen any myself, but I don't know every program on the platform^^)
Maybe even some frameworks?
Is pattern/handwriting recognition a bad idea to implement in ComputerCraft?
Maybe theres a chance, that even a fully trained system becomes too large(in layers and neurons) that a single execution takes too much time to be enjoyful, but I think training will become more of a question. Most likely to be reached at a level of word recognition(but I'm not at the level of recurrent networks yet)
Besides digit recognition I could do a lot more even simpler stuff. Like differentiate between circles and crosses and stuff. What are your ideas on things like these? Do you think there are other regions, where the application of neural nets would make sense? Some turtly stuff? Prediction stuff for breaking bad encryptions which I see a lot around here? Any other ideas?
Edited by InDieTasten, 19 February 2015 - 03:52 AM.












