Gallery of examples

This gallery contains several application examples for the nnbma package to illustrate diverse features.

For simplicity’s sake, these notebooks will apply the approximation of a known analytical function rather than a complex physical model.

Neural networks training:

• training.ipynb: neural network-based approximation of an analytical vectorial function

• masking.ipynb: illustrate the use of a masked loss function to ignore unreliable labels

• operators.ipynb: illustrate the use of pre and post processing operators

Derivatives computing:

• derivatives.ipynb: illustrate how to compute the Jacobian or the Hessian of a neural network efficiently

• derivatives-time.ipynb: illustrate the difference in computation time between the different ways of calculating the matrix

Neural architectures:

• fully-connected.ipynb: classic multilayer perceptron

• densely-connected.ipynb: perceptron architecture with dense shortcuts

• merging-network.ipynb : illustrate the use of the MergingNetwork network

• embedding-network.ipynb : illustrate the use of the EmbeddingNetwork network and the AdditionalModule layer

Advanced neural networks features:

• polynomial-expansion.ipynb: illustrate the use of the PolynomialExpansion layer

• restrictable-layer.ipynb: illustrate the use of the RestrictableLayer layer

Gallery

• Neural networks based approximation of an analytical function
• Training with anomaly labels in the datasets
• Pre and post processing operators
• Neural networks auto-differentiation using PyTorch 2.0
• Multilayer perceptron
• Dense multilayer perceptron
• Neural networks merging
• Neural networks and modules assembly
• Polynomial expansion of inputs
• Restriction of outputs