Neural Networks

Complete reference for all 100+ neural network architectures in AiDotNet.

Convolutional Networks (CNN)

Standard Architectures

var model = new ResNet<float>(
    variant: ResNetVariant.ResNet50,
    numClasses: 1000,
    pretrained: true);

Lightweight Networks

Attention-Enhanced CNNs

Recurrent Networks (RNN)

Core Architectures

var model = new LSTM<float>(
    inputSize: 100,
    hiddenSize: 256,
    numLayers: 2,
    dropout: 0.2f,
    bidirectional: true);

Advanced RNNs

Transformers

Text Transformers

var model = new Transformer<float>(
    vocabSize: 30522,
    dModel: 768,
    numHeads: 12,
    numLayers: 12,
    dFf: 3072,
    maxSeqLen: 512);

Vision Transformers

Multi-Modal Transformers

Generative Adversarial Networks (GAN)

Standard GANs

var gan = new DCGAN<float>(
    latentDim: 100,
    imageSize: 64,
    numChannels: 3);

Conditional GANs

Style GANs

Image-to-Image

Variational Autoencoders (VAE)

var vae = new VAE<float>(
    inputDim: 784,
    latentDim: 32,
    hiddenDims: [512, 256]);

Diffusion Models

var diffusion = new DDPM<float>(
    imageSize: 256,
    timesteps: 1000,
    betaSchedule: BetaSchedule.Linear);

Graph Neural Networks (GNN)

var gnn = new GAT<float>(
    inputDim: 64,
    hiddenDim: 128,
    outputDim: 7,
    numHeads: 8);

Capsule Networks

Neural Radiance Fields (NeRF)

Physics-Informed Neural Networks

Specialized Architectures

Sequence Models

Memory Networks

Attention Mechanisms

Usage Examples

Image Classification

var model = new EfficientNet<float>(
    variant: EfficientNetVariant.B4,
    numClasses: 100);

var result = await new AiModelBuilder<float, Tensor<float>, int>()
    .ConfigureModel(model)
    .ConfigureOptimizer(new AdamWOptimizer<float>())
    .ConfigureGpuAcceleration(new GpuAccelerationConfig { Enabled = true })
    .BuildAsync(images, labels);

Text Generation

var model = new GPT<float>(
    vocabSize: 50257,
    dModel: 768,
    numHeads: 12,
    numLayers: 12);

var generated = model.Generate(
    prompt: "Once upon a time",
    maxTokens: 100,
    temperature: 0.7f);

Graph Classification

var model = new GIN<float>(
    inputDim: 32,
    hiddenDim: 64,
    outputDim: 10,
    numLayers: 5);

var prediction = model.Classify(graph);