If you want to research in the field of neural networks, I offer you a promising idea - Border Pairs Method, which has 11 advantages over the glorious backpropagation method. Programming experiences are welcome.
https://www.theguardian.com/technology/2016/jun/28/google-says-machine-learning-is-the-future-so-i-tried-it-myself
Thanks for the reply. What software language or tool do you master? Would you like to write and publish a neural network learning program that uses very modern algorithms. I am using MATLAB until now, but now I would like an open code version, for example, in TensorFlow.
That is perfect.The MatLab code has about 1000 lines and is not yet optimized. If you are a researcher, you can use this new code for your research work. Have you already worked with neural networks or machine learning?
You can start right now. Look this:
Presentation New Deep Learning Algorithms beyond Backpropagation IBM Deve...
There are three algorithms.
Bipropagation is most simple ~100÷200 lines of code
Border Pairs Method (BPM) is the best one ~ 1000 lines of code
More info for BPM:
Article Border Pairs Method—constructive MLP learning classification algorithm
Thanks,
I am going through the paper first, then I will start implementing this.
- Badges
- Science topic