Python. Machine learning from scratch.
First step: read the sonnets result in the documentation, where a 358,052 parameter transformer brought validation loss from 5.297 to 2.413 in 250 steps, then, from the code on GitHub, run the gradient test: python -m unittest discover -s tests.
Machine serves autonomy, Layer 5. Here you learn how a neural network works by building one from scratch with only NumPy. You get code you can read, run, and change.
The code builds a neural network from basic parts, with only NumPy. A custom Tensor tracks each step so the code can work out gradients on its own. Layers include Dense, Dropout, and BatchNorm. Tools that update the model during training include SGD, Adam, and RMSprop. Ways to measure error include MSE and Cross-Entropy loss.
You can plot training with Matplotlib and watch the loss fall as your network learns.
The layer below: Graph, shared truth
The layer above: Lumen, trust and meaning
Built on Graph: shared truth.
Unlocks the ability to build.
The research under this layer: Foundations, Artificial intelligence and human computer interaction