Examples

Examples may be used as templates for new projects... All examples are at GitHub/examples:

Simple MLP: A simple multi-layer perceptron for MNIST classification, build with Knet and Helferlein-types in just one line of code (or so).

Simple LeNet: A simple LeNet for MNIST classification, build with help of the Helferlein layers in just two (ok: long) lines of code.
Training unbalanced data with help of a focal loss function: A simple MLP with focal loss demonstrate classification of highly unbalanced data.

Vanilla Autoencoder: A simple autoencoder design with help of Knet in Helferlein-style.
Convolutional Autoencoder: A convolutional autoencoder design with help of Knet in Helferlein-style.
Variational Autoencoder: Example for a simple VAE utilising the NNHelferlein-type VAE and demonstrating the fascinating regularisation of a VAE.
Simple sequence-to-sequence network: Simple s2s network to demonstrate how to setup macghine translation with a rnn.
Sequence-to-sequence RNN for machine translation: RNN to demonstrate how to setup machine translation with a bidirectional encoder RNN and attention.
RNN Sequence tagger for annotation of ECGs: RNN to demonstrate how to set-up a sequence tagger to detect heart beats. Only one layer with 8 units is necessary to achieve almost 100% correct predictions. The example includes the definition on peephole LSTMs to display how to integrate non-standard rnn-units with the NNHelfrelein framework.
Import a Keras model: The notebook shows the import of a pretrained VGG16 model from Tensorflow/Keras into a Knet-style CNN and its application to example images utilising the Helferlein imagenet-utilities.

Transformer for machine translation: A simple transformer architecture is set up according to the 2017 Vaswani paper Attention is All You Need with help of NNHelferlein-utils.

Simple Transformer API for Bert-like architectures: A simple transformer architecture is set up with the NNHelferlein transformer API.

Based on the Keras import constructors, it is easy to import pretrained models from the TF/Keras ecosystem.