Neuralatex: A machine learning library written in pure LATEX
3 days ago
- #autograd
- #LaTeX
- #neural-networks
- Implementation based on MicroGrad with a better choice of language.
- Neuralatex implements backpropagation over a dynamically-constructed DAG.
- Autograd engine requires 700 lines of pure LaTeX, neural network library 400 lines.
- Neuralatex is object-oriented using the TiKZ PGF module oo.
- Gradients can be obtained by calling .backward().
- Example code demonstrates Value object creation, multiplication, and gradient computation.
- MLP (Multi-Layer Perceptron) can be easily defined and called.