Home

Nerve — Neural Network Library

Zero-dependency multilayer perceptron in a single C header.

Copy nerve.h into your project. Define NERVE_IMPLEMENTATION in one translation unit. Compile with -lm. That is the entire integration.

#define NERVE_IMPLEMENTATION
#include "nerve.h"

nerve_t *net = nerve_new("2->8->1");
nerve_fit(net, X, y, n_samples, 5000);
printf("accuracy: %.1f%%\n", nerve_score(net, X, y, n_samples) * 100.f);
nerve_free(net);

gcc -O2 main.c -o main -lm

---

Features

Optimizers	SGD + momentum · Adam (Kingma & Ba, 2015)
Activations	Sigmoid · Tanh · ReLU · Leaky ReLU
Init	Uniform · Xavier/Glorot (2010) · He (2015)
Regularisation	L2 weight decay · Inverted Dropout
Training	Online · Mini-batch · Shuffled epoch
Persistence	Text (.net) · Binary (.bin)
Metrics	Accuracy · Confusion matrix · MSE
API levels	Easy API (nerve_*) · Core API (net_*)

---

Pages

Page	Contents
Getting Started	Installation, first program, compilation
Core API	All net_* functions — full reference
Easy API	nerve_* convenience layer
Algorithms	Activation functions, optimizers, initializers, regularization

---

Quick benchmark

Adam vs SGD on a 4-class identity problem (4-6-4 network, 7 runs):

Optimizer	Avg iterations	Speedup
SGD / Uniform / Sigmoid	22 689	1×
Adam / He / ReLU	1 214	18.7×

---

License

Copyright (C) 2026 Fatih Küçükkarakurt — GPL-3.0