Nerve is a multilayer perceptron library implemented as a single C99 header
file with zero external dependencies.

• Single-header (nerve.h, ~1 250 lines) — copy and compile, nothing else required
• Adam optimizer (Kingma & Ba, 2015) and SGD with momentum
• Xavier/Glorot and He weight initialization
• Dropout (inverted) and L2 regularization
• Four activation functions: Sigmoid, Tanh, ReLU, Leaky ReLU
• Text and binary model persistence
• Tested on Linux, macOS, Windows (CI: GitHub Actions)

Repository: https://github.com/fkkarakurt/nerve
License: GPL-3.0-or-later

Checklist:

• Entry is in alphabetical order within the section
• Description follows existing format (one sentence, no markdown links inside)
• Reference number is unused
• License identifier matches SPDX