aicaramba ✨

a simple neural network implementation from a perspective of linear algebra.

the library is mostly developed for recreational and educational purposes for myself and only depends on the rand-crate for randomization of newly created weight- and bias matrices.

for a usage example see src/bin/xor.rs, which simulates an XOR-logic-gate using a small neural network.

---

features

currently available features of the library:

• ReLU and Sigmoid activation functions
• the MSE loss function
• a single, down-to-earth struct that contains the whole network

roadmap

what might happen down the road:

• BCE loss function (requires output layer sigmoid activation - not a trivial addition)
• serde (de-)serialization to easily store checkpoints/training progress.
• perhaps a MNIST example (?)