Just adding this here so it is part of my "collection" of hacks. Sourcecode and writeup can be found here:
https://github.com/cpldcpu/BitNetPDK
https://cpldcpu.com/2024/05/02/machine-learning-mnist-inference-on-the-3-cent-microcontroller/
MNIST Neural Network Inference on a 3 cent Microcontroller
To make the experience fit your profile, pick a username and tell us what interests you.
We found and based on your interests.
Just adding this here so it is part of my "collection" of hacks. Sourcecode and writeup can be found here:
https://github.com/cpldcpu/BitNetPDK
https://cpldcpu.com/2024/05/02/machine-learning-mnist-inference-on-the-3-cent-microcontroller/
Create an account to leave a comment. Already have an account? Log In.
Of couse, the basic arithmetic element would just be an accumulator and some multiplexers or xor gates. What is a bit more complex, is how to get all the data in. But you could start small and use a microcontroller to do the shuffling.
Also, analog neurons could be made the same way.
Well, I am working on a TTL + uPD7220 video card, and getting all the data in is a big task there. For sure, there are good ways to load the inputs or strobe the weights - it is a similar issue, at least if you care about speed. What may be really interesting is making it both forward propagate and learn - although I'm sure this is also much more difficult. Anyway - thank you for sharing your project, it is really cool, as well as inspiring.
Become a member to follow this project and never miss any updates
Hmm, you're making me wonder: maybe this could be done also with TTL logic, without too much hassle? Seems like it could be an interesting project.