"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 50
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
A layer is a union of neurons not dependent on the same iteration and having the same type.
How do you do that? How are you going to squeeze the juice out of the GPU without vector representation? That's one of the things I'm taking into account.
Why without vector? The data arrays are external, it only remains to prescribe the correspondence of what data do with what data.
You have data about what neurons are in one layer, and pass to GPU calculation of vector of this layer. and so on through the layers.
SZY ah yes in neuron model that I drew, you need to introduce the concept of data to store intermediate calculations (well, to work with GPU they should be external too).
1) Why?
2) Why?
1) because in my formulation the layer can contain different types of neurons, and this is atypical tasks that cannot be transferred to the GPU
2) because in Andrew's formulation one neuron can be a layer, and that threatens to underload the GPU
ZY generally choose the lesser of two evils, underloading the GPU is not as terrible as the potential inability to use the GPU.
It's too bad that mql doesn't have pointers to data, otherwise we could just aggregate data from vectors into neurons directly.
HH wait, why do we need aggregation, if we can just pass the index of data vector instead. it will be the same binding with direct access.
HINT create a data reference object in the neuron instead of a data object.
2) because in Andrei's formulation one neuron can be a layer, and this threatens to underload the GPU
If it can be merged... Then it should be merged.
I have scraps of thoughts on how to organize a growing neural gas with such an engine, but they have not yet formed into words.
Only one thesis: network reinitialization will be necessary. The initialization itself should be available to the learning algorithm.
The project manager can be gpwr. Partially can me.
Thank you for your confidence, but it seems to me that I would not fit as a leader of this project. I will explain why.
From all of the above, my recommendation is this.
Good morning.
I saw that you were looking for a name for the project.
Even NeuRobot/NeuroBot/NeuRobotic -- Neurobot/Neurobotic might be an idea, in the case of neural robots.
A little childish as a name, but you get the idea of neural and financial robotics.