Implementing neural network functions on HDL

I have trained a classifier using the neural networks toolbox, I don't need to do any training on the hardware so that part is not what I want to convert into HDL, what I do need is to convert a function that runs the forward-pass ,using the trained model, on input data for classification.

The main function I use for classification calls lots of other functions from the NN toolbox, so I'm not sure if it is even possible to convert everything to HDL, maybe I should try going with Simulink first then HDL?

Not sure how mych details do I need to provide, so if I need to do so, please let me know.

Answer 1

[Nick Choi]

The following documentation page provides a list of the MATLAB functions that are supported for HDL code generation:

https://www.mathworks.com/help/hdlcoder/ug/functions-supported-for-hdl-code-generation-alphabetical-list.html

This documentation page has a list of the Simulink Blocks that are supported for HDL code generation: https://www.mathworks.com/help/hdlcoder/block-support.html

These pages should help give you a better idea of what could be converted into HDL code.

[Alla]

Thank you Nick, I'm coming back to this with an other question if you'd be so kind to help.

I have contacted staff about this again and it would seem that the FPGA coder will be released soon, but since there is no set date I might try and work on this on my own, my question is if it is possible to use the HDL coder to put together a decent forward-pass CNN code?

I have been looking at the supported function list but it is quite difficult to compare and match with the functions required for a forward pass using the deep CNN toolbox.

Answer 2

[Alexander Vorobyev]

Example hdl code generation from CNN https://www.youtube.com/watch?v=dSWkawpTiM4