IA Generador neural network library for microcontroller

Hello, please let me know what you think of this AI web application for improvement. It allows you to generate trained neural network models for use with microcontrollers like Raspberry Pi Pico, ESP32, etc. These models don't require an internet connection and are trained by loading a CSV file. It dynamically maps inputs and outputs and has two hidden layers whose architecture can be configured.

The link is:

https://traindeep.ai

thanks

colporteur 13 days ago

What would be my motivation, to have me click on a link, from a user with a profile of 25 points.

That score would suggest you are new to the E14 community. Welcome.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
rbasulto53 13 days ago

Thanks for the welcome, this application is to help the community and has no commercial purpose.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
Dipeshkachhi 12 days ago

This is a really interesting project! I've been exploring the challenges of running neural networks on microcontrollers myself. Have you had a chance to compare the performance of IA Generador against other lightweight libraries like emlearn or NNoM? I'd be particularly interested in seeing a benchmark for inference speed on a Raspberry Pi Pico or an STM32 board. I'm excited to see where this project goes. Keep up the great work!
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
rbasulto53 12 days ago in reply to Dipeshkachhi

Thank you very much for your comment. Yes, I will be running performance tests, although the generated library is very lightweight for high performance and low resource consumption.
Another important difference is that it doesn't require installing anything to create the model, as it's trained on the web, and you don't need any libraries to run the trained model; it's very simple.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
DAB 10 days ago

If you want anyone to bother with your link, you should greatly expand your description of what is at the end of the link.

It generates a trained NN. So what?

Give me a useful example so I can be intrigued about the possibilities.

Provide a walk through of the process so I can see how easy it is to use.

What is the size of the resulting output? How useful is it? How limited is its application?

You have to make a reasonable effort to sell it if you want anyone to kick the tires.
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
rbasulto53 3 days ago in reply to DAB

thanks for you comment. Please check out the new version 2.0. It has much of what you mentioned and much more.
https://traindeep.ai
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
rbasulto53 3 days ago

Ready version 2.0.
https://traindeep.ai
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel
rbasulto53 2 days ago in reply to Dipeshkachhi

hi,
We ran tests on the rp2040 and obtained these measurements (execution time of already trained model)
- 3 inputs, 2 hidden layers, and 6 neurons per layer with ReLU activation function: 3.5 ms
- 3 inputs, 4 hidden layers, and 10 neurons per layer with ReLU and Leaky activation function: 7.1 ms
Cancel
Vote Up 0 Vote Down

Sign in to reply

Cancel